vulkan: subgroup size tuning (#12087)

* vulkan: subgroup size test

* Vulkan: Add device architecture enum and logic to recognize AMD generations

* vulkan: use new architecture logic to specify subgroup size

* Initial vulkan subgroup size tuning for RDNA3

* vulkan: commonize RDNA subgroup tuning

* vulkan: override subgroup size if required_subgroup_size = 0

* vulkan: disable warp 32 for RDNA3

* vulkan: fine tuned RDNA1 subgroup sizes

* vulkan: adjusted subgroup size map

* vulkan: fixed RDNA2 subgroup map

---------

Co-authored-by: 0cc4m <picard12@live.de>

.github/workflows/build.yml

@@ -467,6 +467,7 @@ jobs:
         run: |
           cmake -B build -S . \
             -DCMAKE_HIP_COMPILER="$(hipconfig -l)/clang" \
+            -DGGML_HIP_ROCWMMA_FATTN=ON \
             -DGGML_HIP=ON
           cmake --build build --config Release -j $(nproc)
 
@@ -476,6 +477,7 @@ jobs:
           cmake -B build2 -S . \
             -DCMAKE_C_COMPILER=hipcc \
             -DCMAKE_CXX_COMPILER=hipcc \
+            -DGGML_HIP_ROCWMMA_FATTN=ON \
             -DGGML_HIP=ON
           cmake --build build2 --config Release -j $(nproc)
 
@@ -710,12 +712,11 @@ jobs:
             -DLLAMA_BUILD_SERVER=OFF \
             -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64"
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu)
-          sudo cmake --install build --config Release
 
       - name: xcodebuild for swift package
         id: xcodebuild
         run: |
-          xcodebuild -scheme llama-Package -destination "${{ matrix.destination }}"
+          ./build-xcframework.sh
 
   windows-msys2:
     runs-on: windows-latest
@@ -773,7 +774,7 @@ jobs:
     env:
       OPENBLAS_VERSION: 0.3.23
       SDE_VERSION: 9.33.0-2024-01-07
-      VULKAN_VERSION: 1.3.261.1
+      VULKAN_VERSION: 1.4.304.1
 
     strategy:
       matrix:
@@ -1203,6 +1204,11 @@ jobs:
         id: checkout
         uses: actions/checkout@v4
 
+      - name: Clone rocWMMA repository
+        id: clone_rocwmma
+        run: |
+          git clone https://github.com/rocm/rocwmma --branch rocm-6.2.4 --depth 1
+
       - name: Install
         id: depends
         run: |
@@ -1232,8 +1238,10 @@ jobs:
           cmake -G "Unix Makefiles" -B build -S . `
             -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
             -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
+            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/rocwmma/library/include/" `
             -DCMAKE_BUILD_TYPE=Release `
             -DGGML_HIP=ON `
+            -DGGML_HIP_ROCWMMA_FATTN=ON `
             -DGGML_RPC=ON
           cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
 
@@ -1252,6 +1260,11 @@ jobs:
         with:
             fetch-depth: 0
 
+      - name: Clone rocWMMA repository
+        id: clone_rocwmma
+        run: |
+          git clone https://github.com/rocm/rocwmma --branch rocm-6.2.4 --depth 1
+
       - name: ccache
         uses: hendrikmuhs/ccache-action@v1.2.16
         with:
@@ -1281,8 +1294,10 @@ jobs:
           cmake -G "Unix Makefiles" -B build -S . `
             -DCMAKE_C_COMPILER="${env:HIP_PATH}\bin\clang.exe" `
             -DCMAKE_CXX_COMPILER="${env:HIP_PATH}\bin\clang++.exe" `
+            -DCMAKE_CXX_FLAGS="-I$($PWD.Path.Replace('\', '/'))/rocwmma/library/include/" `
             -DCMAKE_BUILD_TYPE=Release `
             -DAMDGPU_TARGETS=${{ matrix.gpu_target }} `
+            -DGGML_HIP_ROCWMMA_FATTN=ON `
             -DGGML_HIP=ON `
             -DGGML_RPC=ON
           cmake --build build -j ${env:NUMBER_OF_PROCESSORS}
@@ -1321,6 +1336,8 @@ jobs:
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
 
       - name: Build
         id: cmake_build
@@ -1336,15 +1353,40 @@ jobs:
             -DCMAKE_OSX_DEPLOYMENT_TARGET=14.0 \
             -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
           cmake --build build --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
-          sudo cmake --install build --config Release
 
       - name: xcodebuild for swift package
         id: xcodebuild
         run: |
-          xcodebuild -scheme llama-Package -destination 'generic/platform=iOS'
+          ./build-xcframework.sh
 
       - name: Build Xcode project
-        run: xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' build
+        run: xcodebuild -project examples/llama.swiftui/llama.swiftui.xcodeproj -scheme llama.swiftui -sdk iphoneos CODE_SIGNING_REQUIRED=NO CODE_SIGN_IDENTITY= -destination 'generic/platform=iOS' FRAMEWORK_FOLDER_PATH=./build-ios build
+
+      - name: Determine tag name
+        id: tag
+        shell: bash
+        run: |
+          BUILD_NUMBER="$(git rev-list --count HEAD)"
+          SHORT_HASH="$(git rev-parse --short=7 HEAD)"
+          if [[ "${{ env.BRANCH_NAME }}" == "master" ]]; then
+            echo "name=b${BUILD_NUMBER}" >> $GITHUB_OUTPUT
+          else
+            SAFE_NAME=$(echo "${{ env.BRANCH_NAME }}" | tr '/' '-')
+            echo "name=${SAFE_NAME}-b${BUILD_NUMBER}-${SHORT_HASH}" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Pack artifacts
+        id: pack_artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        run: |
+          zip --symlinks -r llama-${{ steps.tag.outputs.name }}-xcframework.zip build-apple/llama.xcframework
+
+      - name: Upload artifacts
+        if: ${{ ( github.event_name == 'push' && github.ref == 'refs/heads/master' ) || github.event.inputs.create_release == 'true' }}
+        uses: actions/upload-artifact@v4
+        with:
+          path: llama-${{ steps.tag.outputs.name }}-xcframework.zip
+          name: llama-${{ steps.tag.outputs.name }}-xcframework
 
   android-build:
     runs-on: ubuntu-latest

AUTHORS

@@ -1,4 +1,4 @@
-# date: Tue Feb  4 13:04:05 EET 2025
+# date: Sat Mar  8 18:23:52 EET 2025
 # this file is auto-generated by scripts/gen-authors.sh
 
 0cc4m <picard12@live.de>
@@ -8,10 +8,12 @@
 3ooabkhxtn <31479382+3ooabkhxtn@users.noreply.github.com>
 44670 <44670@users.noreply.github.com>
 65a <10104049+65a@users.noreply.github.com>
+708-145 <40387547+708-145@users.noreply.github.com>
 AN Long <aisk@users.noreply.github.com>
 AT <manyoso@users.noreply.github.com>
 Aarni Koskela <akx@iki.fi>
 Aaron Miller <apage43@ninjawhale.com>
+Aaron Teo <57927438+taronaeo@users.noreply.github.com>
 Aaryaman Vasishta <aaryaman.vasishta@amd.com>
 Abheek Gulati <abheekg@hotmail.com>
 Abhilash Majumder <30946547+abhilash1910@users.noreply.github.com>
@@ -20,6 +22,7 @@ Adithya Balaji <adithya.b94@gmail.com>
 AdithyanI <adithyan.i4internet@gmail.com>
 Adrian <smith.adriane@gmail.com>
 Adrian Hesketh <a-h@users.noreply.github.com>
+Adrian Kretz <me@akretz.com>
 Adrien Gallouët <adrien@gallouet.fr>
 Adrien Gallouët <angt@huggingface.co>
 Ahmad Tameem <113388789+Tameem-10xE@users.noreply.github.com>
@@ -28,15 +31,18 @@ AidanBeltonS <87009434+AidanBeltonS@users.noreply.github.com>
 AidanBeltonS <aidan.belton@codeplay.com>
 Aisuko <urakiny@gmail.com>
 Akarshan Biswas <akarshan.biswas@gmail.com>
+Akarshan Biswas <akarshan@menlo.ai>
 Akarshan Biswas <akarshanbiswas@fedoraproject.org>
 Al Mochkin <14274697+amochkin@users.noreply.github.com>
 Albert Jin <albert.jin@gmail.com>
 Alberto <57916483+albbus-stack@users.noreply.github.com>
 Alberto Cabrera Pérez <alberto.cabrera@codeplay.com>
 Alberto Cabrera Pérez <alberto.cabrera@intel.com>
+Aleksei Nikiforov <103434461+AlekseiNikiforovIBM@users.noreply.github.com>
 Alex <awhill19@icloud.com>
 Alex Azarov <alex@azarov.by>
 Alex Azarov <alexander.azarov@mapbox.com>
+Alex Brooks <alex.brooks@ibm.com>
 Alex Klinkhamer <from.github.com.917@grencez.dev>
 Alex Klinkhamer <git@grencez.dev>
 Alex Nguyen <tiendung@users.noreply.github.com>
@@ -67,6 +73,7 @@ Andrew Minh Nguyen <40281306+amqdn@users.noreply.github.com>
 Andy Salerno <andysalerno@gmail.com>
 Andy Tai <andy-tai@users.noreply.github.com>
 Anthony Van de Gejuchte <anthonyvdgent@gmail.com>
+Antoine Viallon <antoine@lesviallon.fr>
 Antonis Makropoulos <benuix@gmail.com>
 Arik Poznanski <arikpoz@users.noreply.github.com>
 Armen Kaleshian <kriation@users.noreply.github.com>
@@ -83,6 +90,7 @@ Atsushi Tatsuma <yoshoku@outlook.com>
 Austin <77757836+teleprint-me@users.noreply.github.com>
 AustinMroz <austinmroz@utexas.edu>
 BADR <contact@pythops.com>
+BB-fat <45072480+BB-fat@users.noreply.github.com>
 Bach Le <bach@bullno1.com>
 Bailey Chittle <39804642+bachittle@users.noreply.github.com>
 BarfingLemurs <128182951+BarfingLemurs@users.noreply.github.com>
@@ -101,6 +109,7 @@ Bert Wagner <github@bertwagner.com>
 Billel Mokeddem <billel.mokeddem.ml@gmail.com>
 Bingan <70050083+binganao@users.noreply.github.com>
 Bjarke Viksøe <164612031+bviksoe@users.noreply.github.com>
+Bodhi <3882561+BodhiHu@users.noreply.github.com>
 Bodo Graumann <mail@bodograumann.de>
 Bono Lv <lvscar@users.noreply.github.com>
 Borislav Stanimirov <b.stanimirov@abv.bg>
@@ -128,6 +137,7 @@ CentricStorm <CentricStorm@users.noreply.github.com>
 Chad Brewbaker <crb002@gmail.com>
 Changyeon Kim <cyzero.kim@samsung.com>
 Chao Jiang <jc19chaoj@zoho.com>
+Charles Duffy <charles@dyfis.net>
 Charles Xu <63788048+chaxu01@users.noreply.github.com>
 Charles Xu <charles.xu@arm.com>
 Chen Xi <xi2.chen@intel.com>
@@ -139,12 +149,14 @@ Chris Kuehl <ckuehl@ckuehl.me>
 Christian Demsar <christian@github.email.demsar.us>
 Christian Demsar <crasm@git.vczf.us>
 Christian Falch <875252+chrfalch@users.noreply.github.com>
+Christian Fillion <cfillion@users.noreply.github.com>
 Christian Kastner <ckk@kvr.at>
 Christian Kögler <ck3d@gmx.de>
 Christian Köhnenkamp <cvk5@me.com>
 Christian Zhou-Zheng <59622928+christianazinn@users.noreply.github.com>
 Christopher Nielsen <62156882+mascguy@users.noreply.github.com>
 Clark Saben <76020733+csaben@users.noreply.github.com>
+Clauszy <zhangyub@uniontech.com>
 Clint Herron <hanclinto@gmail.com>
 Conrad Kramer <conrad@conradkramer.com>
 Corentin REGAL <corentin.regal@gmail.com>
@@ -163,6 +175,7 @@ Daniel Hiltgen <dhiltgen@users.noreply.github.com>
 Daniel Illescas Romero <illescas.daniel@protonmail.com>
 Daniel Kleine <53251018+d-kleine@users.noreply.github.com>
 Daniele <57776841+daniandtheweb@users.noreply.github.com>
+Danny Milosavljevic <dannym@friendly-machines.com>
 DannyDaemonic <DannyDaemonic@gmail.com>
 Dat Quoc Nguyen <2412555+datquocnguyen@users.noreply.github.com>
 Dave <dave-fl@users.noreply.github.com>
@@ -170,6 +183,7 @@ Dave Airlie <airlied@gmail.com>
 Dave Airlie <airlied@redhat.com>
 Dave Della Costa <ddellacosta+github@gmail.com>
 David Friehs <david@friehs.info>
+David Huang <1969802+hjc4869@users.noreply.github.com>
 David Kennedy <dakennedyd@gmail.com>
 David Pflug <david@pflug.email>
 David Renshaw <dwrenshaw@gmail.com>
@@ -236,6 +250,7 @@ Felix <stenbackfelix@gmail.com>
 Finn Voorhees <finnvoorhees@gmail.com>
 Firat <firatkiral@gmail.com>
 FirstTimeEZ <179362031+FirstTimeEZ@users.noreply.github.com>
+Florent BENOIT <fbenoit@redhat.com>
 Folko-Ven <71110216+Folko-Ven@users.noreply.github.com>
 Foul-Tarnished <107711110+Foul-Tarnished@users.noreply.github.com>
 Francisco Melo <43780565+francis2tm@users.noreply.github.com>
@@ -254,6 +269,7 @@ Gary Mulder <gjmulder@gmail.com>
 Gavin Zhao <gavinzhaojw@protonmail.com>
 Genkagaku.GPT <hlhr202@163.com>
 Georgi Gerganov <ggerganov@gmail.com>
+Gian-Carlo Pascutto <gcp@sjeng.org>
 Gilad S <giladgd@users.noreply.github.com>
 Gilad S. <7817232+giladgd@users.noreply.github.com>
 Giuseppe Scrivano <giuseppe@scrivano.org>
@@ -267,7 +283,9 @@ Guspan Tanadi <36249910+guspan-tanadi@users.noreply.github.com>
 Gustavo Rocha Dias <91472747+gustrd@users.noreply.github.com>
 Haggai Nuchi <h.nuchi@gmail.com>
 Halalaluyafail3 <55773281+Halalaluyafail3@users.noreply.github.com>
+Hale Chan <halechan@qq.com>
 Hamdoud Hakem <90524568+hamdoudhakem@users.noreply.github.com>
+Han Yin <han.yin@arm.com>
 HanishKVC <hanishkvc@gmail.com>
 Haohui Mai <ricetons@gmail.com>
 Haoxiang Fei <tonyfettes@tonyfettes.com>
@@ -278,6 +296,7 @@ Haus1 <haus.xda@gmail.com>
 Henk Poley <HenkPoley@gmail.com>
 Henri Vasserman <henv@hot.ee>
 Henrik Forstén <henrik.forsten@gmail.com>
+Henry Linjamäki <henry.linjamaki@gmail.com>
 Herman Semenov <GermanAizek@yandex.ru>
 Hesen Peng <hesen.peng@gmail.com>
 HimariO <dsfhe49854@gmail.com>
@@ -307,6 +326,7 @@ Ivan <nekotekina@gmail.com>
 Ivan Filipov <159561759+vanaka11@users.noreply.github.com>
 Ivan Komarov <Ivan.Komarov@dfyz.info>
 Ivan Stepanov <ivanstepanovftw@gmail.com>
+JC <43374599+MrSMlT@users.noreply.github.com>
 JFLFY2255 <JFLFY2255@163.com>
 JH23X <165871467+JH23X@users.noreply.github.com>
 Jack Mousseau <jack@software.inc>
@@ -325,6 +345,7 @@ Jan Ploski <jpl@plosquare.com>
 Jannis Schönleber <joennlae@gmail.com>
 Jared Van Bortel <cebtenzzre@gmail.com>
 Jared Van Bortel <jared@nomic.ai>
+Jason C.H <ctrysbita@outlook.com>
 Jason McCartney <jmac@theroot.org>
 Jason Stillerman <jason.t.stillerman@gmail.com>
 Jean-Christophe Hoelt <hoelt@fovea.cc>
@@ -342,6 +363,7 @@ Jiahao Li <liplus17@163.com>
 Jian Liao <jianliao@users.noreply.github.com>
 JidongZhang-THU <1119708529@qq.com>
 Jinwoo Jeong <33892306+williamjeong2@users.noreply.github.com>
+Jinyang He <hejinyang@loongson.cn>
 Jiří Podivín <66251151+jpodivin@users.noreply.github.com>
 Jiří Sejkora <Sejseloid@gmail.com>
 Joan Fontanals <jfontanalsmartinez@gmail.com>
@@ -379,6 +401,7 @@ Justine Tunney <jtunney@mozilla.com>
 Juuso Alasuutari <juuso.alasuutari@gmail.com>
 KASR <karim.asrih@gmail.com>
 Kamil Tomšík <info@tomsik.cz>
+Kante Yin <kerthcet@gmail.com>
 Karol Kontny <82021046+kkontny@users.noreply.github.com>
 Karsten Weiss <knweiss@gmail.com>
 Karthick <j.karthic2004@gmail.com>
@@ -419,6 +442,7 @@ LoganDark <github@logandark.mozmail.com>
 Loïc Carrère <loic.carrere@gmail.com>
 LostRuins <39025047+LostRuins@users.noreply.github.com>
 LostRuins Concedo <39025047+LostRuins@users.noreply.github.com>
+Lucas Moura Belo <lucas.belo@live.com>
 Luciano <lucianostrika44@gmail.com>
 Luo Tian <lt@basecity.com>
 Lyle Dean <dean@lyle.dev>
@@ -463,6 +487,7 @@ Matthew Tejo <matthew.tejo@gmail.com>
 Matvey Soloviev <blackhole89@gmail.com>
 Max Krasnyansky <max.krasnyansky@gmail.com>
 Max Krasnyansky <quic_maxk@quicinc.com>
+Maxim Evtush <154841002+maximevtush@users.noreply.github.com>
 Maxime <672982+maximegmd@users.noreply.github.com>
 Maximilian Winter <maximilian.winter.91@gmail.com>
 Meng Zhang <meng@tabbyml.com>
@@ -494,6 +519,7 @@ Miwa / Ensan <63481257+ensan-hcl@users.noreply.github.com>
 Mohammadreza Hendiani <hendiani.mohammadreza@gmail.com>
 Mohammadreza Hendiani <mohammad.r.hendiani@gmail.com>
 Molly Sophia <mollysophia379@gmail.com>
+MoonRide303 <130458190+MoonRide303@users.noreply.github.com>
 MorganRO8 <47795945+MorganRO8@users.noreply.github.com>
 Murilo Santana <mvrilo@gmail.com>
 Musab Gultekin <musabgultekin@users.noreply.github.com>
@@ -524,6 +550,7 @@ Nikolas <127742645+nneubacher@users.noreply.github.com>
 Nindaleth <Nindaleth@users.noreply.github.com>
 Nuno <rare-magma@posteo.eu>
 OSecret <135510162+OLSecret@users.noreply.github.com>
+Oleksandr Kuvshynov <661042+okuvshynov@users.noreply.github.com>
 Oleksandr Nikitin <oleksandr@tvori.info>
 Oleksii Maryshchenko <oleksii.maryshchenko@gmail.com>
 Olivier Chafik <ochafik@users.noreply.github.com>
@@ -533,6 +560,7 @@ PAB <pierreantoine.bannier@gmail.com>
 Pablo Duboue <pablo.duboue@gmail.com>
 Pascal Patry <ppatry@mtacitlabs.com>
 Patrice Ferlet <metal3d@gmail.com>
+Patrick Peng <retr0@retr0.blog>
 Paul Tsochantaris <ptsochantaris@icloud.com>
 Pavel Zloi <github.com@drteam.rocks>
 Pavol Rusnak <pavol@rusnak.io>
@@ -549,6 +577,7 @@ Pieter Ouwerkerk <pieter.ouwerkerk@gmail.com>
 Plamen Minev <pacominev@gmail.com>
 Prashant Vithule <119530321+Vithulep@users.noreply.github.com>
 Przemysław Pawełczyk <przemoc@gmail.com>
+PureJourney <edward.pong@qq.com>
 Qin Yue Chen <71813199+chenqiny@users.noreply.github.com>
 Qingyou Meng <meng.qingyou@gmail.com>
 Qu Zongfu <43257352+yancaoweidaode@users.noreply.github.com>
@@ -564,14 +593,17 @@ Rand Xie <randxiexyy29@gmail.com>
 Randall Fitzgerald <randall@dasaku.net>
 Random Fly <renfei8@live.cn>
 Reinforce-II <fate@eastal.com>
+Rémy O <remyoudompheng@gmail.com>
 Rémy Oudompheng <oudomphe@phare.normalesup.org>
 Ren Xuancheng <jklj077@users.noreply.github.com>
 Rene Leonhardt <65483435+reneleonhardt@users.noreply.github.com>
 Reza Kakhki <rezakakhki.de@gmail.com>
+Reza Rahemtola <49811529+RezaRahemtola@users.noreply.github.com>
 RhinoDevel <RhinoDevel@users.noreply.github.com>
 Riccardo Orlando <Riccorl@users.noreply.github.com>
 Riceball LEE <snowyu.lee@gmail.com>
 Rich Dougherty <rich@rd.nz>
+Richard <r-burton@hotmail.co.uk>
 Richard Kiss <him@richardkiss.com>
 Richard Roberson <richardr1126@gmail.com>
 Rick G <26732651+TheFlipbook@users.noreply.github.com>
@@ -588,6 +620,7 @@ Robert Sung-wook Shin <edp1096@users.noreply.github.com>
 Robey Holderith <robey@flaminglunchbox.net>
 Robyn <robyngraf@users.noreply.github.com>
 Roger Meier <r.meier@siemens.com>
+Rohanjames1997 <rohan.james4@gmail.com>
 Roland <14355895+rbur0425@users.noreply.github.com>
 Romain Biessy <romain.biessy@codeplay.com>
 Romain D <90720+Artefact2@users.noreply.github.com>
@@ -610,6 +643,7 @@ Ryan Landay <rlanday@gmail.com>
 Ryder Wishart <ryderwishart@gmail.com>
 Ryuei <louixs@users.noreply.github.com>
 Rőczey Barnabás <31726601+An0nie@users.noreply.github.com>
+SAMI <samuel.koesnadi@stud.uni-due.de>
 SRHMorris <69468379+SRHMorris@users.noreply.github.com>
 SXX <sxx1136965276@gmail.com>
 SakuraUmi <yukinon244@gmail.com>
@@ -634,6 +668,8 @@ Shane A <shanea@allenai.org>
 Shangning Xu <32517059+xushangning@users.noreply.github.com>
 Shankar <gshankar.87@gmail.com>
 Shanshan Shen <467638484@qq.com>
+Shelby Jenkins <47464908+ShelbyJenkins@users.noreply.github.com>
+Sheldon Robinson <sheldon.robinson@live.com>
 Shijie <821898965@qq.com>
 Shintarou Okada <kokuzen@gmail.com>
 Shouzheng Liu <61452103+lshzh-ww@users.noreply.github.com>
@@ -713,18 +749,24 @@ Victor Nogueira <felladrin@gmail.com>
 Victor Z. Peng <ziliangdotme@gmail.com>
 Viet-Anh NGUYEN (Andrew) <vietanh.dev@gmail.com>
 Vinesh Janarthanan <36610342+VJHack@users.noreply.github.com>
+Vitali Lovich <vlovich+github@gmail.com>
+Vivian <vynride@gmail.com>
 Vlad <spitfireage@gmail.com>
 Vladimir <bogdad@gmail.com>
 Vladimir Malyutin <first-leon@yandex.ru>
+Vladimir Vuksanovic <109677816+vvuksanovic@users.noreply.github.com>
 Vladimir Zorin <vladimir@deviant.guru>
 VoidIsVoid <343750470@qq.com>
 Volodymyr Vitvitskyi <72226+signalpillar@users.noreply.github.com>
+Wagner Bruna <wbruna@users.noreply.github.com>
 Wang Qin <37098874+wangqin0@users.noreply.github.com>
 Wang Ran (汪然) <wangr@smail.nju.edu.cn>
 WangHaoranRobin <56047610+WangHaoranRobin@users.noreply.github.com>
 Weird Constructor <weirdconstructor@gmail.com>
+Weizhao Ouyang <o451686892@gmail.com>
 Welby Seely <welbyseely@gmail.com>
 Wentai Zhang <rchardx@gmail.com>
+Wilken Gottwalt <12194808+wgottwalt@users.noreply.github.com>
 WillCorticesAI <150854901+WillCorticesAI@users.noreply.github.com>
 William Tambellini <william.tambellini@gmail.com>
 William Tambellini <wtambellini@sdl.com>
@@ -816,6 +858,8 @@ chaihahaha <chai836275709@gmail.com>
 chiranko <96988916+chiranko@users.noreply.github.com>
 clibdev <52199778+clibdev@users.noreply.github.com>
 clyang <clyang@clyang.net>
+cmdr2 <secondary.cmdr2@gmail.com>
+cmdr2 <shashank.shekhar.global@gmail.com>
 cocktailpeanut <121128867+cocktailpeanut@users.noreply.github.com>
 codezjx <code.zjx@gmail.com>
 coezbek <c.oezbek@gmail.com>
@@ -835,6 +879,7 @@ deepdiffuser <112834445+deepdiffuser@users.noreply.github.com>
 devojony <61173062+devojony@users.noreply.github.com>
 ditsuke <ditsuke@protonmail.com>
 divinity76 <divinity76@gmail.com>
+dm4 <dm4@secondstate.io>
 dm4 <sunrisedm4@gmail.com>
 dotpy314 <33351922+dotpy314@users.noreply.github.com>
 drbh <david.richard.holtz@gmail.com>
@@ -849,6 +894,7 @@ fairydreaming <166155368+fairydreaming@users.noreply.github.com>
 fengerhu1 <2748250768@qq.com>
 fj-y-saito <85871716+fj-y-saito@users.noreply.github.com>
 fraxy-v <65565042+fraxy-v@users.noreply.github.com>
+fxzjshm <11426482+fxzjshm@users.noreply.github.com>
 github-actions[bot] <github-actions[bot]@users.noreply.github.com>
 gliptic <gliptic@users.noreply.github.com>
 gn64 <yukikaze.jp@gmail.com>
@@ -873,6 +919,7 @@ hydai <z54981220@gmail.com>
 iSma <ismail.senhaji@gmail.com>
 iacore <74560659+iacore@users.noreply.github.com>
 icppWorld <124377669+icppWorld@users.noreply.github.com>
+igardev <49397134+igardev@users.noreply.github.com>
 igarnier <igarnier@protonmail.com>
 intelmatt <61025942+intelmatt@users.noreply.github.com>
 iohub <rickyang.pro@gmail.com>
@@ -880,6 +927,7 @@ issixx <46835150+issixx@users.noreply.github.com>
 jacobi petrucciani <8117202+jpetrucciani@users.noreply.github.com>
 jaime-m-p <167997752+jaime-m-p@users.noreply.github.com>
 jameswu2014 <545426914@qq.com>
+jason_w <jason.wang@126.com>
 jdomke <28772296+jdomke@users.noreply.github.com>
 jiahao su <damow890@gmail.com>
 jiez <373447296@qq.com>
@@ -891,6 +939,7 @@ jon-chuang <9093549+jon-chuang@users.noreply.github.com>
 jp-x-g <jpxg-dev@protonmail.com>
 jukofyork <69222624+jukofyork@users.noreply.github.com>
 junchao-loongson <68935141+junchao-loongson@users.noreply.github.com>
+junchao-zhao <68935141+junchao-loongson@users.noreply.github.com>
 jwj7140 <32943891+jwj7140@users.noreply.github.com>
 k.h.lai <adrian.k.h.lai@outlook.com>
 kaizau <kaizau@users.noreply.github.com>
@@ -925,6 +974,7 @@ ltoniazzi <61414566+ltoniazzi@users.noreply.github.com>
 luoyu-intel <yu.luo@intel.com>
 m3ndax <adrian.goessl@outlook.com>
 maddes8cht <55592906+maddes8cht@users.noreply.github.com>
+magicse <magicse@users.noreply.github.com>
 mahorozte <41834471+mahorozte@users.noreply.github.com>
 makomk <makosoft@googlemail.com>
 manikbhandari <mbbhandarimanik2@gmail.com>
@@ -935,6 +985,7 @@ matt23654 <matthew.webber@protonmail.com>
 matteo <matteogeniaccio@yahoo.it>
 mdrokz <mohammadmunshi@gmail.com>
 mgroeber9110 <45620825+mgroeber9110@users.noreply.github.com>
+midnight <midnightmagic@users.noreply.github.com>
 minarchist <minarchist@users.noreply.github.com>
 mj-shifu <77107165+mj-shifu@users.noreply.github.com>
 mmyjona <jonathan.gonse@gmail.com>
@@ -958,10 +1009,12 @@ omahs <73983677+omahs@users.noreply.github.com>
 oobabooga <112222186+oobabooga@users.noreply.github.com>
 opparco <parco.opaai@gmail.com>
 ostix360 <55257054+ostix360@users.noreply.github.com>
+pascal-lc <49066376+pascal-lc@users.noreply.github.com>
 pculliton <phillipculliton@gmail.com>
 peidaqi <peidaqi@gmail.com>
 pengxin99 <pengxin.yuan@intel.com>
 perserk <perserk@gmail.com>
+petterreinholdtsen <pere-github@hungry.com>
 piDack <104877312+piDack@users.noreply.github.com>
 pmysl <piotr.myslinski@outlook.com>
 postmasters <namnguyen@google.com>
@@ -983,6 +1036,7 @@ semidark <me@semidark.net>
 serhii-nakon <57632032+serhii-nakon@users.noreply.github.com>
 sharpHL <132747147+sharpHL@users.noreply.github.com>
 shibe2 <shibe@tuta.io>
+simon886212 <37953122+simon886212@users.noreply.github.com>
 singularity <12184989+singularity-s0@users.noreply.github.com>
 sjinzh <sjinzh@gmail.com>
 sjxx <63994076+ylsdamxssjxxdd@users.noreply.github.com>
@@ -1000,10 +1054,12 @@ tarcey <cey.tarik@gmail.com>
 tc-mb <157115220+tc-mb@users.noreply.github.com>
 texmex76 <40733439+texmex76@users.noreply.github.com>
 thement <40525767+thement@users.noreply.github.com>
+theraininsky <76763719+theraininsky@users.noreply.github.com>
 thewh1teagle <61390950+thewh1teagle@users.noreply.github.com>
 tjohnman <tjohnman@users.noreply.github.com>
 toyer <2042519524@qq.com>
 tslmy <tslmy@users.noreply.github.com>
+tv1wnd <55383215+tv1wnd@users.noreply.github.com>
 ubik2 <ubik2@users.noreply.github.com>
 uint256_t <konndennsa@gmail.com>
 uint256_t <maekawatoshiki1017@gmail.com>
@@ -1014,6 +1070,7 @@ valiray <133289098+valiray@users.noreply.github.com>
 vb <vaibhavs10@gmail.com>
 vik <vikhyatk@gmail.com>
 viric <viric@viric.name>
+vmobilis <75476228+vmobilis@users.noreply.github.com>
 vodkaslime <646329483@qq.com>
 vvhg1 <94630311+vvhg1@users.noreply.github.com>
 vxiiduu <73044267+vxiiduu@users.noreply.github.com>
@@ -1028,6 +1085,8 @@ wzy <32936898+Freed-Wu@users.noreply.github.com>
 xaedes <xaedes@gmail.com>
 xaedes <xaedes@googlemail.com>
 xctan <axunlei@gmail.com>
+xiaobing318 <71554036+xiaobing318@users.noreply.github.com>
+xiaofei <hbuxiaofei@gmail.com>
 xloem <0xloem@gmail.com>
 yangli2 <yangli2@gmail.com>
 ymcki <84055651+ymcki@users.noreply.github.com>

CMakeLists.txt

@@ -29,6 +29,8 @@ else()
     set(LLAMA_STANDALONE OFF)
 endif()
 
+option(LLAMA_USE_SYSTEM_GGML "Use system libggml" OFF)
+
 if (EMSCRIPTEN)
     set(BUILD_SHARED_LIBS_DEFAULT OFF)
 
@@ -145,7 +147,13 @@ endif()
 # 3rd-party
 #
 
-if (NOT TARGET ggml)
+if (LLAMA_USE_SYSTEM_GGML)
+    message(STATUS "Using system-provided libggml, skipping ggml build")
+    find_package(ggml REQUIRED)
+    add_library(ggml ALIAS ggml::ggml)
+endif()
+
+if (NOT TARGET ggml AND NOT LLAMA_USE_SYSTEM_GGML)
     add_subdirectory(ggml)
     # ... otherwise assume ggml is added by a parent CMakeLists.txt
 endif()

Makefile

@@ -836,7 +836,7 @@ ifdef GGML_MUSA
 	else
 		MUSA_PATH ?= /opt/musa
 	endif
-	MUSA_ARCHITECTURES ?= 21;22
+	MUSA_ARCHITECTURES ?= 21;22;31
 
 	MK_CPPFLAGS += -DGGML_USE_MUSA -DGGML_USE_CUDA
 	MK_LDFLAGS += -L$(MUSA_PATH)/lib -Wl,-rpath=$(MUSA_PATH)/lib

Package.swift

@@ -1,19 +0,0 @@
-// swift-tools-version:5.5
-
-import PackageDescription
-
-let package = Package(
-    name: "llama",
-    platforms: [
-        .macOS(.v12),
-        .iOS(.v14),
-        .watchOS(.v4),
-        .tvOS(.v14)
-    ],
-    products: [
-        .library(name: "llama", targets: ["llama"]),
-    ],
-    targets: [
-        .systemLibrary(name: "llama", pkgConfig: "llama"),
-    ]
-)

README.md

@@ -5,7 +5,7 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![Server](https://github.com/ggml-org/llama.cpp/actions/workflows/server.yml/badge.svg)](https://github.com/ggml-org/llama.cpp/actions/workflows/server.yml)
 
-[Roadmap](https://github.com/users/ggml-org/projects/7) / [Project status](https://github.com/ggml-org/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggml-org/llama.cpp/discussions/205) / [ggml](https://github.com/ggml-org/ggml)
+[Roadmap](https://github.com/users/ggerganov/projects/7) / [Project status](https://github.com/ggml-org/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggml-org/llama.cpp/discussions/205) / [ggml](https://github.com/ggml-org/ggml)
 
 Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others) in pure C/C++
 
@@ -25,7 +25,7 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others)
 
 - **How to use [MTLResidencySet](https://developer.apple.com/documentation/metal/mtlresidencyset?language=objc) to keep the GPU memory active?** https://github.com/ggml-org/llama.cpp/pull/11427
 - **VS Code extension for FIM completions:** https://github.com/ggml-org/llama.vscode
-- Universal tool call support in `llama-server`: https://github.com/ggml-org/llama.cpp/pull/9639
+- Universal [tool call support](./docs/function-calling.md) in `llama-server` https://github.com/ggml-org/llama.cpp/pull/9639
 - Vim/Neovim plugin for FIM completions: https://github.com/ggml-org/llama.vim
 - Introducing GGUF-my-LoRA https://github.com/ggml-org/llama.cpp/discussions/10123
 - Hugging Face Inference Endpoints now support GGUF out of the box! https://github.com/ggml-org/llama.cpp/discussions/9669
@@ -157,6 +157,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - Guile Scheme: [guile_llama_cpp](https://savannah.nongnu.org/projects/guile-llama-cpp)
 - Swift [srgtuszy/llama-cpp-swift](https://github.com/srgtuszy/llama-cpp-swift)
 - Swift [ShenghaiWang/SwiftLlama](https://github.com/ShenghaiWang/SwiftLlama)
+- Delphi [Embarcadero/llama-cpp-delphi](https://github.com/Embarcadero/llama-cpp-delphi)
 
 </details>
 
@@ -171,6 +172,7 @@ Instructions for adding support for new models: [HOWTO-add-model.md](docs/develo
 - [eva](https://github.com/ylsdamxssjxxdd/eva) (MIT)
 - [iohub/collama](https://github.com/iohub/coLLaMA) (Apache-2.0)
 - [janhq/jan](https://github.com/janhq/jan) (AGPL)
+- [johnbean393/Sidekick](https://github.com/johnbean393/Sidekick) (MIT)
 - [KanTV](https://github.com/zhouwg/kantv?tab=readme-ov-file) (Apache-2.0)
 - [KodiBot](https://github.com/firatkiral/kodibot) (GPL)
 - [llama.vim](https://github.com/ggml-org/llama.vim) (MIT)

Sources/llama/llama.h

@@ -1,4 +0,0 @@
-#pragma once
-
-#include <llama.h>
-

Sources/llama/module.modulemap

@@ -1,5 +0,0 @@
-module llama [system] {
-    header "llama.h"
-    link "llama"
-    export *
-}

build-xcframework.sh

@@ -0,0 +1,519 @@
+#!/bin/bash
+#
+# Options
+IOS_MIN_OS_VERSION=16.4
+MACOS_MIN_OS_VERSION=13.3
+VISIONOS_MIN_OS_VERSION=1.0
+TVOS_MIN_OS_VERSION=16.4
+
+BUILD_SHARED_LIBS=OFF
+LLAMA_BUILD_EXAMPLES=OFF
+LLAMA_BUILD_TESTS=OFF
+LLAMA_BUILD_SERVER=OFF
+GGML_METAL=ON
+GGML_METAL_EMBED_LIBRARY=ON
+GGML_BLAS_DEFAULT=ON
+GGML_METAL_USE_BF16=ON
+GGML_OPENMP=OFF
+
+COMMON_C_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"
+COMMON_CXX_FLAGS="-Wno-macro-redefined -Wno-shorten-64-to-32 -Wno-unused-command-line-argument -g"
+
+# Common options for all builds
+COMMON_CMAKE_ARGS=(
+    -DCMAKE_XCODE_ATTRIBUTE_CODE_SIGNING_REQUIRED=NO
+    -DCMAKE_XCODE_ATTRIBUTE_CODE_SIGN_IDENTITY=""
+    -DCMAKE_XCODE_ATTRIBUTE_CODE_SIGNING_ALLOWED=NO
+    -DCMAKE_XCODE_ATTRIBUTE_DEBUG_INFORMATION_FORMAT="dwarf-with-dsym"
+    -DCMAKE_XCODE_ATTRIBUTE_GCC_GENERATE_DEBUGGING_SYMBOLS=YES
+    -DCMAKE_XCODE_ATTRIBUTE_COPY_PHASE_STRIP=NO
+    -DCMAKE_XCODE_ATTRIBUTE_STRIP_INSTALLED_PRODUCT=NO
+    -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
+    -DBUILD_SHARED_LIBS=${BUILD_SHARED_LIBS}
+    -DLLAMA_BUILD_EXAMPLES=${LLAMA_BUILD_EXAMPLES}
+    -DLLAMA_BUILD_TESTS=${LLAMA_BUILD_TESTS}
+    -DLLAMA_BUILD_SERVER=${LLAMA_BUILD_SERVER}
+    -DGGML_METAL_EMBED_LIBRARY=${GGML_METAL_EMBED_LIBRARY}
+    -DGGML_BLAS_DEFAULT=${GGML_BLAS_DEFAULT}
+    -DGGML_METAL=${GGML_METAL}
+    -DGGML_METAL_USE_BF16=${GGML_METAL_USE_BF16}
+    -DGGML_NATIVE=OFF
+    -DGGML_OPENMP=${GGML_OPENMP}
+)
+
+check_required_tool() {
+    local tool=$1
+    local install_message=$2
+
+    if ! command -v $tool &> /dev/null; then
+        echo "Error: $tool is required but not found."
+        echo "$install_message"
+        exit 1
+    fi
+}
+echo "Checking for required tools..."
+check_required_tool "cmake" "Please install CMake 3.28.0 or later (brew install cmake)"
+check_required_tool "xcodebuild" "Please install Xcode and Xcode Command Line Tools (xcode-select --install)"
+check_required_tool "libtool" "Please install libtool which should be available with Xcode Command Line Tools (CLT). Make sure Xcode CLT is installed (xcode-select --install)"
+check_required_tool "dsymutil" "Please install Xcode and Xcode Command Line Tools (xcode-select --install)"
+
+set -e
+
+## Clean up previous builds
+rm -rf build-apple
+rm -rf build-ios-sim
+rm -rf build-ios-device
+rm -rf build-macos
+rm -rf build-visionos
+rm -rf build-visionos-sim
+rm -rf build-tvos-sim
+rm -rf build-tvos-device
+
+# Setup the xcframework build directory structure
+setup_framework_structure() {
+    local build_dir=$1
+    local min_os_version=$2
+    local platform=$3  # "ios", "macos", "visionos", or "tvos"
+    local framework_name="llama"
+
+    echo "Creating ${platform}-style framework structure for ${build_dir}"
+
+    if [[ "$platform" == "macos" ]]; then
+        # macOS versioned structure uses versioned directories
+        mkdir -p ${build_dir}/framework/${framework_name}.framework/Versions/A/Headers
+        mkdir -p ${build_dir}/framework/${framework_name}.framework/Versions/A/Modules
+        mkdir -p ${build_dir}/framework/${framework_name}.framework/Versions/A/Resources
+
+        # Create symbolic links
+        ln -sf A ${build_dir}/framework/${framework_name}.framework/Versions/Current
+        ln -sf Versions/Current/Headers ${build_dir}/framework/${framework_name}.framework/Headers
+        ln -sf Versions/Current/Modules ${build_dir}/framework/${framework_name}.framework/Modules
+        ln -sf Versions/Current/Resources ${build_dir}/framework/${framework_name}.framework/Resources
+        ln -sf Versions/Current/${framework_name} ${build_dir}/framework/${framework_name}.framework/${framework_name}
+
+        # Set header and module paths
+        local header_path=${build_dir}/framework/${framework_name}.framework/Versions/A/Headers/
+        local module_path=${build_dir}/framework/${framework_name}.framework/Versions/A/Modules/
+    else
+        # iOS/VisionOS/tvOS use a flat structure
+        mkdir -p ${build_dir}/framework/${framework_name}.framework/Headers
+        mkdir -p ${build_dir}/framework/${framework_name}.framework/Modules
+
+        # Remove any existing structure to ensure clean build
+        rm -rf ${build_dir}/framework/${framework_name}.framework/Versions
+
+        # Set header and module paths
+        local header_path=${build_dir}/framework/${framework_name}.framework/Headers/
+        local module_path=${build_dir}/framework/${framework_name}.framework/Modules/
+    fi
+
+    # Copy all required headers (common for all platforms)
+    cp include/llama.h             ${header_path}
+    cp ggml/include/ggml.h         ${header_path}
+    cp ggml/include/ggml-alloc.h   ${header_path}
+    cp ggml/include/ggml-backend.h ${header_path}
+    cp ggml/include/ggml-metal.h   ${header_path}
+    cp ggml/include/ggml-cpu.h     ${header_path}
+    cp ggml/include/ggml-blas.h    ${header_path}
+    cp ggml/include/gguf.h         ${header_path}
+
+    # Create module map (common for all platforms)
+    cat > ${module_path}module.modulemap << EOF
+framework module llama {
+    header "llama.h"
+    header "ggml.h"
+    header "ggml-alloc.h"
+    header "ggml-backend.h"
+    header "ggml-metal.h"
+    header "ggml-cpu.h"
+    header "ggml-blas.h"
+    header "gguf.h"
+
+    link "c++"
+    link framework "Accelerate"
+    link framework "Metal"
+    link framework "Foundation"
+
+    export *
+}
+EOF
+
+    # Platform-specific settings for Info.plist
+    local platform_name=""
+    local sdk_name=""
+    local supported_platform=""
+
+    case "$platform" in
+        "ios")
+            platform_name="iphoneos"
+            sdk_name="iphoneos${min_os_version}"
+            supported_platform="iPhoneOS"
+            local plist_path="${build_dir}/framework/${framework_name}.framework/Info.plist"
+            local device_family='    <key>UIDeviceFamily</key>
+    <array>
+        <integer>1</integer>
+        <integer>2</integer>
+    </array>'
+            ;;
+        "macos")
+            platform_name="macosx"
+            sdk_name="macosx${min_os_version}"
+            supported_platform="MacOSX"
+            local plist_path="${build_dir}/framework/${framework_name}.framework/Versions/A/Resources/Info.plist"
+            local device_family=""
+            ;;
+        "visionos")
+            platform_name="xros"
+            sdk_name="xros${min_os_version}"
+            supported_platform="XRPlatform"
+            local plist_path="${build_dir}/framework/${framework_name}.framework/Info.plist"
+            local device_family=""
+            ;;
+        "tvos")
+            platform_name="appletvos"
+            sdk_name="appletvos${min_os_version}"
+            supported_platform="AppleTVOS"
+            local plist_path="${build_dir}/framework/${framework_name}.framework/Info.plist"
+            local device_family='    <key>UIDeviceFamily</key>
+    <array>
+        <integer>3</integer>
+    </array>'
+            ;;
+    esac
+
+    # Create Info.plist
+    cat > ${plist_path} << EOF
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+    <key>CFBundleDevelopmentRegion</key>
+    <string>en</string>
+    <key>CFBundleExecutable</key>
+    <string>llama</string>
+    <key>CFBundleIdentifier</key>
+    <string>org.ggml.llama</string>
+    <key>CFBundleInfoDictionaryVersion</key>
+    <string>6.0</string>
+    <key>CFBundleName</key>
+    <string>llama</string>
+    <key>CFBundlePackageType</key>
+    <string>FMWK</string>
+    <key>CFBundleShortVersionString</key>
+    <string>1.0</string>
+    <key>CFBundleVersion</key>
+    <string>1</string>
+    <key>MinimumOSVersion</key>
+    <string>${min_os_version}</string>
+    <key>CFBundleSupportedPlatforms</key>
+    <array>
+        <string>${supported_platform}</string>
+    </array>${device_family}
+    <key>DTPlatformName</key>
+    <string>${platform_name}</string>
+    <key>DTSDKName</key>
+    <string>${sdk_name}</string>
+</dict>
+</plist>
+EOF
+}
+
+# Create dynamic libraries from static libraries.
+combine_static_libraries() {
+    local build_dir="$1"
+    local release_dir="$2"
+    local platform="$3"  # "ios", "macos", "visionos", or "tvos"
+    local is_simulator="$4"
+    local base_dir="$(pwd)"
+    local framework_name="llama"
+
+    # Determine output path based on platform
+    local output_lib=""
+    if [[ "$platform" == "macos" ]]; then
+        # macOS uses versioned structure
+        output_lib="${build_dir}/framework/${framework_name}.framework/Versions/A/${framework_name}"
+    else
+        # iOS, visionOS, and tvOS use a directory flat structure
+        output_lib="${build_dir}/framework/${framework_name}.framework/${framework_name}"
+    fi
+
+    local libs=(
+        "${base_dir}/${build_dir}/src/${release_dir}/libllama.a"
+        "${base_dir}/${build_dir}/ggml/src/${release_dir}/libggml.a"
+        "${base_dir}/${build_dir}/ggml/src/${release_dir}/libggml-base.a"
+        "${base_dir}/${build_dir}/ggml/src/${release_dir}/libggml-cpu.a"
+        "${base_dir}/${build_dir}/ggml/src/ggml-metal/${release_dir}/libggml-metal.a"
+        "${base_dir}/${build_dir}/ggml/src/ggml-blas/${release_dir}/libggml-blas.a"
+    )
+
+    # Create temporary directory for processing
+    local temp_dir="${base_dir}/${build_dir}/temp"
+    mkdir -p "${temp_dir}"
+
+    # Since we have multiple architectures libtool will find object files that do not
+    # match the target architecture. We suppress these warnings.
+    libtool -static -o "${temp_dir}/combined.a" "${libs[@]}" 2> /dev/null
+
+    # Determine SDK, architectures, and install_name based on platform and simulator flag.
+    local sdk=""
+    local archs=""
+    local min_version_flag=""
+    local install_name=""
+
+    case "$platform" in
+        "ios")
+            if [[ "$is_simulator" == "true" ]]; then
+                sdk="iphonesimulator"
+                archs="arm64 x86_64"
+                min_version_flag="-mios-simulator-version-min=${IOS_MIN_OS_VERSION}"
+            else
+                sdk="iphoneos"
+                archs="arm64"
+                min_version_flag="-mios-version-min=${IOS_MIN_OS_VERSION}"
+            fi
+            install_name="@rpath/llama.framework/llama"
+            ;;
+        "macos")
+            sdk="macosx"
+            archs="arm64 x86_64"
+            min_version_flag="-mmacosx-version-min=${MACOS_MIN_OS_VERSION}"
+            install_name="@rpath/llama.framework/Versions/Current/llama"
+            ;;
+        "visionos")
+            if [[ "$is_simulator" == "true" ]]; then
+                sdk="xrsimulator"
+                archs="arm64 x86_64"
+                min_version_flag="-mtargetos=xros${VISIONOS_MIN_OS_VERSION}-simulator"
+            else
+                sdk="xros"
+                archs="arm64"
+                min_version_flag="-mtargetos=xros${VISIONOS_MIN_OS_VERSION}"
+            fi
+            # Use flat structure for visionOS, same as iOS
+            install_name="@rpath/llama.framework/llama"
+            ;;
+        "tvos")
+            if [[ "$is_simulator" == "true" ]]; then
+                sdk="appletvsimulator"
+                archs="arm64 x86_64"
+                min_version_flag="-mtvos-simulator-version-min=${TVOS_MIN_OS_VERSION}"
+            else
+                sdk="appletvos"
+                archs="arm64"
+                min_version_flag="-mtvos-version-min=${TVOS_MIN_OS_VERSION}"
+            fi
+            install_name="@rpath/llama.framework/llama"
+            ;;
+    esac
+
+    # Build architecture flags
+    local arch_flags=""
+    for arch in $archs; do
+        arch_flags+=" -arch $arch"
+    done
+
+    # Create dynamic library
+    echo "Creating dynamic library for ${platform}."
+    xcrun -sdk $sdk clang++ -dynamiclib \
+        -isysroot $(xcrun --sdk $sdk --show-sdk-path) \
+        $arch_flags \
+        $min_version_flag \
+        -Wl,-force_load,"${temp_dir}/combined.a" \
+        -framework Foundation -framework Metal -framework Accelerate \
+        -install_name "$install_name" \
+        -o "${base_dir}/${output_lib}"
+
+    # Platform-specific post-processing for device builds
+    if [[ "$is_simulator" == "false" ]]; then
+        if command -v vtool &>/dev/null; then
+            case "$platform" in
+                "ios")
+                    echo "Marking binary as a framework binary for iOS..."
+                    vtool -set-build-version ios ${IOS_MIN_OS_VERSION} ${IOS_MIN_OS_VERSION} -replace \
+                        -output "${base_dir}/${output_lib}" "${base_dir}/${output_lib}"
+                    ;;
+                "visionos")
+                    echo "Marking binary as a framework binary for visionOS..."
+                    vtool -set-build-version xros ${VISIONOS_MIN_OS_VERSION} ${VISIONOS_MIN_OS_VERSION} -replace \
+                        -output "${base_dir}/${output_lib}" "${base_dir}/${output_lib}"
+                    ;;
+                "tvos")
+                    echo "Marking binary as a framework binary for tvOS..."
+                    vtool -set-build-version tvos ${TVOS_MIN_OS_VERSION} ${TVOS_MIN_OS_VERSION} -replace \
+                        -output "${base_dir}/${output_lib}" "${base_dir}/${output_lib}"
+                    ;;
+            esac
+        else
+            echo "Warning: vtool not found. Binary may not pass App Store validation."
+        fi
+    fi
+
+    echo "Creating properly formatted dSYM..."
+    # Create a separate directory for dSYMs for all platforms
+    mkdir -p "${base_dir}/${build_dir}/dSYMs"
+
+    # iOS and visionOS style dSYM (flat structure)
+    if [[ "$platform" == "ios" || "$platform" == "visionos" || "$platform" == "tvos" ]]; then
+        # Generate dSYM in the dSYMs directory
+        xcrun dsymutil "${base_dir}/${output_lib}" -o "${base_dir}/${build_dir}/dSYMs/llama.dSYM"
+
+        # Create a copy of the binary that will be stripped
+        cp "${base_dir}/${output_lib}" "${temp_dir}/binary_to_strip"
+
+        # Strip debug symbols from the copy
+        xcrun strip -S "${temp_dir}/binary_to_strip" -o "${temp_dir}/stripped_lib"
+
+        # Replace the original with the stripped version
+        mv "${temp_dir}/stripped_lib" "${base_dir}/${output_lib}"
+    else
+        # macOS style dSYM
+        # First strip debug info to a separate file
+        xcrun strip -S "${base_dir}/${output_lib}" -o "${temp_dir}/stripped_lib"
+
+        # Generate dSYM in the dSYMs directory
+        xcrun dsymutil "${base_dir}/${output_lib}" -o "${base_dir}/${build_dir}/dSYMs/llama.dSYM"
+
+        # Replace original binary with stripped version
+        mv "${temp_dir}/stripped_lib" "${base_dir}/${output_lib}"
+    fi
+
+    # Remove any automatically generated dSYM files in the framework structure as they will
+    # otherwise case Invalid Bundle Structure validation errors.
+    if [ -d "${base_dir}/${output_lib}.dSYM" ]; then
+        echo "Removing generated dSYM file in framework structure: ${base_dir}/${output_lib}.dSYM"
+        rm -rf "${base_dir}/${output_lib}.dSYM"
+    fi
+
+    # Clean up
+    rm -rf "${temp_dir}"
+}
+
+echo "Building for iOS simulator..."
+cmake -B build-ios-sim -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${IOS_MIN_OS_VERSION} \
+    -DIOS=ON \
+    -DCMAKE_SYSTEM_NAME=iOS \
+    -DCMAKE_OSX_SYSROOT=iphonesimulator \
+    -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
+    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphonesimulator \
+    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-ios-sim --config Release -- -quiet
+
+echo "Building for iOS devices..."
+cmake -B build-ios-device -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${IOS_MIN_OS_VERSION} \
+    -DCMAKE_OSX_SYSROOT=iphoneos \
+    -DCMAKE_OSX_ARCHITECTURES="arm64" \
+    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphoneos \
+    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-ios-device --config Release -- -quiet
+
+echo "Building for macOS..."
+cmake -B build-macos -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${MACOS_MIN_OS_VERSION} \
+    -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
+    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-macos --config Release -- -quiet
+
+echo "Building for visionOS..."
+cmake -B build-visionos -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${VISIONOS_MIN_OS_VERSION} \
+    -DCMAKE_OSX_ARCHITECTURES="arm64" \
+    -DCMAKE_SYSTEM_NAME=visionOS \
+    -DCMAKE_OSX_SYSROOT=xros \
+    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=xros \
+    -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 -Du_int=unsigned\ int -Du_char=unsigned\ char -Du_short=unsigned\ short ${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 -Du_int=unsigned\ int -Du_char=unsigned\ char -Du_short=unsigned\ short ${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-visionos --config Release -- -quiet
+
+echo "Building for visionOS simulator..."
+cmake -B build-visionos-sim -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${VISIONOS_MIN_OS_VERSION} \
+    -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
+    -DCMAKE_SYSTEM_NAME=visionOS \
+    -DCMAKE_OSX_SYSROOT=xrsimulator \
+    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=xrsimulator \
+    -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 -Du_int=unsigned\ int -Du_char=unsigned\ char -Du_short=unsigned\ short ${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 -Du_int=unsigned\ int -Du_char=unsigned\ char -Du_short=unsigned\ short ${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-visionos-sim --config Release -- -quiet
+
+# Add tvOS builds (might need the same u_int definitions as watchOS and visionOS)
+echo "Building for tvOS simulator..."
+cmake -B build-tvos-sim -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${TVOS_MIN_OS_VERSION} \
+    -DCMAKE_SYSTEM_NAME=tvOS \
+    -DCMAKE_OSX_SYSROOT=appletvsimulator \
+    -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
+    -DGGML_METAL=ON \
+    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=appletvsimulator \
+    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-tvos-sim --config Release -- -quiet
+
+echo "Building for tvOS devices..."
+cmake -B build-tvos-device -G Xcode \
+    "${COMMON_CMAKE_ARGS[@]}" \
+    -DCMAKE_OSX_DEPLOYMENT_TARGET=${TVOS_MIN_OS_VERSION} \
+    -DCMAKE_SYSTEM_NAME=tvOS \
+    -DCMAKE_OSX_SYSROOT=appletvos \
+    -DCMAKE_OSX_ARCHITECTURES="arm64" \
+    -DGGML_METAL=ON \
+    -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=appletvos \
+    -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
+    -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -S .
+cmake --build build-tvos-device --config Release -- -quiet
+
+# Setup frameworks and copy binaries and headers
+echo "Setting up framework structures..."
+setup_framework_structure "build-ios-sim" ${IOS_MIN_OS_VERSION} "ios"
+setup_framework_structure "build-ios-device" ${IOS_MIN_OS_VERSION} "ios"
+setup_framework_structure "build-macos" ${MACOS_MIN_OS_VERSION} "macos"
+setup_framework_structure "build-visionos" ${VISIONOS_MIN_OS_VERSION} "visionos"
+setup_framework_structure "build-visionos-sim" ${VISIONOS_MIN_OS_VERSION} "visionos"
+setup_framework_structure "build-tvos-sim" ${TVOS_MIN_OS_VERSION} "tvos"
+setup_framework_structure "build-tvos-device" ${TVOS_MIN_OS_VERSION} "tvos"
+
+# Create dynamic libraries from static libraries
+echo "Creating dynamic libraries from static libraries..."
+combine_static_libraries "build-ios-sim" "Release-iphonesimulator" "ios" "true"
+combine_static_libraries "build-ios-device" "Release-iphoneos" "ios" "false"
+combine_static_libraries "build-macos" "Release" "macos" "false"
+combine_static_libraries "build-visionos" "Release-xros" "visionos" "false"
+combine_static_libraries "build-visionos-sim" "Release-xrsimulator" "visionos" "true"
+combine_static_libraries "build-tvos-sim" "Release-appletvsimulator" "tvos" "true"
+combine_static_libraries "build-tvos-device" "Release-appletvos" "tvos" "false"
+
+# Create XCFramework with correct debug symbols paths
+echo "Creating XCFramework..."
+xcodebuild -create-xcframework \
+    -framework $(pwd)/build-ios-sim/framework/llama.framework \
+    -debug-symbols $(pwd)/build-ios-sim/dSYMs/llama.dSYM \
+    -framework $(pwd)/build-ios-device/framework/llama.framework \
+    -debug-symbols $(pwd)/build-ios-device/dSYMs/llama.dSYM \
+    -framework $(pwd)/build-macos/framework/llama.framework \
+    -debug-symbols $(pwd)/build-macos/dSYMS/llama.dSYM \
+    -framework $(pwd)/build-visionos/framework/llama.framework \
+    -debug-symbols $(pwd)/build-visionos/dSYMs/llama.dSYM \
+    -framework $(pwd)/build-visionos-sim/framework/llama.framework \
+    -debug-symbols $(pwd)/build-visionos-sim/dSYMs/llama.dSYM \
+    -framework $(pwd)/build-tvos-device/framework/llama.framework \
+    -debug-symbols $(pwd)/build-tvos-device/dSYMs/llama.dSYM \
+    -framework $(pwd)/build-tvos-sim/framework/llama.framework \
+    -debug-symbols $(pwd)/build-tvos-sim/dSYMs/llama.dSYM \
+    -output $(pwd)/build-apple/llama.xcframework

ci/run.sh

@@ -352,10 +352,10 @@ function gg_run_open_llama_7b_v2 {
 
     (time ./bin/llama-imatrix --model ${model_f16} -f ${wiki_test} -t 1 -ngl 99 -c 2048 -b 512 --chunks 4 ) 2>&1 | tee -a $OUT/${ci}-imatrix.log
 
-    (time ./bin/llama-save-load-state--model ${model_q4_0} -ngl 10 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/llama-save-load-state--model ${model_q4_0} -ngl 10 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/llama-save-load-state--model ${model_q4_0} -ngl 99 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
-    (time ./bin/llama-save-load-state--model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 10 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0     ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
+    (time ./bin/llama-save-load-state --model ${model_q4_0} -ngl 99 -c 0 -fa ) 2>&1 | tee -a $OUT/${ci}-save-load-state.log
 
     function check_ppl {
         qnt="$1"

cmake/common.cmake

@@ -1,3 +1,5 @@
+include("ggml/cmake/common.cmake")
+
 function(llama_add_compile_flags)
     if (LLAMA_FATAL_WARNINGS)
         if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")

common/arg.cpp

@@ -764,7 +764,11 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_env("LLAMA_ARG_CTX_SIZE"));
     add_opt(common_arg(
         {"-n", "--predict", "--n-predict"}, "N",
-        string_format("number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)", params.n_predict),
+        string_format(
+            ex == LLAMA_EXAMPLE_MAIN || ex == LLAMA_EXAMPLE_INFILL
+                ? "number of tokens to predict (default: %d, -1 = infinity, -2 = until context filled)"
+                : "number of tokens to predict (default: %d, -1 = infinity)",
+            params.n_predict),
         [](common_params & params, int value) {
             params.n_predict = value;
         }
@@ -849,6 +853,20 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             }
         }
     ).set_excludes({LLAMA_EXAMPLE_SERVER}));
+    add_opt(common_arg(
+        {"-sysf", "--system-prompt-file"}, "FNAME",
+        "a file containing the system prompt (default: none)",
+        [](common_params & params, const std::string & value) {
+            std::ifstream file(value);
+            if (!file) {
+                throw std::runtime_error(string_format("error: failed to open file '%s'\n", value.c_str()));
+            }
+            std::copy(std::istreambuf_iterator<char>(file), std::istreambuf_iterator<char>(), back_inserter(params.system_prompt));
+            if (!params.system_prompt.empty() && params.system_prompt.back() == '\n') {
+                params.system_prompt.pop_back();
+            }
+        }
+    ).set_examples({LLAMA_EXAMPLE_MAIN}));
     add_opt(common_arg(
         {"--in-file"}, "FNAME",
         "an input file (repeat to specify multiple files)",
@@ -949,6 +967,15 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
             params.conversation_mode = COMMON_CONVERSATION_MODE_DISABLED;
         }
     ).set_examples({LLAMA_EXAMPLE_MAIN}));
+    add_opt(common_arg(
+        {"-st", "--single-turn"},
+        "run conversation for a single turn only, then exit when done\n"
+        "will not be interactive if first turn is predefined with --prompt\n"
+        "(default: false)",
+        [](common_params & params) {
+            params.single_turn = true;
+        }
+    ).set_examples({LLAMA_EXAMPLE_MAIN}));
     add_opt(common_arg(
         {"-i", "--interactive"},
         string_format("run in interactive mode (default: %s)", params.interactive ? "true" : "false"),
@@ -1858,18 +1885,11 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_examples({LLAMA_EXAMPLE_PASSKEY}));
     add_opt(common_arg(
         {"-o", "--output", "--output-file"}, "FNAME",
-        string_format("output file (default: '%s')",
-            ex == LLAMA_EXAMPLE_EXPORT_LORA
-                ? params.lora_outfile.c_str()
-                : ex == LLAMA_EXAMPLE_CVECTOR_GENERATOR
-                    ? params.cvector_outfile.c_str()
-                    : params.out_file.c_str()),
+        string_format("output file (default: '%s')", params.out_file.c_str()),
         [](common_params & params, const std::string & value) {
             params.out_file = value;
-            params.cvector_outfile = value;
-            params.lora_outfile = value;
         }
-    ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA}));
+    ).set_examples({LLAMA_EXAMPLE_IMATRIX, LLAMA_EXAMPLE_CVECTOR_GENERATOR, LLAMA_EXAMPLE_EXPORT_LORA, LLAMA_EXAMPLE_TTS}));
     add_opt(common_arg(
         {"-ofreq", "--output-frequency"}, "N",
         string_format("output the imatrix every N iterations (default: %d)", params.n_out_freq),
@@ -2562,5 +2582,43 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         }
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
 
+    add_opt(common_arg(
+        {"--fim-qwen-7b-spec"},
+        string_format("use Qwen 2.5 Coder 7B + 0.5B draft for speculative decoding (note: can download weights from the internet)"),
+        [](common_params & params) {
+            params.hf_repo = "ggml-org/Qwen2.5-Coder-7B-Q8_0-GGUF";
+            params.hf_file = "qwen2.5-coder-7b-q8_0.gguf";
+            params.speculative.hf_repo = "ggml-org/Qwen2.5-Coder-0.5B-Q8_0-GGUF";
+            params.speculative.hf_file = "qwen2.5-coder-0.5b-q8_0.gguf";
+            params.speculative.n_gpu_layers = 99;
+            params.port = 8012;
+            params.n_gpu_layers = 99;
+            params.flash_attn = true;
+            params.n_ubatch = 1024;
+            params.n_batch = 1024;
+            params.n_ctx = 0;
+            params.n_cache_reuse = 256;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}));
+
+    add_opt(common_arg(
+        {"--fim-qwen-14b-spec"},
+        string_format("use Qwen 2.5 Coder 14B + 0.5B draft for speculative decoding (note: can download weights from the internet)"),
+        [](common_params & params) {
+            params.hf_repo = "ggml-org/Qwen2.5-Coder-14B-Q8_0-GGUF";
+            params.hf_file = "qwen2.5-coder-14b-q8_0.gguf";
+            params.speculative.hf_repo = "ggml-org/Qwen2.5-Coder-0.5B-Q8_0-GGUF";
+            params.speculative.hf_file = "qwen2.5-coder-0.5b-q8_0.gguf";
+            params.speculative.n_gpu_layers = 99;
+            params.port = 8012;
+            params.n_gpu_layers = 99;
+            params.flash_attn = true;
+            params.n_ubatch = 1024;
+            params.n_batch = 1024;
+            params.n_ctx = 0;
+            params.n_cache_reuse = 256;
+        }
+    ).set_examples({LLAMA_EXAMPLE_SERVER}));
+
     return ctx_arg;
 }

common/chat.cpp

@@ -60,7 +60,9 @@ std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messa
             }
             msg.role = message.at("role");
 
-            if (message.contains("content")) {
+            auto has_content = message.contains("content");
+            auto has_tool_calls = message.contains("tool_calls");
+            if (has_content) {
                 const auto & content = message.at("content");
                 if (content.is_string()) {
                     msg.content = content;
@@ -81,19 +83,8 @@ std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messa
                 } else if (!content.is_null()) {
                     throw std::runtime_error("Invalid 'content' type: expected string or array, got " + content.dump() + " (ref: https://github.com/ggml-org/llama.cpp/issues/8367)");
                 }
-            } else {
-                throw std::runtime_error("Expected 'content' (ref: https://github.com/ggml-org/llama.cpp/issues/8367)");
             }
-            if (message.contains("reasoning_content")) {
-                msg.reasoning_content = message.at("reasoning_content");
-            }
-            if (message.contains("name")) {
-                msg.tool_name = message.at("name");
-            }
-            if (message.contains("tool_call_id")) {
-                msg.tool_call_id = message.at("tool_call_id");
-            }
-            if (message.contains("tool_calls")) {
+            if (has_tool_calls) {
                 for (const auto & tool_call : message.at("tool_calls")) {
                     common_chat_tool_call tc;
                     if (!tool_call.contains("type")) {
@@ -118,6 +109,18 @@ std::vector<common_chat_msg> common_chat_msgs_parse_oaicompat(const json & messa
                     msg.tool_calls.push_back(tc);
                 }
             }
+            if (!has_content && !has_tool_calls) {
+                throw std::runtime_error("Expected 'content' or 'tool_calls' (ref: https://github.com/ggml-org/llama.cpp/issues/8367 & https://github.com/ggml-org/llama.cpp/issues/12279)");
+            }
+            if (message.contains("reasoning_content")) {
+                msg.reasoning_content = message.at("reasoning_content");
+            }
+            if (message.contains("name")) {
+                msg.tool_name = message.at("name");
+            }
+            if (message.contains("tool_call_id")) {
+                msg.tool_call_id = message.at("tool_call_id");
+            }
 
             msgs.push_back(msg);
         }
@@ -442,6 +445,7 @@ std::string common_chat_format_name(common_chat_format format) {
         case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2: return "Functionary v3.2";
         case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1: return "Functionary v3.1 Llama 3.1";
         case COMMON_CHAT_FORMAT_HERMES_2_PRO: return "Hermes 2 Pro";
+        case COMMON_CHAT_FORMAT_HERMES_2_PRO_EXTRACT_REASONING: return "Hermes 2 Pro (extract reasoning)";
         case COMMON_CHAT_FORMAT_COMMAND_R7B: return "Command R7B";
         case COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING: return "Command R7B (extract reasoning)";
         default:
@@ -449,12 +453,6 @@ std::string common_chat_format_name(common_chat_format format) {
     }
 }
 
-const common_grammar_options grammar_options {
-    /* .dotall = */ false,
-    /* .compact_spaces = */ false,
-    // /* .compact_spaces = */ true,
-};
-
 static bool parse_json(std::string::const_iterator & it, const std::string::const_iterator & end, json & out) {
     // // https://json.nlohmann.me/features/parsing/sax_interface/
     struct json_error_locator : public nlohmann::json_sax<json> {
@@ -500,6 +498,34 @@ static bool parse_json(std::string::const_iterator & it, const std::string::cons
     }
 }
 
+static bool parse_literal(std::string::const_iterator & it, const std::string::const_iterator & end, const std::string & expected) {
+    auto expected_it = expected.begin();
+    auto tmp_it = it;
+    while (tmp_it != end && expected_it != expected.end() && *tmp_it == *expected_it) {
+        ++tmp_it;
+        ++expected_it;
+    }
+    if (expected_it == expected.end()) {
+        it = tmp_it;
+        return true;
+    }
+    return false;
+}
+
+static std::optional<std::smatch> parse_pattern(std::string::const_iterator & it, const std::string::const_iterator & end, const std::regex & expected) {
+    std::smatch match;
+    if (std::regex_match(it, end, match, expected)) {
+        it = match.suffix().first;
+        return match;
+    }
+    return std::nullopt;
+}
+
+static void consume_spaces(std::string::const_iterator & it, const std::string::const_iterator & end) {
+    while (it != end && std::isspace(*it)) {
+        ++it;
+    }
+}
 
 /**
  * Takes a prefix regex that must have 1 group to capture the function name, a closing suffix, and expects json parameters in between.
@@ -509,7 +535,8 @@ static common_chat_msg parse_json_tool_calls(
     const std::string& input,
     const std::optional<std::regex> & trigger_opt,
     const std::regex & function_regex,
-    const std::regex & close_regex) {
+    const std::regex & close_regex,
+    bool allow_raw_python = false) {
     std::smatch match;
 
     common_chat_msg result;
@@ -540,14 +567,19 @@ static common_chat_msg parse_json_tool_calls(
         it = rit->suffix().first;
 
         json arguments;
-        if (!parse_json(it, end, arguments)) {
+        if (parse_json(it, end, arguments)) {
+            if (!std::regex_search(it, end, match, close_regex)) {
+                throw std::runtime_error("Malformed input, missing closing pattern: " + input);
+            }
+            it = match.suffix().first;
+            result.tool_calls.push_back({name, arguments.is_string() ? arguments.get<std::string>() : arguments.dump(), /* id= */ ""});
+        } else {
+            if (allow_raw_python && name == "python") {
+                result.tool_calls.push_back({name, json({{"code", std::string(it, end)}}).dump(), /* id= */ ""});
+                break;
+            }
             throw std::runtime_error("Failed to parse json tool call arguments: " + input);
         }
-        if (!std::regex_search(it, end, match, close_regex)) {
-            throw std::runtime_error("Malformed input, missing closing pattern: " + input);
-        }
-        it = match.suffix().first;
-        result.tool_calls.push_back({name, arguments.is_string() ? arguments.get<std::string>() : arguments.dump(), /* id= */ ""});
     }
 
     if (!result.tool_calls.empty()) {
@@ -559,29 +591,29 @@ static common_chat_msg parse_json_tool_calls(
     return result;
 }
 
+static common_chat_tool_call process_tool_call(const json & tool_call) {
+    const auto & arguments = tool_call.at("arguments");
+    return {
+        /* .name = */ tool_call.at("name"),
+        /* .arguments = */ arguments.is_string() ? arguments.get<std::string>() : arguments.dump(),
+        /* .id = */ tool_call.contains("id") ? tool_call.at("id") : "",
+    };
+}
 static common_chat_msg parse_prefixed_json_tool_call_array(const std::string& input, const std::string & prefix, size_t rstrip_prefix = 0) {
     auto content_end = input.find(prefix);
     size_t tc_start = std::string::npos;
 
     common_chat_msg result;
     result.role = "assistant";
-    const auto process_tool_calls = [&](const json & tool_calls) {
-        for (const auto & tool_call : tool_calls) {
-            const auto & arguments = tool_call.at("arguments");
-            result.tool_calls.push_back({
-                tool_call.at("name"),
-                arguments.is_string() ? arguments.get<std::string>() : arguments.dump(),
-                tool_call.contains("id") ? tool_call.at("id") : "",
-            });
-        }
-    };
     if (content_end == std::string::npos) {
         result.content = input;
     } else {
         tc_start = content_end + prefix.size() - rstrip_prefix;
         result.content = input.substr(0, content_end);
         auto tool_calls = json::parse(input.substr(tc_start));
-        process_tool_calls(tool_calls);
+        for (const auto & tool_call : tool_calls) {
+            result.tool_calls.emplace_back(process_tool_call(tool_call));
+        }
     }
     return result;
 }
@@ -700,7 +732,7 @@ static common_chat_params common_chat_params_init_generic(const common_chat_temp
     data.grammar_lazy = false;
     data.grammar = build_grammar([&](const common_grammar_builder & builder) {
         builder.add_schema("root", schema);
-    }, grammar_options);
+    });
 
     auto tweaked_messages = common_chat_template::add_system(
         inputs.messages,
@@ -770,8 +802,11 @@ static common_chat_params common_chat_params_init_mistral_nemo(const common_chat
             schema["maxItems"] = 1;
         }
         builder.add_rule("root", "\"[TOOL_CALLS]\" " + builder.add_schema("tool_calls", schema));
-    }, grammar_options);
-    data.grammar_triggers.push_back({"[TOOL_CALLS]", /* .at_start = */ true});
+    });
+    data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "[TOOL_CALLS]"});
+    data.preserved_tokens = {
+        "[TOOL_CALLS]",
+    };
     data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
     data.format = COMMON_CHAT_FORMAT_MISTRAL_NEMO;
     return data;
@@ -813,14 +848,18 @@ static common_chat_params common_chat_params_init_command_r7b(const common_chat_
             schema["maxItems"] = 1;
         }
         builder.add_rule("root", "\"<|START_ACTION|>\" " + builder.add_schema("tool_calls", schema) + " \"<|END_ACTION|>\"");
-    }, grammar_options);
-    data.grammar_triggers.push_back({"<|START_ACTION|>", /* .at_start = */ false});
+    });
+    data.grammar_triggers.push_back({
+        COMMON_GRAMMAR_TRIGGER_TYPE_WORD,
+        "<|START_ACTION|>",
+    });
     data.preserved_tokens = {
+        "<|START_ACTION|>",
+        "<|END_ACTION|>",
         "<|START_RESPONSE|>",
         "<|END_RESPONSE|>",
         "<|START_THINKING|>",
         "<|END_THINKING|>",
-        "<|END_ACTION|>",
     };
     auto adjusted_messages = json::array();
     for (const auto & msg : inputs.messages) {
@@ -840,9 +879,9 @@ static common_chat_params common_chat_params_init_command_r7b(const common_chat_
     return data;
 }
 static common_chat_msg common_chat_parse_command_r7b(const std::string & input, bool extract_reasoning) {
-    static std::regex thought_regex("(<\\|START_THINKING\\|>([\\s\\S\\n\\r]*?)<\\|END_THINKING\\|>)([\\s\\S\\n\\r]*)");
-    static std::regex action_regex("<\\|START_ACTION\\|>([\\s\\S\\n\\r]*?)<\\|END_ACTION\\|>");
-    static std::regex response_regex("(?:<\\|START_RESPONSE\\|>)?([\\s\\S\\n\\r]*?)<\\|END_RESPONSE\\|>");
+    static const std::regex thought_regex("(<\\|START_THINKING\\|>([\\s\\S]*?)<\\|END_THINKING\\|>)([\\s\\S]*)");
+    static const std::regex action_regex("<\\|START_ACTION\\|>([\\s\\S]*?)<\\|END_ACTION\\|>");
+    static const std::regex response_regex("(?:<\\|START_RESPONSE\\|>)?([\\s\\S]*?)<\\|END_RESPONSE\\|>");
 
     std::smatch match;
 
@@ -945,23 +984,23 @@ static common_chat_params common_chat_params_init_llama_3_1_tool_calls(const com
                 builder.add_rule(
                     name + "-call",
                     "\"{\" space "
-                    "( \"\\\"type\\\":\" space \"\\\"function\\\",\" space )? "
-                    "\"\\\"name\\\": \\\"" + name + "\\\", \\\"parameters\\\": \" " +
-                        builder.add_schema(name + "-args", parameters) +
-                    " \"}\""));
-            data.grammar_triggers.push_back({"{\"name\": \"" + name + "\"", /* .at_start = */ true});
+                    "( \"\\\"type\\\"\"       space \":\" space \"\\\"function\\\"\"     space \",\" space )? "
+                    "  \"\\\"name\\\"\"       space \":\" space \"\\\"" + name + "\\\"\" space \",\" space "
+                    "  \"\\\"parameters\\\"\" space \":\" space " + builder.add_schema(name + "-args", parameters) + " "
+                    "\"}\" space"));
+        });
+        // Small models may hallucinate function names so we match anything (*at the start*) that looks like the JSON of a function call, regardless of the name.
+        data.grammar_triggers.push_back({
+            COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START,
+            "\\{\\s*(?:\"type\"\\s*:\\s*\"function\"\\s*,\\s*)?\"name\"\\s*:\\s*\"", // + name + "\"[\\s\\S]*",
         });
-        data.grammar_triggers.push_back({"{\"name\":", /* .at_start = */ true});
-        data.grammar_triggers.push_back({"{\n  \"name\":", /* .at_start = */ true});
-        data.grammar_triggers.push_back({"{\n    \"name\":", /* .at_start = */ true});
-        data.grammar_triggers.push_back({"{\"type\": \"function\"", /* .at_start = */ true});
-        data.grammar_triggers.push_back({"{\n  \"type\": \"function\"", /* .at_start = */ true});
-        data.grammar_triggers.push_back({"{\n    \"type\": \"function\"", /* .at_start = */ true});
         if (!builtin_tools.empty()) {
-            data.grammar_triggers.push_back({"<|python_tag|>", /* .at_start = */ false});
+            data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<|python_tag|>"});
+            data.preserved_tokens.push_back("<|python_tag|>");
         }
+        // Allow a few empty lines on top of the usual constrained json schema space rule.
         builder.add_rule("root", string_join(tool_rules, " | "));
-    }, grammar_options);
+    });
     data.additional_stops.push_back("<|eom_id|>");
     data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt, {
         {"tools_in_user_message", false},
@@ -974,33 +1013,33 @@ static common_chat_params common_chat_params_init_llama_3_1_tool_calls(const com
 }
 static common_chat_msg common_chat_parse_llama_3_1(const std::string & input, bool with_builtin_tools = false) {
     // TODO: tighten & simplify the parser, don't accept leading text context.
-    static std::regex function_regex("\\{[\\s\\n\\r]*(?:\"type\"[\\s\\n\\r]*:[\\s\\n\\r]*\"function\"[\\s\\n\\r]*,[\\s\\n\\r]*|[\\s\\n\\r]*)\"name\"[\\s\\n\\r]*:[\\s\\n\\r]*\"([^\"]+)\"[\\s\\n\\r]*,[\\s\\n\\r]*\"parameters\": ");
-    static std::regex close_regex("\\}");
-    static std::regex builtin_call_regex("<\\|python_tag\\|>([^.(]+)\\.call\\((.*)\\)");
+    static const std::regex function_regex(
+        "\\s*\\{\\s*(?:\"type\"\\s*:\\s*\"function\"\\s*,\\s*)?\"name\"\\s*:\\s*\"([^\"]+)\"\\s*,\\s*\"parameters\"\\s*: ");
+    static const std::regex close_regex("\\}\\s*");
+    static const std::regex builtin_call_regex("<\\|python_tag\\|>\\s*([^.(]+)\\s*\\.\\s*call\\s*\\(\\s*([\\w]+)\\s*=\\s*([\\s\\S]*?)\\)");
 
     if (with_builtin_tools) {
         std::smatch match;
         if (std::regex_match(input, match, builtin_call_regex)) {
-            auto name = match[1].str();
-            auto raw_args = match[2].str();
+            try {
+                auto name = match[1].str();
+                auto arg_name = match[2].str();
+                auto arg_value_str = match[3].str();
+                auto arg_value = json::parse(arg_value_str);
 
-            // TODO: if/when builtin tools start accepting more than 1 argument, use parse_json for real parsing.
-            auto it_eq = raw_args.find('=');
-            auto arg_name = raw_args.substr(0, it_eq);
-            auto arg_value_str = raw_args.substr(it_eq + 1);
-            auto arg_value = json::parse(arg_value_str);
-
-            common_chat_msg msg;
-            msg.role = "assistant";
-            msg.content = match.prefix().str();
-            msg.tool_calls.push_back({
-                /* .name = */ name,
-                /* .arguments = */ (json {
-                    {arg_name, arg_value},
-                }).dump(),
-                /* .id = */ "",
-            });
-            return msg;
+                common_chat_msg msg;
+                msg.role = "assistant";
+                msg.tool_calls.push_back({
+                    /* .name = */ name,
+                    /* .arguments = */ (json {
+                        {arg_name, arg_value},
+                    }).dump(),
+                    /* .id = */ "",
+                });
+                return msg;
+            } catch (const std::exception & e) {
+                LOG_WRN("Failed to parse builtin tool call arguments (%s): %s", e.what(), input.c_str());
+            }
         }
     }
     return parse_json_tool_calls(input, std::nullopt, function_regex, close_regex);
@@ -1017,10 +1056,10 @@ static common_chat_params common_chat_params_init_deepseek_r1(const common_chat_
                 std::string name = function.at("name");
                 auto parameters = function.at("parameters");
                 builder.resolve_refs(parameters);
-                auto args_rule = builder.add_schema(name + "-args", parameters);
                 tool_rules.push_back(builder.add_rule(name + "-call",
                     "\"<｜tool▁call▁begin｜>function<｜tool▁sep｜>" + name + "\\n"
-                    "```json\\n\" " + args_rule + " \"```<｜tool▁call▁end｜>\""));
+                    "```json\\n\" " + builder.add_schema(name + "-args", parameters) + " "
+                    "\"```<｜tool▁call▁end｜>\""));
             });
             // Distill Qwen 7B & 32B models seem confused re/ syntax of their tool call opening tag,
             // so we accept common variants (then it's all constrained)
@@ -1029,18 +1068,20 @@ static common_chat_params common_chat_params_init_deepseek_r1(const common_chat_
                 "(" + string_join(tool_rules, " | ") + ")" + (inputs.parallel_tool_calls ? "*" : "") + " "
                 "\"<｜tool▁calls▁end｜>\""
                 " space");
-            data.grammar_triggers.push_back({"<｜tool▁calls▁begin｜>", /* .at_start = */ false});
-            data.grammar_triggers.push_back({"<｜tool_calls_begin｜>", /* .at_start = */ false});
-            data.grammar_triggers.push_back({"<｜tool calls begin｜>", /* .at_start = */ false});
-            data.grammar_triggers.push_back({"<｜tool\\_calls\\_begin｜>", /* .at_start = */ false});
+            data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<｜tool▁calls▁begin｜>"});
+            data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<｜tool_calls_begin｜>"});
+            data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<｜tool calls begin｜>"});
+            data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<｜tool\\_calls\\_begin｜>"});
             data.preserved_tokens = {
                 "<think>",
                 "</think>",
+                "<｜tool▁calls▁begin｜>",
+                "<｜tool▁call▁begin｜>",
                 "<｜tool▁sep｜>",
-                "<｜tool▁calls▁end｜",
                 "<｜tool▁call▁end｜>",
+                "<｜tool▁calls▁end｜",
             };
-        }, grammar_options);
+        });
     }
     auto prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
 
@@ -1065,34 +1106,42 @@ static common_chat_params common_chat_params_init_deepseek_r1(const common_chat_
     data.format = inputs.extract_reasoning ? COMMON_CHAT_FORMAT_DEEPSEEK_R1_EXTRACT_REASONING : COMMON_CHAT_FORMAT_DEEPSEEK_R1;
     return data;
 }
-static common_chat_msg common_chat_parse_deepseek_r1(const std::string & input, bool extract_reasoning) {
-    static std::regex function_regex("<｜tool▁call▁begin｜>function<｜tool▁sep｜>([^\n]+)\n```json\n");
-    static std::regex close_regex("```[\\s\\r\\n]*<｜tool▁call▁end｜>");
-    static std::regex reasoning_content_regex("((?:<think>)?([\\s\\S\\r\\n]*?)</think>)?([\\s\\S\\r\\n]*)");
-    static std::regex tool_calls_regex("[\\s\\r\\n]*(?:<｜tool▁calls▁begin｜>|<｜tool_calls_begin｜>|<｜tool calls begin｜>|<｜tool\\\\_calls\\\\_begin｜>)([\\s\\S\\r\\n]*?)<｜tool▁calls▁end｜>");
-    common_chat_msg msg;
-    msg.role = "assistant";
+static common_chat_msg handle_think_tag_prelude(const std::string & input, bool extract_reasoning, const std::function<common_chat_msg(const std::string &)> & rest_parser) {
     std::smatch match;
+    static const std::regex reasoning_content_regex("((?:<think>)?([\\s\\S\\r\\n]*?)</think>)?([\\s\\S\\r\\n]*)");
     if (std::regex_match(input, match, reasoning_content_regex)) {
-        std::string rest;
+        auto rest = match[3].str();
+        auto msg = rest_parser(rest);
+        auto reasoning_content = string_strip(match[2].str());
         if (extract_reasoning) {
-            msg.reasoning_content = string_strip(match[2].str());
-        } else {
-            msg.content = match[1].str();
+            msg.reasoning_content = reasoning_content;
+        } else if (!reasoning_content.empty()) {
+            std::ostringstream content;
+            content << "<think>" << reasoning_content << "</think>" << msg.content;
+            msg.content = content.str();
         }
-        rest = match[3].str();
+        return msg;
+    }
+    return rest_parser(input);
+}
+static common_chat_msg common_chat_parse_deepseek_r1(const std::string & input, bool extract_reasoning) {
+    return handle_think_tag_prelude(input, extract_reasoning, [](const std::string & input) {
+        static const std::regex function_regex("<｜tool▁call▁begin｜>function<｜tool▁sep｜>([^\n]+)\n```json\n");
+        static const std::regex close_regex("```[\\s\\r\\n]*<｜tool▁call▁end｜>");
+        static const std::regex tool_calls_regex("[\\s\\r\\n]*(?:<｜tool▁calls▁begin｜>|<｜tool_calls_begin｜>|<｜tool calls begin｜>|<｜tool\\\\_calls\\\\_begin｜>)([\\s\\S\\r\\n]*?)<｜tool▁calls▁end｜>");
 
-        if (std::regex_search(rest, match, tool_calls_regex)) {
+        common_chat_msg msg;
+        msg.role = "assistant";
+        std::smatch match;
+        if (std::regex_search(input, match, tool_calls_regex)) {
             auto tool_calls = match[1].str();
             auto msg2 = parse_json_tool_calls(tool_calls, std::nullopt, function_regex, close_regex);
             msg.tool_calls = std::move(msg2.tool_calls);
         } else {
-            msg.content += std::string(rest.begin() + rest.find_first_not_of(" \r\n"), rest.end());
+            msg.content = input;
         }
-    } else {
-        msg.content = input;
-    }
-    return msg;
+        return msg;
+    });
 }
 
 static common_chat_params common_chat_params_init_firefunction_v2(const common_chat_template & tmpl, const struct templates_params & inputs) {
@@ -1129,8 +1178,11 @@ static common_chat_params common_chat_params_init_firefunction_v2(const common_c
                 schema["maxItems"] = 1;
             }
             builder.add_rule("root", "\" functools\"? " + builder.add_schema("tool_calls", schema));
-        }, grammar_options);
-        data.grammar_triggers.push_back({" functools[", /* .at_start = */ false});
+        });
+        data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, " functools["});
+        data.preserved_tokens = {
+            " functools[",
+        };
         data.format = COMMON_CHAT_FORMAT_FIREFUNCTION_V2;
     } else {
         data.format = COMMON_CHAT_FORMAT_CONTENT_ONLY;
@@ -1158,11 +1210,28 @@ static common_chat_params common_chat_params_init_functionary_v3_2(const common_
                 auto parameters = function.at("parameters");
                 builder.resolve_refs(parameters);
                 auto args_rule = builder.add_schema(name + "-args", parameters);
-                first_tool_rules.push_back(builder.add_rule(name + "-call", "\"" + name + "\\n\" " + args_rule));
+                first_tool_rules.push_back(builder.add_rule(name + "-call", "( \"assistant<|end_header_id|>\\n\" )? \"" + name + "\\n\" " + args_rule));
                 subsequent_tool_rules.push_back(builder.add_rule(name + "-call2", "\">>>" + name + "\\n\" " + args_rule));
-                data.grammar_triggers.push_back({name, /* .at_start = */ true});
-                data.grammar_triggers.push_back({">>>" + name, /* .at_start = */ false});
+                data.grammar_triggers.push_back({
+                    COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START,
+                    regex_escape(name + "\n"),
+                });
+                data.grammar_triggers.push_back({
+                    COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START,
+                    regex_escape("assistant<|end_header_id|>\n" + name + "\n"),
+                });
+                data.grammar_triggers.push_back({
+                    COMMON_GRAMMAR_TRIGGER_TYPE_WORD,
+                    regex_escape(">>>" + name + "\n"),
+                });
+                data.grammar_triggers.push_back({
+                    COMMON_GRAMMAR_TRIGGER_TYPE_WORD,
+                    ">>>assistant<|end_header_id|>\n" + name,
+                });
             });
+            data.preserved_tokens = {
+                "<|end_header_id|>",
+            };
             auto first_rule = first_tool_rules.empty() ? "" : builder.add_rule("first_tool_call", string_join(first_tool_rules, " | ")) + " space";
             if (inputs.parallel_tool_calls) {
                 auto subsequent_rule = builder.add_rule("subsequent_tool_call", string_join(subsequent_tool_rules, " | ")) + " space";
@@ -1171,34 +1240,20 @@ static common_chat_params common_chat_params_init_functionary_v3_2(const common_
                 builder.add_rule("root", first_rule);
             }
 
-        }, grammar_options);
+        });
     }
     return data;
 }
 
-static bool consume(std::string::const_iterator & it, const std::string::const_iterator & end, const std::string & expected) {
-    auto expected_it = expected.begin();
-    auto tmp_it = it;
-    while (tmp_it != end && expected_it != expected.end() && *tmp_it == *expected_it) {
-        ++tmp_it;
-        ++expected_it;
-    }
-    if (expected_it == expected.end()) {
-        it = tmp_it;
-        return true;
-    }
-    return false;
-}
-
 static common_chat_msg common_chat_parse_functionary_v3_2(const std::string & input) {
-    static std::regex function_regex(R"((?:>>>)?(\w+)\n)");
-    static std::regex close_regex(R"($|(?=>>>))");
+    static const std::regex function_regex(R"((?:>>>)?(?:assistant<|end_header_id|>\n)?(\w+)\n)");
+    static const std::regex close_regex(R"($|(?=>>>))");
 
     std::string content;
     auto it = input.begin();
     const auto end = input.end();
 
-    if (consume(it, end, "all\n")) {
+    if (parse_literal(it, end, "all\n")) {
         std::smatch match;
         if (std::regex_search(it, end, match, function_regex)) {
             auto fun_it = match.prefix().second;
@@ -1213,7 +1268,7 @@ static common_chat_msg common_chat_parse_functionary_v3_2(const std::string & in
     }
     // TODO: tighten & simplify.
     try {
-        auto res = parse_json_tool_calls(std::string(it, end), std::nullopt, function_regex, close_regex);
+        auto res = parse_json_tool_calls(std::string(it, end), std::nullopt, function_regex, close_regex, /* allow_raw_python= */ true);
         res.content = content + res.content;
         return res;
     } catch (const std::exception & e) {
@@ -1266,12 +1321,13 @@ static common_chat_params common_chat_params_init_functionary_v3_1_llama_3_1(con
         });
         if (has_raw_python) {
             tool_rules.push_back(builder.add_rule("python-call", "\"<|python_tag|>\" .*"));
-            data.grammar_triggers.push_back({"<|python_tag|>", /* .at_start = */ false});
+            data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<|python_tag|>"});
+            data.preserved_tokens.push_back("<|python_tag|>");
         }
         auto tool_call = builder.add_rule("tool_call", string_join(tool_rules, " | ")) + " space";
         builder.add_rule("root", inputs.parallel_tool_calls ? "(" + tool_call + ")+" : tool_call);
-        data.grammar_triggers.push_back({"<function=", /* .at_start = */ false});
-    }, grammar_options);
+        data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<function="});
+    });
 
     data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
     // TODO: if (has_raw_python)
@@ -1280,7 +1336,7 @@ static common_chat_params common_chat_params_init_functionary_v3_1_llama_3_1(con
 }
 static common_chat_msg common_chat_parse_functionary_v3_1_llama_3_1(const std::string & input) {
     // This version of Functionary still supports the llama 3.1 tool call format for the python tool.
-    static std::regex python_tag_regex(R"(<\|python_tag\|>([\s\S\n]*)$)");
+    static const std::regex python_tag_regex(R"(<\|python_tag\|>([\s\S\n]*)$)");
     std::smatch match;
     if (std::regex_search(input, match, python_tag_regex)) {
         auto code = match[1].str();
@@ -1294,8 +1350,8 @@ static common_chat_msg common_chat_parse_functionary_v3_1_llama_3_1(const std::s
         });
         return msg;
     }
-    static std::regex function_regex(R"(<function=(\w+)>)");
-    static std::regex close_regex(R"(</function>)");
+    static const std::regex function_regex(R"(<function=(\w+)>)");
+    static const std::regex close_regex(R"(</function>)");
     // TODO: tighten & simplify.
     return parse_json_tool_calls(input, std::nullopt, function_regex, close_regex);
 }
@@ -1306,6 +1362,7 @@ static common_chat_params common_chat_params_init_hermes_2_pro(const common_chat
     data.grammar_lazy = inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_REQUIRED;
     data.grammar = build_grammar([&](const common_grammar_builder & builder) {
         std::vector<std::string> tool_rules;
+        std::vector<std::string> tool_call_alts;
         foreach_function(inputs.tools, [&](const json & tool) {
             const auto & function = tool.at("function");
             std::string name = function.at("name");
@@ -1319,68 +1376,187 @@ static common_chat_params common_chat_params_init_hermes_2_pro(const common_chat
                 }},
                 {"required", json::array({"name", "arguments"})},
             }));
+            tool_call_alts.push_back(builder.add_rule(
+                name + "-function-tag",
+                "\"<function\" ( \"=" + name + "\" | \" name=\\\"" + name + "\\\"\" ) \">\" space " +
+                builder.add_schema(name + "-args", parameters) + " "
+                "\"</function>\" space"));
+
+            data.grammar_triggers.push_back({
+                COMMON_GRAMMAR_TRIGGER_TYPE_WORD,
+                "<function=" + name + ">",
+            });
+            auto escaped_name = regex_escape(name);
+            data.grammar_triggers.push_back({
+                COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN,
+                "<function\\s+name\\s*=\\s*\"" + escaped_name + "\"",
+            });
         });
-        auto tool_call = "\"<tool_call>\" space " + builder.add_rule("tool_call", string_join(tool_rules, " | ")) + " \"</tool_call>\" space";
+        auto any_tool_call = builder.add_rule("any_tool_call", "( " + string_join(tool_rules, " | ") + " ) space");
+        std::vector<std::string> alt_tags {
+            any_tool_call,
+            "\"<tool_call>\" space "     + any_tool_call + " \"</tool_call>\"",
+            // The rest is just to accommodate common "good bad" outputs.
+            "\"<function_call>\" space " + any_tool_call + " \"</function_call>\"",
+            "\"<response>\"  space "     + any_tool_call + " \"</response>\"",
+            "\"<tools>\"     space "     + any_tool_call + " \"</tools>\"",
+            "\"<json>\"      space "     + any_tool_call + " \"</json>\"",
+            "\"<xml>\"      space "     + any_tool_call + " \"</xml>\"",
+            "\"<JSON>\"      space "     + any_tool_call + " \"</JSON>\"",
+        };
+        auto wrappable_tool_call = builder.add_rule("wrappable_tool_call", "( " + string_join(alt_tags, " | ") + " ) space");
+        tool_call_alts.push_back(wrappable_tool_call);
+        tool_call_alts.push_back(
+            "( \"```\\n\" | \"```json\\n\" | \"```xml\\n\" ) space " + wrappable_tool_call + " space \"```\" space ");
+        auto tool_call = builder.add_rule("tool_call", string_join(tool_call_alts, " | "));
         builder.add_rule("root", inputs.parallel_tool_calls ? "(" + tool_call + ")+" : tool_call);
-        data.grammar_triggers.push_back({"<tool_call>", /* .at_start = */ false});
-        data.preserved_tokens = { "</tool_call>" };
-    }, grammar_options);
+        data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<tool_call>"});
+        data.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, "<function"});
+        // Trigger on some common known "good bad" outputs (only from the start and with a json that's about a specific argument name to avoid false positives)
+        data.grammar_triggers.push_back({
+            COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START,
+            "(?:```(?:json|xml)?\n\\s*)?(?:<function_call>|<tools>|<xml><json>|<response>)?\\s*\\{\\s*\"", //name\"\\s*:\\s*\"" + escaped_name + "\"",
+        });
+        data.preserved_tokens = {
+            "<think>",
+            "</think>",
+            "<tool_call>",
+            "</tool_call>",
+            "<function",
+            "<tools>",
+            "</tools>",
+            "<response>",
+            "</response>",
+            "<function_call>",
+            "</function_call>",
+            "<json>",
+            "</json>",
+            "<JSON>",
+            "</JSON>",
+            "```",
+            "```json",
+            "```xml",
+        };
+    });
 
     data.prompt = apply(tmpl, inputs.messages, inputs.tools.empty() ? json() : inputs.tools, inputs.add_generation_prompt);
-    data.format = COMMON_CHAT_FORMAT_HERMES_2_PRO;
+    data.format = inputs.extract_reasoning ? COMMON_CHAT_FORMAT_HERMES_2_PRO_EXTRACT_REASONING : COMMON_CHAT_FORMAT_HERMES_2_PRO;
     return data;
 }
-static common_chat_msg common_chat_parse_hermes_2_pro(const std::string & input) {
-    try {
-        std::regex start_pattern(R"([\n\s]*<tool_call>)");
-        std::regex middle_pattern(R"([\n\s]*</tool_call>[\n\s]*<tool_call>)");
-        std::regex end_pattern(R"([\n\s]*</tool_call>[\n\s]*$)");
+static common_chat_msg common_chat_parse_hermes_2_pro(const std::string& input, bool extract_reasoning) {
+    return handle_think_tag_prelude(input, extract_reasoning, [](const std::string & input) {
+        static const std::regex open_regex(
+            "(?:"
+            "(```(?:xml|json)?\\n\\s*)?"         // match 1 (block_start)
+            "(<tool_call>"                   // match 2 (open_tag)
+            "|<function_call>"
+            "|<tool>"
+            "|<tools>"
+            "|<response>"
+            "|<json>"
+            "|<xml>"
+            "|<JSON>"
+            ")?"
+            "(\\s*\\{\\s*\"name\"\\s*:[\\s\\S]*)"    // match 3 (named tool call + rest)
+            ")"
+            "|"
+            "(?:<function=([^>]+)>"            // match 4 (function name)
+            "|<function name=\"([^\"]+)\">)" // match 5 (function name again)
+            "([\\s\\S]*)"                   // match 6 (function arguments + rest)})"
+        );
 
-        common_chat_msg msg;
-        msg.role = "assistant";
+        try {
+            common_chat_msg msg;
+            msg.role = "assistant";
 
-        auto end = input.end();
-        std::sregex_iterator rend;
-        std::sregex_iterator rit(input.begin(), end, start_pattern);
-        if (rit == rend) {
+            std::string::const_iterator it = input.begin();
+            const std::string::const_iterator end = input.end();
+            std::smatch match;
+
+            while (it != end) {
+                if (std::regex_search(it, end, match, open_regex)) {
+                    // Add content before the match
+                    msg.content += std::string(it, match[0].first);
+
+                    auto block_start = match[1].str();
+                    std::string block_end = block_start.empty() ? "" : "```";
+
+                    auto open_tag = match[2].str();
+                    std::string close_tag;
+
+                    if (match[3].matched) {
+                        close_tag = open_tag.empty() ? "" : "</" + open_tag.substr(1);
+                        auto json_it = match[3].first;
+                        json tool_call;
+                        if (parse_json(json_it, end, tool_call) && tool_call.contains("name") && tool_call.contains("arguments")) {
+
+                            msg.tool_calls.emplace_back(process_tool_call(tool_call));
+                            it = json_it;  // Move iterator past parsed JSON
+
+                            // Handle close tags
+                            consume_spaces(it, end);
+                            if (!close_tag.empty() && !parse_literal(it, end, close_tag)) {
+                                throw std::runtime_error("Failed to parse closing tag");
+                            }
+                            consume_spaces(it, end);
+                            if (!block_end.empty() && !parse_literal(it, end, block_end)) {
+                                throw std::runtime_error("Failed to parse block end");
+                            }
+                            consume_spaces(it, end);
+                        } else {
+                            // Not a valid tool call, treat as content
+                            msg.content += std::string(match[0].first, match[0].second);
+                            it = match[0].second;
+                        }
+                    } else {
+                        auto function_name = match[4].str();
+                        if (function_name.empty()) {
+                            function_name = match[5].str();
+                        }
+                        GGML_ASSERT(!function_name.empty());
+
+                        close_tag = "</function>";
+                        // Start parsing from after the opening tags
+                        auto json_it = match[6].first;
+                        json arguments;
+                        if (parse_json(json_it, end, arguments)) {
+                            msg.tool_calls.emplace_back(process_tool_call({
+                                {"name", function_name},
+                                {"arguments", arguments},
+                            }));
+                            it = json_it;  // Move iterator past parsed JSON
+
+                            // Handle close tags
+                            consume_spaces(it, end);
+                            if (!close_tag.empty() && !parse_literal(it, end, close_tag)) {
+                                throw std::runtime_error("Failed to parse closing tag");
+                            }
+                            consume_spaces(it, end);
+                            if (!block_end.empty() && !parse_literal(it, end, block_end)) {
+                                throw std::runtime_error("Failed to parse block end");
+                            }
+                            consume_spaces(it, end);
+                        } else {
+                            // Not a valid tool call, treat as content
+                            msg.content += std::string(match[0].first, match[0].second);
+                            it = match[0].second;
+                        }
+                    }
+                } else {
+                    // Add remaining content
+                    msg.content += std::string(it, end);
+                    break;
+                }
+            }
+            return msg;
+        } catch (const std::exception & e) {
+            LOG_ERR("Failed to parse hermes 2 pro input: %s\n", e.what());
+            common_chat_msg msg;
+            msg.role = "assistant";
             msg.content = input;
             return msg;
         }
-
-        msg.content = rit->prefix();
-
-        auto it = rit->suffix().first;
-        while (it != end) {
-            json call;
-            if (!parse_json(it, end, call)) {
-                throw std::runtime_error("Failed to parse json tool call");
-            }
-            const auto & arguments = call.at("arguments");
-            msg.tool_calls.push_back({
-                call.at("name"),
-                arguments.dump(),
-                // arguments.is_string() ? arguments.get<std::string>() : arguments.dump(),
-                /* id= */ "",
-            });
-            rit = {it, end, middle_pattern};
-            if (rit != rend) {
-                it = rit->suffix().first;
-            } else {
-                rit = {it, end, end_pattern};
-                if (rit == rend) {
-                    throw std::runtime_error("Malformed input, missing </tool_call>");
-                }
-                break;
-            }
-        }
-        return msg;
-    } catch (const std::exception & e) {
-        LOG_ERR("Failed to parse hermes 2 pro input: %s\n", e.what());
-        common_chat_msg msg;
-        msg.role = "assistant";
-        msg.content = input;
-        return msg;
-    }
+    });
 }
 
 static common_chat_params common_chat_params_init_without_tools(const common_chat_template & tmpl, const struct templates_params & inputs) {
@@ -1445,6 +1621,11 @@ static common_chat_params common_chat_templates_apply_jinja(
         return common_chat_params_init_command_r7b(tmpl, params);
     }
 
+    // Hermes 2/3 Pro, Qwen 2.5 Instruct (w/ tools)
+    if (src.find("<tool_call>") != std::string::npos && params.json_schema.is_null()) {
+        return common_chat_params_init_hermes_2_pro(tmpl, params);
+    }
+
     // Use generic handler when mixing tools + JSON schema.
     // TODO: support that mix in handlers below.
     if ((params.tools.is_array() && params.json_schema.is_object())) {
@@ -1466,11 +1647,6 @@ static common_chat_params common_chat_templates_apply_jinja(
         return common_chat_params_init_without_tools(tmpl, params);
     }
 
-    // Hermes 2/3 Pro, Qwen 2.5 Instruct (w/ tools)
-    if (src.find("<tool_call>") != std::string::npos) {
-        return common_chat_params_init_hermes_2_pro(tmpl, params);
-    }
-
     // Functionary v3.1 (w/ tools)
     if (src.find("<|start_header_id|>") != std::string::npos
         && src.find("<function=") != std::string::npos) {
@@ -1588,7 +1764,9 @@ common_chat_msg common_chat_parse(const std::string & input, common_chat_format
         case COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1:
             return common_chat_parse_functionary_v3_1_llama_3_1(input);
         case COMMON_CHAT_FORMAT_HERMES_2_PRO:
-            return common_chat_parse_hermes_2_pro(input);
+            return common_chat_parse_hermes_2_pro(input, /* extract_reasoning= */ false);
+        case COMMON_CHAT_FORMAT_HERMES_2_PRO_EXTRACT_REASONING:
+            return common_chat_parse_hermes_2_pro(input, /* extract_reasoning= */ true);
         case COMMON_CHAT_FORMAT_FIREFUNCTION_V2:
             return common_chat_parse_firefunction_v2(input);
         case COMMON_CHAT_FORMAT_COMMAND_R7B:

common/chat.h

@@ -53,6 +53,7 @@ enum common_chat_format {
     COMMON_CHAT_FORMAT_FUNCTIONARY_V3_2,
     COMMON_CHAT_FORMAT_FUNCTIONARY_V3_1_LLAMA_3_1,
     COMMON_CHAT_FORMAT_HERMES_2_PRO,
+    COMMON_CHAT_FORMAT_HERMES_2_PRO_EXTRACT_REASONING,
     COMMON_CHAT_FORMAT_COMMAND_R7B,
     COMMON_CHAT_FORMAT_COMMAND_R7B_EXTRACT_REASONING,
 

common/common.cpp

@@ -10,7 +10,6 @@
 // Change JSON_ASSERT from assert() to GGML_ASSERT:
 #define JSON_ASSERT GGML_ASSERT
 #include "json.hpp"
-#include "json-schema-to-grammar.h"
 #include "llama.h"
 
 #include <algorithm>
@@ -483,6 +482,11 @@ void string_replace_all(std::string & s, const std::string & search, const std::
     s = std::move(builder);
 }
 
+std::string regex_escape(const std::string & s) {
+    static const std::regex special_chars("[.^$|()*+?\\[\\]{}\\\\]");
+    return std::regex_replace(s, special_chars, "\\$0");
+}
+
 std::string string_join(const std::vector<std::string> & values, const std::string & separator) {
     std::ostringstream result;
     for (size_t i = 0; i < values.size(); ++i) {
@@ -951,8 +955,8 @@ struct common_init_result common_init_from_params(common_params & params) {
         return iparams;
     }
 
-    if (params.ctx_shift && !llama_kv_cache_can_shift(lctx)) {
-        LOG_WRN("%s: KV cache shifting is not supported for this model, disabling KV cache shifting\n", __func__);
+    if (params.ctx_shift && !llama_kv_self_can_shift(lctx)) {
+        LOG_WRN("%s: KV cache shifting is not supported for this context, disabling KV cache shifting\n", __func__);
         params.ctx_shift = false;
     }
 
@@ -1029,6 +1033,8 @@ struct common_init_result common_init_from_params(common_params & params) {
     if (params.warmup) {
         LOG_WRN("%s: warming up the model with an empty run - please wait ... (--no-warmup to disable)\n", __func__);
 
+        llama_set_warmup(lctx, true);
+
         std::vector<llama_token> tmp;
         llama_token bos = llama_vocab_bos(vocab);
         llama_token eos = llama_vocab_eos(vocab);
@@ -1056,9 +1062,10 @@ struct common_init_result common_init_from_params(common_params & params) {
         if (llama_model_has_decoder(model)) {
             llama_decode(lctx, llama_batch_get_one(tmp.data(), std::min(tmp.size(), (size_t) params.n_batch)));
         }
-        llama_kv_cache_clear(lctx);
+        llama_kv_self_clear(lctx);
         llama_synchronize(lctx);
         llama_perf_context_reset(lctx);
+        llama_set_warmup(lctx, false);
     }
 
     iparams.model.reset(model);
@@ -2026,3 +2033,25 @@ common_control_vector_data common_control_vector_load(const std::vector<common_c
     return result;
 }
 
+template <>
+json common_grammar_trigger::to_json() const {
+    json out {
+        {"type", (int) type},
+        {"value", value},
+    };
+    if (type == COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN) {
+        out["token"] = (int) token;
+    }
+    return out;
+}
+
+template <>
+common_grammar_trigger common_grammar_trigger::from_json(const json & in) {
+    common_grammar_trigger out;
+    out.type = (common_grammar_trigger_type) in.at("type").get<int>();
+    out.value = in.at("value").get<std::string>();
+    if (out.type == COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN) {
+        out.token = (llama_token) in.at("token").get<int>();
+    }
+    return out;
+}

common/common.h

@@ -110,9 +110,21 @@ enum common_conversation_mode {
     COMMON_CONVERSATION_MODE_AUTO     = 2,
 };
 
+enum common_grammar_trigger_type {
+    COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN,
+    COMMON_GRAMMAR_TRIGGER_TYPE_WORD,
+    COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN,
+    COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START,
+};
+
 struct common_grammar_trigger {
-    std::string word;
-    bool at_start;
+    common_grammar_trigger_type type;
+    std::string value;
+    llama_token token = LLAMA_TOKEN_NULL;
+
+    // T can only be nlohmann::ordered_json
+    template <class T> T to_json() const;
+    template <class T> static common_grammar_trigger from_json(const T & in);
 };
 
 // sampling parameters
@@ -163,8 +175,7 @@ struct common_params_sampling {
 
     std::string                         grammar; // optional BNF-like grammar to constrain sampling
     bool                                grammar_lazy = false;
-    std::vector<common_grammar_trigger> grammar_trigger_words;  // optional trigger words to trigger lazy grammar
-    std::vector<llama_token>            grammar_trigger_tokens; // optional trigger tokens to trigger lazy grammar and print trigger special tokens.
+    std::vector<common_grammar_trigger> grammar_triggers; // optional triggers (for lazy grammars)
     std::set<llama_token>               preserved_tokens;
 
     std::vector<llama_logit_bias> logit_bias; // logit biases to apply
@@ -328,6 +339,8 @@ struct common_params {
     bool warmup            = true;  // warmup run
     bool check_tensors     = false; // validate tensor data
 
+    bool single_turn       = false; // single turn chat conversation
+
     ggml_type cache_type_k = GGML_TYPE_F16; // KV cache data type for the K
     ggml_type cache_type_v = GGML_TYPE_F16; // KV cache data type for the V
 
@@ -394,8 +407,6 @@ struct common_params {
     int32_t i_pos  = -1;  // position of the passkey in the junk text
 
     // imatrix params
-    std::string out_file = "imatrix.dat"; // save the resulting imatrix to this file
-
     int32_t n_out_freq  = 10; // output the imatrix every n_out_freq iterations
     int32_t n_save_freq =  0; // save the imatrix every n_save_freq iterations
     int32_t i_chunk     =  0; // start processing from this chunk
@@ -407,16 +418,16 @@ struct common_params {
     int n_pca_batch = 100;
     int n_pca_iterations = 1000;
     dimre_method cvector_dimre_method = DIMRE_METHOD_PCA;
-    std::string cvector_outfile       = "control_vector.gguf";
     std::string cvector_positive_file = "examples/cvector-generator/positive.txt";
     std::string cvector_negative_file = "examples/cvector-generator/negative.txt";
 
     bool spm_infill = false; // suffix/prefix/middle pattern for infill
 
-    std::string lora_outfile = "ggml-lora-merged-f16.gguf";
-
     // batched-bench params
     bool batched_bench_output_jsonl = false;
+
+    // common params
+    std::string out_file; // output filename for all example programs
 };
 
 // call once at the start of a program if it uses libcommon
@@ -456,6 +467,8 @@ std::string string_repeat(const std::string & str, size_t n);
 
 void string_replace_all(std::string & s, const std::string & search, const std::string & replace);
 
+std::string regex_escape(const std::string & s);
+
 template<class T>
 static std::vector<T> string_split(const std::string & str, char delim) {
     static_assert(!std::is_same<T, std::string>::value, "Please use the specialized version for std::string");

common/json-schema-to-grammar.cpp

@@ -264,7 +264,7 @@ static void _build_min_max_int(int min_value, int max_value, std::stringstream &
     throw std::runtime_error("At least one of min_value or max_value must be set");
 }
 
-const std::string SPACE_RULE = "| \" \" | \"\\n\" [ \\t]{0,20}";
+const std::string SPACE_RULE = "| \" \" | \"\\n\"{1,2} [ \\t]{0,20}";
 
 struct BuiltinRule {
     std::string content;
@@ -764,11 +764,10 @@ private:
 public:
     SchemaConverter(
         const std::function<json(const std::string &)> & fetch_json,
-        bool dotall,
-        bool compact_spaces)
+        bool dotall)
           : _fetch_json(fetch_json), _dotall(dotall)
     {
-        _rules["space"] = compact_spaces ? "\" \"?" : SPACE_RULE;
+        _rules["space"] = SPACE_RULE;
     }
 
     void resolve_refs(json & schema, const std::string & url) {
@@ -1007,7 +1006,7 @@ std::string json_schema_to_grammar(const json & schema, bool force_gbnf) {
 }
 
 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options) {
-    SchemaConverter converter([&](const std::string &) { return json(); }, options.dotall, options.compact_spaces);
+    SchemaConverter converter([&](const std::string &) { return json(); }, options.dotall);
     common_grammar_builder builder {
         /* .add_rule = */ [&](const std::string & name, const std::string & rule) {
             return converter._add_rule(name, rule);

common/json-schema-to-grammar.h

@@ -16,7 +16,6 @@ struct common_grammar_builder {
 
 struct common_grammar_options {
     bool dotall = false;
-    bool compact_spaces = false;
 };
 
 std::string build_grammar(const std::function<void(const common_grammar_builder &)> & cb, const common_grammar_options & options = {});

common/minja/minja.hpp

@@ -1378,13 +1378,27 @@ struct ArgumentsExpression {
     }
 };
 
-static std::string strip(const std::string & s) {
-  auto start = s.find_first_not_of(" \t\n\r");
+static std::string strip(const std::string & s, const std::string & chars = "", bool left = true, bool right = true) {
+  auto charset = chars.empty() ? " \t\n\r" : chars;
+  auto start = left ? s.find_first_not_of(charset) : 0;
   if (start == std::string::npos) return "";
-  auto end = s.find_last_not_of(" \t\n\r");
+  auto end = right ? s.find_last_not_of(charset) : s.size() - 1;
   return s.substr(start, end - start + 1);
 }
 
+static std::vector<std::string> split(const std::string & s, const std::string & sep) {
+  std::vector<std::string> result;
+  size_t start = 0;
+  size_t end = s.find(sep);
+  while (end != std::string::npos) {
+    result.push_back(s.substr(start, end - start));
+    start = end + sep.length();
+    end = s.find(sep, start);
+  }
+  result.push_back(s.substr(start));
+  return result;
+}
+
 static std::string capitalize(const std::string & s) {
   if (s.empty()) return s;
   auto result = s;
@@ -1467,8 +1481,26 @@ public:
         } else if (obj.is_string()) {
           auto str = obj.get<std::string>();
           if (method->get_name() == "strip") {
-            vargs.expectArgs("strip method", {0, 0}, {0, 0});
-            return Value(strip(str));
+            vargs.expectArgs("strip method", {0, 1}, {0, 0});
+            auto chars = vargs.args.empty() ? "" : vargs.args[0].get<std::string>();
+            return Value(strip(str, chars));
+          } else if (method->get_name() == "lstrip") {
+            vargs.expectArgs("lstrip method", {0, 1}, {0, 0});
+            auto chars = vargs.args.empty() ? "" : vargs.args[0].get<std::string>();
+            return Value(strip(str, chars, /* left= */ true, /* right= */ false));
+          } else if (method->get_name() == "rstrip") {
+            vargs.expectArgs("rstrip method", {0, 1}, {0, 0});
+            auto chars = vargs.args.empty() ? "" : vargs.args[0].get<std::string>();
+            return Value(strip(str, chars, /* left= */ false, /* right= */ true));
+          } else if (method->get_name() == "split") {
+            vargs.expectArgs("split method", {1, 1}, {0, 0});
+            auto sep = vargs.args[0].get<std::string>();
+            auto parts = split(str, sep);
+            Value result = Value::array();
+            for (const auto& part : parts) {
+              result.push_back(Value(part));
+            }
+            return result;
           } else if (method->get_name() == "capitalize") {
             vargs.expectArgs("capitalize method", {0, 0}, {0, 0});
             return Value(capitalize(str));

common/ngram-cache.cpp

@@ -7,6 +7,7 @@
 #include <cstdio>
 #include <fstream>
 #include <thread>
+#include <algorithm>
 
 void common_ngram_cache_update(common_ngram_cache & ngram_cache, int ngram_min, int ngram_max,
                               std::vector<llama_token> & inp, int nnew, bool print_progress) {

common/sampling.cpp

@@ -4,6 +4,7 @@
 
 #include <cmath>
 #include <unordered_map>
+#include <algorithm>
 
 // the ring buffer works similarly to std::deque, but with a fixed capacity
 // TODO: deduplicate with llama-impl.h
@@ -159,16 +160,53 @@ struct common_sampler * common_sampler_init(const struct llama_model * model, co
         GGML_ABORT("llguidance (cmake -DLLAMA_LLGUIDANCE=ON) is not enabled");
 #endif // LLAMA_USE_LLGUIDANCE
     } else {
-        std::vector<const char *> trigger_words;
-        trigger_words.reserve(params.grammar_trigger_words.size());
-        for (const auto & str : params.grammar_trigger_words) {
-            trigger_words.push_back(str.word.c_str());
+        std::vector<std::string> patterns_at_start;
+        std::vector<std::string> patterns_anywhere;
+        std::vector<llama_token> trigger_tokens;
+        for (const auto & trigger : params.grammar_triggers) {
+            switch (trigger.type) {
+                case COMMON_GRAMMAR_TRIGGER_TYPE_WORD:
+                {
+                    const auto & word = trigger.value;
+                    patterns_anywhere.push_back(regex_escape(word));
+                    break;
+                }
+                case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN:
+                case COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START:
+                {
+                    const auto & pattern = trigger.value;
+                    (trigger.type == COMMON_GRAMMAR_TRIGGER_TYPE_PATTERN_START ? patterns_at_start : patterns_anywhere).push_back(pattern);
+                    break;
+                }
+                case COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN:
+                {
+                    const auto token = trigger.token;
+                    trigger_tokens.push_back(token);
+                    break;
+                }
+                default:
+                    GGML_ASSERT(false && "unknown trigger type");
+            }
+        }
+
+        std::vector<std::string> trigger_patterns;
+        if (!patterns_at_start.empty()) {
+            trigger_patterns.push_back("^(" + string_join(patterns_at_start, "|") + ")[\\s\\S]*");
+        }
+        if (!patterns_anywhere.empty()) {
+            trigger_patterns.push_back("^[\\s\\S]*?(" + string_join(patterns_anywhere, "|") + ")[\\s\\S]*");
+        }
+
+        std::vector<const char *> trigger_patterns_c;
+        trigger_patterns_c.reserve(trigger_patterns.size());
+        for (const auto & regex : trigger_patterns) {
+            trigger_patterns_c.push_back(regex.c_str());
         }
 
         grmr = params.grammar_lazy
-             ? llama_sampler_init_grammar_lazy(vocab, params.grammar.c_str(), "root",
-                                               trigger_words.data(), trigger_words.size(),
-                                               params.grammar_trigger_tokens.data(), params.grammar_trigger_tokens.size())
+             ? llama_sampler_init_grammar_lazy_patterns(vocab, params.grammar.c_str(), "root",
+                                                        trigger_patterns_c.data(), trigger_patterns_c.size(),
+                                                        trigger_tokens.data(), trigger_tokens.size())
              :      llama_sampler_init_grammar(vocab, params.grammar.c_str(), "root");
     }
 

common/speculative.cpp

@@ -5,6 +5,7 @@
 #include "sampling.h"
 
 #include <cstring>
+#include <algorithm>
 
 #define SPEC_VOCAB_MAX_SIZE_DIFFERENCE  128
 #define SPEC_VOCAB_CHECK_START_TOKEN_ID 5
@@ -172,7 +173,7 @@ llama_tokens common_speculative_gen_draft(
     result.reserve(params.n_draft);
 
     if (reuse_n == 0) {
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         prompt.clear();
     } else {
@@ -191,14 +192,14 @@ llama_tokens common_speculative_gen_draft(
         }
 
         if (reuse_i > 0) {
-            llama_kv_cache_seq_rm (ctx, 0, 0, reuse_i);
-            llama_kv_cache_seq_add(ctx, 0, reuse_i, -1, -reuse_i);
+            llama_kv_self_seq_rm (ctx, 0, 0, reuse_i);
+            llama_kv_self_seq_add(ctx, 0, reuse_i, -1, -reuse_i);
 
             prompt.erase(prompt.begin(), prompt.begin() + reuse_i);
         }
 
         if (reuse_n < (int) prompt.size()) {
-            llama_kv_cache_seq_rm (ctx, 0, reuse_n, -1);
+            llama_kv_self_seq_rm (ctx, 0, reuse_n, -1);
 
             prompt.erase(prompt.begin() + reuse_n, prompt.end());
         }

convert_hf_to_gguf.py

@@ -861,6 +861,9 @@ class Model:
                 for token_id, token_data in added_tokens_decoder.items():
                     token_id = int(token_id)
                     token: str = token_data["content"]
+                    if token_id >= vocab_size:
+                        logger.warning(f'ignore token {token_id}: id is out of range, max={vocab_size - 1}')
+                        continue
                     if toktypes[token_id] != SentencePieceTokenTypes.UNUSED:
                         if tokens[token_id] != token.encode("utf-8"):
                             logger.warning(f'replacing token {token_id}: {tokens[token_id].decode("utf-8")!r} -> {token!r}')
@@ -3322,6 +3325,83 @@ class Gemma2Model(Model):
         return [(self.map_tensor_name(name), data_torch)]
 
 
+@Model.register("Gemma3ForCausalLM", "Gemma3ForConditionalGeneration")
+class Gemma3Model(Model):
+    model_arch = gguf.MODEL_ARCH.GEMMA3
+    has_vision: bool = False
+
+    # we need to merge the text_config into the root level of hparams
+    def __init__(self, *args, **kwargs):
+        hparams = Model.load_hparams(kwargs["dir_model"])
+        if "text_config" in hparams:
+            hparams = {**hparams, **hparams["text_config"]}
+            kwargs["hparams"] = hparams
+        super().__init__(*args, **kwargs)
+        if "vision_config" in hparams:
+            logger.info("Has vision encoder, but it will be ignored")
+            self.has_vision = True
+
+    def write(self):
+        super().write()
+        if self.has_vision:
+            logger.info("NOTE: this script only convert the language model to GGUF")
+            logger.info("      for the vision model, please use gemma3_convert_encoder_to_gguf.py")
+
+    def set_vocab(self):
+        self._set_vocab_sentencepiece()
+
+        self.gguf_writer.add_add_space_prefix(False)
+
+    def set_gguf_parameters(self):
+        hparams = self.hparams
+        block_count = hparams["num_hidden_layers"]
+
+        # some default values are not specified in the hparams
+        self.gguf_writer.add_context_length(hparams.get("max_position_embeddings", 131072))
+        self.gguf_writer.add_embedding_length(hparams["hidden_size"])
+        self.gguf_writer.add_block_count(block_count)
+        self.gguf_writer.add_feed_forward_length(hparams["intermediate_size"])
+        self.gguf_writer.add_head_count(hparams.get("num_attention_heads", 8))
+        self.gguf_writer.add_layer_norm_rms_eps(self.hparams.get("rms_norm_eps", 1e-6))
+        self.gguf_writer.add_key_length(hparams.get("head_dim", 256))
+        self.gguf_writer.add_value_length(hparams.get("head_dim", 256))
+        self.gguf_writer.add_file_type(self.ftype)
+        self.gguf_writer.add_rope_freq_base(hparams.get("rope_theta", 1_000_000.0)) # for global layers
+        # both attn_logit_softcapping and final_logit_softcapping are removed in Gemma3
+        assert hparams.get("attn_logit_softcapping") is None
+        assert hparams.get("final_logit_softcapping") is None
+        self.gguf_writer.add_sliding_window(hparams["sliding_window"])
+        self.gguf_writer.add_head_count_kv(hparams.get("num_key_value_heads", 4))
+        if hparams.get("rope_scaling") is not None:
+            assert hparams["rope_scaling"]["rope_type"] == "linear"
+            # important: this rope_scaling is only applied for global layers, and not used by 1B model
+            self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LINEAR)
+            self.gguf_writer.add_rope_scaling_factor(hparams["rope_scaling"]["factor"])
+
+    def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
+        del bid  # unused
+
+        if name.startswith("language_model."):
+            name = name.replace("language_model.", "")
+        elif name.startswith("multi_modal_projector.") or name.startswith("vision_tower.") \
+                or name.startswith("multimodal_projector.") or name.startswith("vision_model."): # this is for old HF model, should be removed later
+            # ignore vision tensors
+            return []
+
+        # remove OOV (out-of-vocabulary) rows in token_embd
+        if "embed_tokens.weight" in name:
+            vocab = self._create_vocab_sentencepiece()
+            tokens = vocab[0]
+            data_torch = data_torch[:len(tokens)]
+
+        # ref code in Gemma3RMSNorm
+        # output = output * (1.0 + self.weight.float())
+        if name.endswith("norm.weight"):
+            data_torch = data_torch + 1
+
+        return [(self.map_tensor_name(name), data_torch)]
+
+
 @Model.register("Starcoder2ForCausalLM")
 class StarCoder2Model(Model):
     model_arch = gguf.MODEL_ARCH.STARCODER2

docs/build.md

@@ -197,29 +197,53 @@ The following compilation options are also available to tweak performance:
 
 ## MUSA
 
-This provides GPU acceleration using the MUSA cores of your Moore Threads MTT GPU. Make sure to have the MUSA SDK installed. You can download it from here: [MUSA SDK](https://developer.mthreads.com/sdk/download/musa).
+This provides GPU acceleration using a Moore Threads GPU. Make sure to have the [MUSA SDK](https://developer.mthreads.com/musa/musa-sdk) installed.
 
-- Using `CMake`:
+#### Download directly from Moore Threads
 
-  ```bash
-  cmake -B build -DGGML_MUSA=ON
-  cmake --build build --config Release
+You may find the official downloads here: [Moore Threads developer site](https://developer.mthreads.com/sdk/download/musa).
+
+### Compilation
+
+```bash
+cmake -B build -DGGML_MUSA=ON
+cmake --build build --config Release
+```
+
+#### Override Compute Capability Specifications
+
+By default, all supported compute capabilities are enabled. To customize this behavior, you can specify the `MUSA_ARCHITECTURES` option in the CMake command:
+
+```bash
+cmake -B build -DGGML_MUSA=ON -DMUSA_ARCHITECTURES="21"
+```
+
+This configuration enables only compute capability `2.1` (MTT S80) during compilation, which can help reduce compilation time.
+
+#### Compilation options
+
+Most of the compilation options available for CUDA should also be available for MUSA, though they haven't been thoroughly tested yet.
+
+- For static builds, add `-DBUILD_SHARED_LIBS=OFF` and `-DCMAKE_POSITION_INDEPENDENT_CODE=ON`:
   ```
-
-  For static build:
-
-  ```bash
   cmake -B build -DGGML_MUSA=ON \
     -DBUILD_SHARED_LIBS=OFF -DCMAKE_POSITION_INDEPENDENT_CODE=ON
   cmake --build build --config Release
   ```
 
-The environment variable [`MUSA_VISIBLE_DEVICES`](https://docs.mthreads.com/musa-sdk/musa-sdk-doc-online/programming_guide/Z%E9%99%84%E5%BD%95/) can be used to specify which GPU(s) will be used.
+### Runtime MUSA environmental variables
+
+You may set the [musa environmental variables](https://docs.mthreads.com/musa-sdk/musa-sdk-doc-online/programming_guide/Z%E9%99%84%E5%BD%95/) at runtime.
+
+```bash
+# Use `MUSA_VISIBLE_DEVICES` to hide the first compute device.
+MUSA_VISIBLE_DEVICES="-0" ./build/bin/llama-server --model /srv/models/llama.gguf
+```
+
+### Unified Memory
 
 The environment variable `GGML_CUDA_ENABLE_UNIFIED_MEMORY=1` can be used to enable unified memory in Linux. This allows swapping to system RAM instead of crashing when the GPU VRAM is exhausted.
 
-Most of the compilation options available for CUDA should also be available for MUSA, though they haven't been thoroughly tested yet.
-
 ## HIP
 
 This provides GPU acceleration on HIP-supported AMD GPUs.
@@ -235,6 +259,12 @@ You can download it from your Linux distro's package manager or from here: [ROCm
   On Linux it is also possible to use unified memory architecture (UMA) to share main memory between the CPU and integrated GPU by setting `-DGGML_HIP_UMA=ON`.
   However, this hurts performance for non-integrated GPUs (but enables working with integrated GPUs).
 
+  To enhance flash attention performance on RDNA3+ or CDNA architectures, you can utilize the rocWMMA library by enabling the `-DGGML_HIP_ROCWMMA_FATTN=ON` option. This requires rocWMMA headers to be installed on the build system.
+
+  The rocWMMA library is included by default when installing the ROCm SDK using the `rocm` meta package provided by AMD. Alternatively, if you are not using the meta package, you can install the library using the `rocwmma-dev` or `rocwmma-devel` package, depending on your system's package manager.
+
+  As an alternative, you can manually install the library by cloning it from the official [GitHub repository](https://github.com/ROCm/rocWMMA), checkout the corresponding version tag (e.g. `rocm-6.2.4`) and set `-DCMAKE_CXX_FLAGS="-I<path/to/rocwmma>/library/include/"` in CMake. This also works under Windows despite not officially supported by AMD.
+
   Note that if you get the following error:
   ```
   clang: error: cannot find ROCm device library; provide its path via '--rocm-path' or '--rocm-device-lib-path', or pass '-nogpulib' to build without ROCm device library

docs/function-calling.md

@@ -287,30 +287,32 @@ Here are some models known to work (w/ chat template override when needed):
 
 llama-server --jinja -fa -hf bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M
 llama-server --jinja -fa -hf bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q6_K_L
-llama-server --jinja -fa -hf bartowski/functionary-small-v3.2-GGUF:Q4_K_M
 llama-server --jinja -fa -hf bartowski/Llama-3.3-70B-Instruct-GGUF:Q4_K_M
 
-# Native support for DeepSeek R1 works best w/ our own template (official template buggy)
+# Native support for DeepSeek R1 works best w/ our template override (official template is buggy, although we do work around it)
 
 llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q6_K_L \
---chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
+    --chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
 
 llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-32B-GGUF:Q4_K_M \
---chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
+    --chat-template-file models/templates/llama-cpp-deepseek-r1.jinja
 
 # Native support requires the right template for these GGUFs:
 
+llama-server --jinja -fa -hf bartowski/functionary-small-v3.2-GGUF:Q4_K_M
+    --chat-template-file models/templates/meetkai-functionary-medium-v3.2.jinja
+
 llama-server --jinja -fa -hf bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M \
---chat-template-file <( python scripts/get_chat_template.py NousResearch/Hermes-2-Pro-Llama-3-8B tool_use )
+    --chat-template-file models/templates/NousResearch-Hermes-2-Pro-Llama-3-8B-tool_use.jinja
 
 llama-server --jinja -fa -hf bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M \
---chat-template-file <( python scripts/get_chat_template.py NousResearch/Hermes-3-Llama-3.1-8B tool_use )
+    --chat-template-file models/templates/NousResearch-Hermes-3-Llama-3.1-8B-tool_use.jinja
 
 llama-server --jinja -fa -hf bartowski/firefunction-v2-GGUF -hff firefunction-v2-IQ1_M.gguf \
---chat-template-file <( python scripts/get_chat_template.py fireworks-ai/llama-3-firefunction-v2 tool_use )
+    --chat-template-file models/templates/fireworks-ai-llama-3-firefunction-v2.jinja
 
 llama-server --jinja -fa -hf bartowski/c4ai-command-r7b-12-2024-GGUF:Q6_K_L \
---chat-template-file <( python scripts/get_chat_template.py CohereForAI/c4ai-command-r7b-12-2024 tool_use )
+    --chat-template-file models/templates/CohereForAI-c4ai-command-r7b-12-2024-tool_use.jinja
 
 # Generic format support
 llama-server --jinja -fa -hf bartowski/phi-4-GGUF:Q4_0
@@ -318,6 +320,8 @@ llama-server --jinja -fa -hf bartowski/gemma-2-2b-it-GGUF:Q8_0
 llama-server --jinja -fa -hf bartowski/c4ai-command-r-v01-GGUF:Q2_K
 ```
 
+To get the official template from original HuggingFace repos, you can use [scripts/get_chat_template.py](../scripts/get_chat_template.py) (see examples invocations in [models/templates/README.md](../models/templates/README.md))
+
 > [!TIP]
 > If there is no official `tool_use` Jinja template, you may want to set `--chat-template chatml` to use a default that works with many models (YMMV!), or write your own (e.g. we provide a custom [llama-cpp-deepseek-r1.jinja](../models/templates/llama-cpp-deepseek-r1.jinja) for DeepSeek R1 distills)
 

examples/batched-bench/batched-bench.cpp

@@ -132,7 +132,7 @@ int main(int argc, char ** argv) {
 
                 const auto t_pp_start = ggml_time_us();
 
-                llama_kv_cache_clear(ctx);
+                llama_kv_self_clear(ctx);
 
                 if (!decode_helper(ctx, batch, ctx_params.n_batch)) {
                     LOG_ERR("%s: llama_decode() failed\n", __func__);
@@ -141,7 +141,7 @@ int main(int argc, char ** argv) {
 
                 if (is_pp_shared) {
                     for (int32_t i = 1; i < pl; ++i) {
-                        llama_kv_cache_seq_cp(ctx, 0, i, -1, -1);
+                        llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
                     }
                 }
 

examples/batched.swift/Sources/main.swift

@@ -116,7 +116,7 @@ if llama_decode(context, batch) != 0 {
 }
 
 for i in 1 ..< n_parallel {
-    llama_kv_cache_seq_cp(context, 0, Int32(i), 0, batch.n_tokens)
+    llama_kv_self_seq_cp(context, 0, Int32(i), 0, batch.n_tokens)
 }
 
 if n_parallel > 1 {

examples/cvector-generator/cvector-generator.cpp

@@ -342,7 +342,7 @@ static bool cb_eval(struct ggml_tensor * t, bool ask, void * user_data) {
 }
 
 static bool get_hidden_layers(llama_context * ctx, std::vector<llama_token> & tokens) {
-    llama_kv_cache_clear(ctx);
+    llama_kv_self_clear(ctx);
     if (llama_decode(ctx, llama_batch_get_one(tokens.data(), tokens.size()))) {
         fprintf(stderr, "%s : failed to eval\n", __func__);
         return false;
@@ -394,6 +394,8 @@ static int prepare_entries(common_params & params, train_context & ctx_train) {
 int main(int argc, char ** argv) {
     common_params params;
 
+    params.out_file = "control_vector.gguf";
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_CVECTOR_GENERATOR, print_usage)) {
         return 1;
     }
@@ -498,7 +500,7 @@ int main(int argc, char ** argv) {
     }
 
     // write output vectors to gguf
-    export_gguf(ctx_train.v_final, params.cvector_outfile, model_hint);
+    export_gguf(ctx_train.v_final, params.out_file, model_hint);
 
     llama_backend_free();
 

examples/embedding/embedding.cpp

@@ -4,6 +4,7 @@
 #include "llama.h"
 
 #include <ctime>
+#include <algorithm>
 
 #if defined(_MSC_VER)
 #pragma warning(disable: 4244 4267) // possible loss of data
@@ -37,7 +38,7 @@ static void batch_decode(llama_context * ctx, llama_batch & batch, float * outpu
     const struct llama_model * model = llama_get_model(ctx);
 
     // clear previous kv_cache values (irrelevant for embeddings)
-    llama_kv_cache_clear(ctx);
+    llama_kv_self_clear(ctx);
 
     // run model
     LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);

examples/export-lora/export-lora.cpp

@@ -413,20 +413,22 @@ static void print_usage(int, char ** argv) {
 int main(int argc, char ** argv) {
     common_params params;
 
+    params.out_file = "ggml-lora-merged-f16.gguf";
+
     if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_EXPORT_LORA, print_usage)) {
         return 1;
     }
 
     g_verbose = (params.verbosity > 1);
     try {
-        lora_merge_ctx ctx(params.model, params.lora_adapters, params.lora_outfile, params.cpuparams.n_threads);
+        lora_merge_ctx ctx(params.model, params.lora_adapters, params.out_file, params.cpuparams.n_threads);
         ctx.run_merge();
     } catch (const std::exception & err) {
         fprintf(stderr, "%s\n", err.what());
         exit(EXIT_FAILURE);
     }
 
-    printf("done, output file is %s\n", params.lora_outfile.c_str());
+    printf("done, output file is %s\n", params.out_file.c_str());
 
     return 0;
 }

examples/gritlm/gritlm.cpp

@@ -45,7 +45,7 @@ static std::vector<std::vector<float>> encode(llama_context * ctx, const std::ve
         }
 
         // clear previous kv_cache values (irrelevant for embeddings)
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
         llama_set_embeddings(ctx, true);
         llama_set_causal_attn(ctx, false);
 
@@ -102,7 +102,7 @@ static std::string generate(llama_context * ctx, llama_sampler * smpl, const std
 
     llama_token eos_token = llama_vocab_eos(vocab);
 
-    llama_kv_cache_clear(ctx);
+    llama_kv_self_clear(ctx);
     llama_set_embeddings(ctx, false);
     llama_set_causal_attn(ctx, true);
 

examples/imatrix/imatrix.cpp

@@ -206,9 +206,6 @@ bool IMatrixCollector::collect_imatrix(struct ggml_tensor * t, bool ask, void *
 
 void IMatrixCollector::save_imatrix(int ncall) const {
     auto fname = m_params.out_file;
-    if (fname.empty()) {
-        fname = "imatrix.dat";
-    }
 
     if (ncall > 0) {
         fname += ".at_";
@@ -498,7 +495,7 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) {
         const auto t_start = std::chrono::high_resolution_clock::now();
 
         // clear the KV cache
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         llama_batch batch = llama_batch_init(n_batch, 0, 1);
 
@@ -583,6 +580,8 @@ static bool compute_imatrix(llama_context * ctx, const common_params & params) {
 int main(int argc, char ** argv) {
     common_params params;
 
+    params.out_file = "imatrix.dat" ;
+
     params.n_ctx = 512;
     params.logits_all = true;
     params.escape = false;

examples/infill/infill.cpp

@@ -332,8 +332,8 @@ int main(int argc, char ** argv) {
                 LOG_DBG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
                     n_past, n_left, n_ctx, params.n_keep, n_discard);
 
-                llama_kv_cache_seq_rm (ctx, 0, params.n_keep + 1            , params.n_keep + n_discard + 1);
-                llama_kv_cache_seq_add(ctx, 0, params.n_keep + 1 + n_discard, n_past, -n_discard);
+                llama_kv_self_seq_rm (ctx, 0, params.n_keep + 1            , params.n_keep + n_discard + 1);
+                llama_kv_self_seq_add(ctx, 0, params.n_keep + 1 + n_discard, n_past, -n_discard);
 
                 n_past -= n_discard;
 

examples/json_schema_to_grammar.py

@@ -195,7 +195,7 @@ class BuiltinRule:
         self.deps = deps or []
 
 # Constraining spaces to prevent model "running away".
-SPACE_RULE = '| " " | "\\n" [ \\t]{0,20}'
+SPACE_RULE = '| " " | "\\n"{1,2} [ \\t]{0,20}'
 
 PRIMITIVE_RULES = {
     'boolean'      : BuiltinRule('("true" | "false") space', []),

examples/llama-bench/llama-bench.cpp

@@ -1578,7 +1578,7 @@ int main(int argc, char ** argv) {
 
         test t(inst, lmodel, ctx);
 
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         // cool off before the test
         if (params.delay) {
@@ -1618,7 +1618,7 @@ int main(int argc, char ** argv) {
         }
 
         for (int i = 0; i < params.reps; i++) {
-            llama_kv_cache_clear(ctx);
+            llama_kv_self_clear(ctx);
 
             uint64_t t_start = get_time_ns();
 

examples/llama.android/llama/src/main/cpp/llama-android.cpp

@@ -194,7 +194,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
         }
 
         batch->logits[batch->n_tokens - 1] = true;
-        llama_kv_cache_clear(context);
+        llama_kv_self_clear(context);
 
         const auto t_pp_start = ggml_time_us();
         if (llama_decode(context, *batch) != 0) {
@@ -206,7 +206,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
 
         LOGi("Benchmark text generation (tg)");
 
-        llama_kv_cache_clear(context);
+        llama_kv_self_clear(context);
         const auto t_tg_start = ggml_time_us();
         for (i = 0; i < tg; i++) {
 
@@ -223,7 +223,7 @@ Java_android_llama_cpp_LLamaAndroid_bench_1model(
 
         const auto t_tg_end = ggml_time_us();
 
-        llama_kv_cache_clear(context);
+        llama_kv_self_clear(context);
 
         const auto t_pp = double(t_pp_end - t_pp_start) / 1000000.0;
         const auto t_tg = double(t_tg_end - t_tg_start) / 1000000.0;
@@ -361,7 +361,7 @@ Java_android_llama_cpp_LLamaAndroid_completion_1init(
     const auto tokens_list = common_tokenize(context, text, true, parse_special);
 
     auto n_ctx = llama_n_ctx(context);
-    auto n_kv_req = tokens_list.size() + (n_len - tokens_list.size());
+    auto n_kv_req = tokens_list.size() + n_len;
 
     LOGi("n_len = %d, n_ctx = %d, n_kv_req = %d", n_len, n_ctx, n_kv_req);
 
@@ -448,5 +448,5 @@ Java_android_llama_cpp_LLamaAndroid_completion_1loop(
 extern "C"
 JNIEXPORT void JNICALL
 Java_android_llama_cpp_LLamaAndroid_kv_1cache_1clear(JNIEnv *, jobject, jlong context) {
-    llama_kv_cache_clear(reinterpret_cast<llama_context *>(context));
+    llama_kv_self_clear(reinterpret_cast<llama_context *>(context));
 }

examples/llama.swiftui/README.md

@@ -5,6 +5,21 @@ point for more advanced projects.
 
 For usage instructions and performance stats, check the following discussion: https://github.com/ggml-org/llama.cpp/discussions/4508
 
+
+### Building
+First llama.cpp need to be built and a XCFramework needs to be created. This can be done by running
+the following script from the llama.cpp project root:
+```console
+$ ./build-xcframework.sh
+```
+Open `llama.swiftui.xcodeproj` project in Xcode and you should be able to build and run the app on
+a simulator or a real device.
+
+To use the framework with a different project, the XCFramework can be added to the project by
+adding `build-apple/llama.xcframework` by dragging and dropping it into the project navigator, or
+by manually selecting the framework in the "Frameworks, Libraries, and Embedded Content" section
+of the project settings.
+
 ![image](https://github.com/ggml-org/llama.cpp/assets/1991296/2b40284f-8421-47a2-b634-74eece09a299)
 
 Video demonstration:

examples/llama.swiftui/llama.cpp.swift/LibLlama.swift

@@ -210,7 +210,7 @@ actor LlamaContext {
             }
             batch.logits[Int(batch.n_tokens) - 1] = 1 // true
 
-            llama_kv_cache_clear(context)
+            llama_kv_self_clear(context)
 
             let t_pp_start = DispatchTime.now().uptimeNanoseconds / 1000;
 
@@ -223,7 +223,7 @@ actor LlamaContext {
 
             // bench text generation
 
-            llama_kv_cache_clear(context)
+            llama_kv_self_clear(context)
 
             let t_tg_start = DispatchTime.now().uptimeNanoseconds / 1000;
 
@@ -242,7 +242,7 @@ actor LlamaContext {
 
             let t_tg_end = DispatchTime.now().uptimeNanoseconds / 1000;
 
-            llama_kv_cache_clear(context)
+            llama_kv_self_clear(context)
 
             let t_pp = Double(t_pp_end - t_pp_start) / 1000000.0
             let t_tg = Double(t_tg_end - t_tg_start) / 1000000.0
@@ -292,7 +292,7 @@ actor LlamaContext {
     func clear() {
         tokens_list.removeAll()
         temporary_invalid_cchars.removeAll()
-        llama_kv_cache_clear(context)
+        llama_kv_self_clear(context)
     }
 
     private func tokenize(text: String, add_bos: Bool) -> [llama_token] {

examples/llama.swiftui/llama.swiftui.xcodeproj/project.pbxproj

@@ -7,7 +7,6 @@
 	objects = {
 
 /* Begin PBXBuildFile section */
-		1809696D2D05A39F00400EE8 /* llama in Frameworks */ = {isa = PBXBuildFile; productRef = 1809696C2D05A39F00400EE8 /* llama */; };
 		549479CB2AC9E16000E0F78B /* Metal.framework in Frameworks */ = {isa = PBXBuildFile; fileRef = 549479CA2AC9E16000E0F78B /* Metal.framework */; };
 		79E1D9CD2B4CD16E005F8E46 /* InputButton.swift in Sources */ = {isa = PBXBuildFile; fileRef = 79E1D9CC2B4CD16E005F8E46 /* InputButton.swift */; };
 		7FA3D2B32B2EA2F600543F92 /* DownloadButton.swift in Sources */ = {isa = PBXBuildFile; fileRef = 7FA3D2B22B2EA2F600543F92 /* DownloadButton.swift */; };
@@ -18,9 +17,25 @@
 		8A3F84242AC4C891005E2EE8 /* models in Resources */ = {isa = PBXBuildFile; fileRef = 8A3F84232AC4C891005E2EE8 /* models */; };
 		8A907F332AC7138A006146EA /* LibLlama.swift in Sources */ = {isa = PBXBuildFile; fileRef = 8A907F322AC7134E006146EA /* LibLlama.swift */; };
 		8A9F7C4D2AC332EE008AE1EA /* LlamaState.swift in Sources */ = {isa = PBXBuildFile; fileRef = 8A9F7C4C2AC332EE008AE1EA /* LlamaState.swift */; };
+		DD84C9FD2D747FED007778EC /* llama.xcframework in Frameworks */ = {isa = PBXBuildFile; fileRef = DD84C9FC2D747FED007778EC /* llama.xcframework */; };
+		DD84C9FE2D747FED007778EC /* llama.xcframework in Embed Frameworks */ = {isa = PBXBuildFile; fileRef = DD84C9FC2D747FED007778EC /* llama.xcframework */; settings = {ATTRIBUTES = (CodeSignOnCopy, RemoveHeadersOnCopy, ); }; };
 		F1FE20E22B465ECA00B45541 /* LoadCustomButton.swift in Sources */ = {isa = PBXBuildFile; fileRef = F1FE20E12B465EC900B45541 /* LoadCustomButton.swift */; };
 /* End PBXBuildFile section */
 
+/* Begin PBXCopyFilesBuildPhase section */
+		DD84C9FF2D747FED007778EC /* Embed Frameworks */ = {
+			isa = PBXCopyFilesBuildPhase;
+			buildActionMask = 2147483647;
+			dstPath = "";
+			dstSubfolderSpec = 10;
+			files = (
+				DD84C9FE2D747FED007778EC /* llama.xcframework in Embed Frameworks */,
+			);
+			name = "Embed Frameworks";
+			runOnlyForDeploymentPostprocessing = 0;
+		};
+/* End PBXCopyFilesBuildPhase section */
+
 /* Begin PBXFileReference section */
 		549479CA2AC9E16000E0F78B /* Metal.framework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.framework; name = Metal.framework; path = System/Library/Frameworks/Metal.framework; sourceTree = SDKROOT; };
 		79E1D9CC2B4CD16E005F8E46 /* InputButton.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = InputButton.swift; sourceTree = "<group>"; };
@@ -33,6 +48,7 @@
 		8A3F84232AC4C891005E2EE8 /* models */ = {isa = PBXFileReference; lastKnownFileType = folder; name = models; path = llama.swiftui/Resources/models; sourceTree = "<group>"; };
 		8A907F322AC7134E006146EA /* LibLlama.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LibLlama.swift; sourceTree = "<group>"; };
 		8A9F7C4C2AC332EE008AE1EA /* LlamaState.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LlamaState.swift; sourceTree = "<group>"; };
+		DD84C9FC2D747FED007778EC /* llama.xcframework */ = {isa = PBXFileReference; lastKnownFileType = wrapper.xcframework; name = llama.xcframework; path = "../../build-apple/llama.xcframework"; sourceTree = "<group>"; };
 		DF2D2FE72B4A59BE00FCB72D /* llama.cpp */ = {isa = PBXFileReference; lastKnownFileType = wrapper; name = llama.cpp; path = ../..; sourceTree = "<group>"; };
 		F1FE20E12B465EC900B45541 /* LoadCustomButton.swift */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.swift; path = LoadCustomButton.swift; sourceTree = "<group>"; };
 /* End PBXFileReference section */
@@ -42,9 +58,9 @@
 			isa = PBXFrameworksBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
-				1809696D2D05A39F00400EE8 /* llama in Frameworks */,
 				549479CB2AC9E16000E0F78B /* Metal.framework in Frameworks */,
 				8A39BE0A2AC7601100BFEB40 /* Accelerate.framework in Frameworks */,
+				DD84C9FD2D747FED007778EC /* llama.xcframework in Frameworks */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};
@@ -86,6 +102,7 @@
 		8A39BE082AC7601000BFEB40 /* Frameworks */ = {
 			isa = PBXGroup;
 			children = (
+				DD84C9FC2D747FED007778EC /* llama.xcframework */,
 				549479CA2AC9E16000E0F78B /* Metal.framework */,
 				8A39BE092AC7601000BFEB40 /* Accelerate.framework */,
 			);
@@ -144,6 +161,7 @@
 				8A1C836F2AC328BD0096AF73 /* Sources */,
 				8A1C83702AC328BD0096AF73 /* Frameworks */,
 				8A1C83712AC328BD0096AF73 /* Resources */,
+				DD84C9FF2D747FED007778EC /* Embed Frameworks */,
 			);
 			buildRules = (
 			);
@@ -151,7 +169,6 @@
 			);
 			name = llama.swiftui;
 			packageProductDependencies = (
-				1809696C2D05A39F00400EE8 /* llama */,
 			);
 			productName = llama.swiftui;
 			productReference = 8A1C83732AC328BD0096AF73 /* llama.swiftui.app */;
@@ -427,13 +444,6 @@
 			defaultConfigurationName = Release;
 		};
 /* End XCConfigurationList section */
-
-/* Begin XCSwiftPackageProductDependency section */
-		1809696C2D05A39F00400EE8 /* llama */ = {
-			isa = XCSwiftPackageProductDependency;
-			productName = llama;
-		};
-/* End XCSwiftPackageProductDependency section */
 	};
 	rootObject = 8A1C836B2AC328BD0096AF73 /* Project object */;
 }

examples/llava/CMakeLists.txt

@@ -51,6 +51,13 @@ install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llava ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
+set(TARGET llama-gemma3-cli)
+add_executable(${TARGET} gemma3-cli.cpp)
+set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama-gemma3-cli)
+install(TARGETS ${TARGET} RUNTIME)
+target_link_libraries(${TARGET} PRIVATE common llava ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PRIVATE cxx_std_17)
+
 set(TARGET llama-llava-clip-quantize-cli)
 add_executable(${TARGET} clip-quantize-cli.cpp)
 set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama-llava-clip-quantize-cli)

examples/llava/README-gemma3.md

@@ -0,0 +1,30 @@
+# Gemma 3 vision
+
+> [!IMPORTANT]
+>
+> This is very experimental, only used for demo purpose.
+
+## How to get mmproj.gguf?
+
+```bash
+cd gemma-3-4b-it
+python ../llama.cpp/examples/llava/gemma3_convert_encoder_to_gguf.py .
+
+# output file is mmproj.gguf
+```
+
+## How to run it?
+
+What you need:
+- The text model GGUF, can be converted using `convert_hf_to_gguf.py`
+- The mmproj file from step above
+- An image file
+
+```bash
+# build
+cmake -B build
+cmake --build build --target llama-gemma3-cli
+
+# run it
+./build/bin/llama-gemma3-cli -m {text_model}.gguf --mmproj mmproj.gguf --image your_image.jpg
+```

examples/llava/README-minicpmo2.6.md

@@ -5,13 +5,25 @@ Currently, this readme only supports minicpm-omni's image capabilities, and we w
 
 Download [MiniCPM-o-2_6](https://huggingface.co/openbmb/MiniCPM-o-2_6) PyTorch model from huggingface to "MiniCPM-o-2_6" folder.
 
+
+### Build llama.cpp
+Readme modification time: 20250206
+
+If there are differences in usage, please refer to the official build [documentation](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)
+
 Clone llama.cpp:
 ```bash
-git clone git@github.com:OpenBMB/llama.cpp.git
+git clone https://github.com/ggerganov/llama.cpp
 cd llama.cpp
-git checkout minicpm-omni
 ```
 
+Build llama.cpp using `CMake`:
+```bash
+cmake -B build
+cmake --build build --config Release
+```
+
+
 ### Usage of MiniCPM-o 2.6
 
 Convert PyTorch model to gguf files (You can also download the converted [gguf](https://huggingface.co/openbmb/MiniCPM-o-2_6-gguf) by us)
@@ -22,25 +34,15 @@ python ./examples/llava/minicpmv-convert-image-encoder-to-gguf.py -m ../MiniCPM-
 python ./convert_hf_to_gguf.py ../MiniCPM-o-2_6/model
 
 # quantize int4 version
-./llama-quantize ../MiniCPM-o-2_6/model/ggml-model-f16.gguf ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M
+./build/bin/llama-quantize ../MiniCPM-o-2_6/model/ggml-model-f16.gguf ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M
 ```
 
-Build llama.cpp using `CMake`:
-https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md
-
-```bash
-cmake -B build
-cmake --build build --config Release
-```
 
 Inference on Linux or Mac
-```
+```bash
 # run f16 version
-./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-f16.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
+./build/bin/llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-f16.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
 
 # run quantized int4 version
-./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
-
-# or run in interactive mode
-./llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i
+./build/bin/llama-minicpmv-cli -m ../MiniCPM-o-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-o-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
 ```

examples/llava/README-minicpmv2.5.md

@@ -4,13 +4,26 @@
 
 Download [MiniCPM-Llama3-V-2_5](https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5) PyTorch model from huggingface to "MiniCPM-Llama3-V-2_5" folder.
 
+
+### Build llama.cpp
+Readme modification time: 20250206
+
+If there are differences in usage, please refer to the official build [documentation](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)
+
 Clone llama.cpp:
 ```bash
 git clone https://github.com/ggml-org/llama.cpp
 cd llama.cpp
 ```
 
-### Usage
+Build llama.cpp using `CMake`:
+```bash
+cmake -B build
+cmake --build build --config Release
+```
+
+
+### Usage of MiniCPM-Llama3-V 2.5
 
 Convert PyTorch model to gguf files (You can also download the converted [gguf](https://huggingface.co/openbmb/MiniCPM-Llama3-V-2_5-gguf) by us)
 
@@ -20,80 +33,15 @@ python ./examples/llava/minicpmv-convert-image-encoder-to-gguf.py -m ../MiniCPM-
 python ./convert_hf_to_gguf.py ../MiniCPM-Llama3-V-2_5/model
 
 # quantize int4 version
-./llama-quantize ../MiniCPM-Llama3-V-2_5/model/model-8B-F16.gguf ../MiniCPM-Llama3-V-2_5/model/ggml-model-Q4_K_M.gguf Q4_K_M
+./build/bin/llama-quantize ../MiniCPM-Llama3-V-2_5/model/model-8B-F16.gguf ../MiniCPM-Llama3-V-2_5/model/ggml-model-Q4_K_M.gguf Q4_K_M
 ```
 
-Build for Linux or Mac
-
-```bash
-make
-make llama-minicpmv-cli
-```
 
 Inference on Linux or Mac
-```
+```bash
 # run f16 version
-./llama-minicpmv-cli -m ../MiniCPM-Llama3-V-2_5/model/model-8B-F16.gguf --mmproj ../MiniCPM-Llama3-V-2_5/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
+./build/bin/llama-minicpmv-cli -m ../MiniCPM-Llama3-V-2_5/model/model-8B-F16.gguf --mmproj ../MiniCPM-Llama3-V-2_5/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
 
 # run quantized int4 version
-./llama-minicpmv-cli -m ../MiniCPM-Llama3-V-2_5/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-Llama3-V-2_5/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
-
-# or run in interactive mode
-./llama-minicpmv-cli -m ../MiniCPM-Llama3-V-2_5/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-Llama3-V-2_5/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i
-```
-
-### Android
-
-#### Build on Android device using Termux
-We found that build on Android device would bring better runtime performance, so we recommend to build on device.
-
-[Termux](https://github.com/termux/termux-app#installation) is a terminal app on Android device (no root required).
-
-Install tools in Termux:
-```
-apt update && apt upgrade -y
-apt install git make cmake
-```
-
-It's recommended to move your model inside the `~/` directory for best performance:
-```
-cd storage/downloads
-mv model.gguf ~/
-```
-
-#### Building the Project using Android NDK
-Obtain the [Android NDK](https://developer.android.com/ndk) and then build with CMake.
-
-Execute the following commands on your computer to avoid downloading the NDK to your mobile. Alternatively, you can also do this in Termux:
-
-```bash
-mkdir build-android
-cd build-android
-export NDK=/your_ndk_path
-cmake -DCMAKE_TOOLCHAIN_FILE=$NDK/build/cmake/android.toolchain.cmake -DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-23 -DCMAKE_C_FLAGS=-march=armv8.4a+dotprod ..
-make
-```
-
-Install [termux](https://github.com/termux/termux-app#installation) on your device and run `termux-setup-storage` to get access to your SD card (if Android 11+ then run the command twice).
-
-Finally, copy these built `llama` binaries and the model file to your device storage. Because the file permissions in the Android sdcard cannot be changed, you can copy the executable files to the `/data/data/com.termux/files/home/bin` path, and then execute the following commands in Termux to add executable permission:
-
-(Assumed that you have pushed the built executable files to the /sdcard/llama.cpp/bin path using `adb push`)
-```
-$cp -r /sdcard/llama.cpp/bin /data/data/com.termux/files/home/
-$cd /data/data/com.termux/files/home/bin
-$chmod +x ./*
-```
-
-Download models and push them to `/sdcard/llama.cpp/`, then move it to `/data/data/com.termux/files/home/model/`
-
-```
-$mv /sdcard/llama.cpp/ggml-model-Q4_K_M.gguf /data/data/com.termux/files/home/model/
-$mv /sdcard/llama.cpp/mmproj-model-f16.gguf /data/data/com.termux/files/home/model/
-```
-
-Now, you can start chatting:
-```
-$cd /data/data/com.termux/files/home/bin
-$./llama-minicpmv-cli -m ../model/ggml-model-Q4_K_M.gguf --mmproj ../model/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
+./build/bin/llama-minicpmv-cli -m ../MiniCPM-Llama3-V-2_5/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-Llama3-V-2_5/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
 ```

examples/llava/README-minicpmv2.6.md

@@ -4,13 +4,25 @@
 
 Download [MiniCPM-V-2_6](https://huggingface.co/openbmb/MiniCPM-V-2_6) PyTorch model from huggingface to "MiniCPM-V-2_6" folder.
 
+
+### Build llama.cpp
+Readme modification time: 20250206
+
+If there are differences in usage, please refer to the official build [documentation](https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)
+
 Clone llama.cpp:
 ```bash
-git clone git@github.com:OpenBMB/llama.cpp.git
+git clone https://github.com/ggerganov/llama.cpp
 cd llama.cpp
-git checkout minicpmv-main
 ```
 
+Build llama.cpp using `CMake`:
+```bash
+cmake -B build
+cmake --build build --config Release
+```
+
+
 ### Usage of MiniCPM-V 2.6
 
 Convert PyTorch model to gguf files (You can also download the converted [gguf](https://huggingface.co/openbmb/MiniCPM-V-2_6-gguf) by us)
@@ -21,87 +33,15 @@ python ./examples/llava/minicpmv-convert-image-encoder-to-gguf.py -m ../MiniCPM-
 python ./convert_hf_to_gguf.py ../MiniCPM-V-2_6/model
 
 # quantize int4 version
-./llama-quantize ../MiniCPM-V-2_6/model/ggml-model-f16.gguf ../MiniCPM-V-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M
+./build/bin/llama-quantize ../MiniCPM-V-2_6/model/ggml-model-f16.gguf ../MiniCPM-V-2_6/model/ggml-model-Q4_K_M.gguf Q4_K_M
 ```
 
-Build for Linux or Mac
-
-```bash
-make
-make llama-minicpmv-cli
-```
 
 Inference on Linux or Mac
-```
+```bash
 # run f16 version
-./llama-minicpmv-cli -m ../MiniCPM-V-2_6/model/ggml-model-f16.gguf --mmproj ../MiniCPM-V-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
+./build/bin/llama-minicpmv-cli -m ../MiniCPM-V-2_6/model/ggml-model-f16.gguf --mmproj ../MiniCPM-V-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -p "What is in the image?"
 
 # run quantized int4 version
-./llama-minicpmv-cli -m ../MiniCPM-V-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-V-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
-
-# or run in interactive mode
-./llama-minicpmv-cli -m ../MiniCPM-V-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-V-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg -i
-```
-
-### Video
-Install FFmpeg
-```
-brew install ffmpeg
-brew install pkg-config
-```
-
-### Android
-
-#### Build on Android device using Termux
-We found that build on Android device would bring better runtime performance, so we recommend to build on device.
-
-[Termux](https://github.com/termux/termux-app#installation) is a terminal app on Android device (no root required).
-
-Install tools in Termux:
-```
-apt update && apt upgrade -y
-apt install git make cmake
-```
-
-It's recommended to move your model inside the `~/` directory for best performance:
-```
-cd storage/downloads
-mv model.gguf ~/
-```
-
-#### Building the Project using Android NDK
-Obtain the [Android NDK](https://developer.android.com/ndk) and then build with CMake.
-
-Execute the following commands on your computer to avoid downloading the NDK to your mobile. Alternatively, you can also do this in Termux:
-
-```bash
-mkdir build-android
-cd build-android
-export NDK=/your_ndk_path
-cmake -DCMAKE_TOOLCHAIN_FILE=$NDK/build/cmake/android.toolchain.cmake -DANDROID_ABI=arm64-v8a -DANDROID_PLATFORM=android-23 -DCMAKE_C_FLAGS=-march=armv8.4a+dotprod ..
-make
-```
-
-Install [termux](https://github.com/termux/termux-app#installation) on your device and run `termux-setup-storage` to get access to your SD card (if Android 11+ then run the command twice).
-
-Finally, copy these built `llama` binaries and the model file to your device storage. Because the file permissions in the Android sdcard cannot be changed, you can copy the executable files to the `/data/data/com.termux/files/home/bin` path, and then execute the following commands in Termux to add executable permission:
-
-(Assumed that you have pushed the built executable files to the /sdcard/llama.cpp/bin path using `adb push`)
-```
-$cp -r /sdcard/llama.cpp/bin /data/data/com.termux/files/home/
-$cd /data/data/com.termux/files/home/bin
-$chmod +x ./*
-```
-
-Download models and push them to `/sdcard/llama.cpp/`, then move it to `/data/data/com.termux/files/home/model/`
-
-```
-$mv /sdcard/llama.cpp/ggml-model-Q4_K_M.gguf /data/data/com.termux/files/home/model/
-$mv /sdcard/llama.cpp/mmproj-model-f16.gguf /data/data/com.termux/files/home/model/
-```
-
-Now, you can start chatting:
-```
-$cd /data/data/com.termux/files/home/bin
-$./llama-minicpmv-cli -m ../model/ggml-model-Q4_K_M.gguf --mmproj ../model/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
+./build/bin/llama-minicpmv-cli -m ../MiniCPM-V-2_6/model/ggml-model-Q4_K_M.gguf --mmproj ../MiniCPM-V-2_6/mmproj-model-f16.gguf -c 4096 --temp 0.7 --top-p 0.8 --top-k 100 --repeat-penalty 1.05 --image xx.jpg  -p "What is in the image?"
 ```

examples/llava/clip.cpp

@@ -4,31 +4,12 @@
 // Note: Even when using identical normalized image inputs (see normalize_image_u8_to_f32()) we have a significant difference in resulting embeddings compared to pytorch
 #include "clip.h"
 #include "ggml.h"
+#include "ggml-cpp.h"
 #include "ggml-cpu.h"
 #include "ggml-alloc.h"
 #include "ggml-backend.h"
 #include "gguf.h"
 
-//#ifdef GGML_USE_CUDA
-//#include "ggml-cuda.h"
-//#endif
-//
-//#ifdef GGML_USE_SYCL
-//#include "ggml-sycl.h"
-//#endif
-//
-//#ifdef GGML_USE_METAL
-//#include "ggml-metal.h"
-//#endif
-//
-//#ifdef GGML_USE_CANN
-//#include "ggml-cann.h"
-//#endif
-//
-//#ifdef GGML_USE_VULKAN
-//#include "ggml-vulkan.h"
-//#endif
-
 #define STB_IMAGE_IMPLEMENTATION
 #include "stb_image.h"
 
@@ -155,6 +136,8 @@ static std::string format(const char * fmt, ...) {
 #define TN_MVLM_PROJ_BLOCK "mm.model.mb_block.%d.block.%d.%s"
 #define TN_MVLM_PROJ_PEG   "mm.model.peg.%d.%s"
 #define TN_IMAGE_NEWLINE   "model.image_newline"
+#define TN_MM_INP_PROJ     "mm.input_projection.weight" // gemma3
+#define TN_MM_SOFT_EMB_N   "mm.soft_emb_norm.weight"    // gemma3
 
 #define TN_MINICPMV_POS_EMBD_K "resampler.pos_embed_k"
 #define TN_MINICPMV_QUERY "resampler.query"
@@ -181,6 +164,7 @@ enum projector_type {
     PROJECTOR_TYPE_RESAMPLER,
     PROJECTOR_TYPE_GLM_EDGE,
     PROJECTOR_TYPE_MERGER,
+    PROJECTOR_TYPE_GEMMA3,
     PROJECTOR_TYPE_UNKNOWN,
 };
 
@@ -191,6 +175,7 @@ static std::map<projector_type, std::string> PROJECTOR_TYPE_NAMES = {
     { PROJECTOR_TYPE_RESAMPLER, "resampler"},
     { PROJECTOR_TYPE_GLM_EDGE, "adapter"},
     { PROJECTOR_TYPE_MERGER, "qwen2vl_merger"},
+    { PROJECTOR_TYPE_GEMMA3, "gemma3"},
 };
 
 
@@ -317,7 +302,7 @@ static projector_type clip_projector_type_from_string(const std::string & name)
             return kv.first;
         }
     }
-    return PROJECTOR_TYPE_UNKNOWN;
+    throw std::runtime_error(format("Unknown projector type: %s", name.c_str()));
 }
 
 #ifdef CLIP_DEBUG_FUNCTIONS
@@ -574,6 +559,10 @@ struct clip_vision_model {
     struct ggml_tensor * mm_model_ln_kv_b;
     struct ggml_tensor * mm_model_ln_post_w;
     struct ggml_tensor * mm_model_ln_post_b;
+
+    // gemma3
+    struct ggml_tensor * mm_input_proj_w;
+    struct ggml_tensor * mm_soft_emb_norm_w;
 };
 
 struct clip_ctx {
@@ -588,7 +577,7 @@ struct clip_ctx {
     struct clip_vision_model vision_model;
     projector_type proj_type = PROJECTOR_TYPE_MLP;
 
-    int32_t max_feature_layer;
+    int32_t max_feature_layer; // unused in newer models like gemma3
     float image_mean[3];
     float image_std[3];
     bool use_gelu = false;
@@ -600,21 +589,209 @@ struct clip_ctx {
     bool has_post_norm = false;
     bool has_patch_bias = false;
 
-    struct gguf_context * ctx_gguf;
-    struct ggml_context * ctx_data;
+    struct gguf_context * ctx_gguf = nullptr;
+    struct ggml_context * ctx_data = nullptr;
 
     std::vector<uint8_t> buf_compute_meta;
 
-    // memory buffers to evaluate the model
-    ggml_backend_buffer_t params_buffer  = NULL;
+    std::vector<ggml_backend_t> backend_ptrs;
+    std::vector<ggml_backend_buffer_type_t> backend_buft;
 
-    ggml_backend_t backend       = NULL;
-    ggml_gallocr_t compute_alloc = NULL;
+    ggml_backend_t backend     = nullptr;
+    ggml_backend_t backend_cpu = nullptr;
+    ggml_backend_buffer_t buf  = nullptr;
 
-    struct clip_image_size * load_image_size;
+    ggml_backend_sched_ptr sched;
+
+    struct clip_image_size * load_image_size = nullptr;
+
+    clip_ctx(clip_context_params & ctx_params) {
+        backend_cpu = ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_CPU, nullptr);
+        backend     = ctx_params.use_gpu
+                        ? ggml_backend_init_by_type(GGML_BACKEND_DEVICE_TYPE_GPU, nullptr)
+                        : nullptr;
+
+        if (backend) {
+            LOG_INF("%s: CLIP using %s backend\n", __func__, ggml_backend_name(backend));
+            backend_ptrs.push_back(backend);
+            backend_buft.push_back(ggml_backend_get_default_buffer_type(backend));
+        } else {
+            backend = backend_cpu;
+            LOG_INF("%s: CLIP using CPU backend\n", __func__);
+        }
+
+        backend_ptrs.push_back(backend_cpu);
+        backend_buft.push_back(ggml_backend_get_default_buffer_type(backend_cpu));
+
+        sched.reset(
+            ggml_backend_sched_new(backend_ptrs.data(), backend_buft.data(), backend_ptrs.size(), 8192, false)
+        );
+    }
+
+    ~clip_ctx() {
+        ggml_free(ctx_data);
+        gguf_free(ctx_gguf);
+        ggml_backend_buffer_free(buf);
+        ggml_backend_free(backend);
+        if (backend_cpu != backend) {
+            ggml_backend_free(backend_cpu);
+        }
+    }
 };
 
-static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32_batch * imgs, struct clip_image_size * load_image_size, bool is_inf = false) {
+static ggml_cgraph * clip_image_build_graph_siglip(clip_ctx * ctx, const clip_image_f32_batch * imgs) {
+    const auto & model = ctx->vision_model;
+    const auto & hparams = model.hparams;
+
+    const int image_size = hparams.image_size;
+    int image_size_width  = image_size;
+    int image_size_height = image_size;
+
+    const int patch_size           = hparams.patch_size;
+    const int num_patches          = ((image_size_width / patch_size) * (image_size_height / patch_size));
+    const int hidden_size          = hparams.hidden_size;
+    const int n_head               = hparams.n_head;
+    const int d_head               = hidden_size / n_head;
+    const int n_layer              = hparams.n_layer;
+    const float eps                = hparams.eps;
+
+    GGML_ASSERT(imgs->size == 1); // batch_size == 1
+
+    struct ggml_init_params params = {
+        /*.mem_size   =*/ ctx->buf_compute_meta.size(),
+        /*.mem_buffer =*/ ctx->buf_compute_meta.data(),
+        /*.no_alloc   =*/ true,
+    };
+
+    struct ggml_context * ctx0 = ggml_init(params);
+    struct ggml_cgraph * gf = ggml_new_graph(ctx0);
+
+    // input raw
+    struct ggml_tensor * inp_raw = ggml_new_tensor_3d(ctx0, GGML_TYPE_F32, image_size_width, image_size_height, 3);
+    ggml_set_name(inp_raw, "inp_raw");
+    ggml_set_input(inp_raw);
+
+    struct ggml_tensor * inp = ggml_conv_2d(ctx0, model.patch_embeddings_0, inp_raw, patch_size, patch_size, 0, 0, 1, 1);
+    inp = ggml_reshape_2d(ctx0, inp, num_patches, hidden_size);
+    inp = ggml_cont(ctx0, ggml_transpose(ctx0, inp));
+    inp = ggml_add(ctx0, inp, model.patch_bias);
+
+    // position embeddings
+    struct ggml_tensor * embeddings = ggml_add(ctx0, inp, model.position_embeddings);
+
+    // loop over layers
+    for (int il = 0; il < n_layer; il++) {
+        struct ggml_tensor * cur = embeddings; // embeddings = residual, cur = hidden_states
+
+        // layernorm1
+        {
+            cur = ggml_norm(ctx0, cur, eps);
+            cur = ggml_add(ctx0, ggml_mul(ctx0, cur, model.layers[il].ln_1_w), model.layers[il].ln_1_b);
+        }
+
+        // self-attention
+        {
+
+            struct ggml_tensor * Q =
+                ggml_add(ctx0, ggml_mul_mat(ctx0, model.layers[il].q_w, cur), model.layers[il].q_b);
+
+            Q = ggml_reshape_3d(ctx0, Q, d_head, n_head, num_patches);
+            Q = ggml_cont(ctx0, ggml_permute(ctx0, Q, 0, 2, 1, 3));
+
+            struct ggml_tensor * K =
+                ggml_add(ctx0, ggml_mul_mat(ctx0, model.layers[il].k_w, cur), model.layers[il].k_b);
+
+            K = ggml_reshape_3d(ctx0, K, d_head, n_head, num_patches);
+            K = ggml_cont(ctx0, ggml_permute(ctx0, K, 0, 2, 1, 3));
+
+            struct ggml_tensor * V =
+                ggml_add(ctx0, ggml_mul_mat(ctx0, model.layers[il].v_w, cur), model.layers[il].v_b);
+
+            V = ggml_reshape_3d(ctx0, V, d_head, n_head, num_patches);
+            V = ggml_cont(ctx0, ggml_permute(ctx0, V, 1, 2, 0, 3));
+
+            struct ggml_tensor * KQ = ggml_mul_mat(ctx0, K, Q);
+            KQ = ggml_scale_inplace(ctx0, KQ, 1.0f / sqrtf((float)d_head));
+            KQ = ggml_soft_max_inplace(ctx0, KQ);
+
+            struct ggml_tensor * KQV = ggml_mul_mat(ctx0, V, KQ);
+            KQV = ggml_reshape_3d(ctx0, KQV, d_head, num_patches, n_head);
+            KQV = ggml_permute(ctx0, KQV, 0, 2, 1, 3);
+
+            cur = ggml_cont_2d(ctx0, KQV, hidden_size, num_patches);
+        }
+
+        // attention output
+        cur = ggml_add(ctx0, ggml_mul_mat(ctx0, model.layers[il].o_w, cur), model.layers[il].o_b);
+
+        // re-add the layer input, e.g., residual
+        cur = ggml_add(ctx0, cur, embeddings);
+
+        embeddings = cur; // embeddings = residual, cur = hidden_states
+
+        // layernorm2
+        {
+            cur = ggml_norm(ctx0, cur, eps);
+            cur = ggml_add(ctx0, ggml_mul(ctx0, cur, model.layers[il].ln_2_w), model.layers[il].ln_2_b);
+        }
+
+        cur = ggml_mul_mat(ctx0, model.layers[il].ff_i_w, cur);
+        cur = ggml_add(ctx0, cur, model.layers[il].ff_i_b);
+
+        // siglip uses gelu
+        cur = ggml_gelu(ctx0, cur);
+
+        cur = ggml_mul_mat(ctx0, model.layers[il].ff_o_w, cur);
+        cur = ggml_add(ctx0, cur, model.layers[il].ff_o_b);
+
+        // residual 2
+        cur = ggml_add(ctx0, embeddings, cur);
+
+        embeddings = cur;
+    }
+
+    // post-layernorm
+    if (ctx->has_post_norm) {
+        embeddings = ggml_norm(ctx0, embeddings, eps);
+        ggml_set_name(embeddings, "post_ln");
+
+        embeddings = ggml_add(ctx0, ggml_mul(ctx0, embeddings, model.post_ln_w), model.post_ln_b);
+    }
+
+    if (ctx->proj_type == PROJECTOR_TYPE_GEMMA3) {
+        const int batch_size = 1;
+        const int mm_tokens_per_image = 256; // default value for gemma3
+        const int tokens_per_side = sqrt(mm_tokens_per_image);
+        const int patches_per_image = sqrt(num_patches);
+        const int kernel_size = patches_per_image / tokens_per_side;
+
+        embeddings = ggml_cont(ctx0, ggml_transpose(ctx0, embeddings));
+        embeddings = ggml_reshape_4d(ctx0, embeddings, patches_per_image, patches_per_image, hidden_size, batch_size);
+
+        // doing a pool2d to reduce the number of output tokens to 256
+        embeddings = ggml_pool_2d(ctx0, embeddings, GGML_OP_POOL_AVG, kernel_size, kernel_size, kernel_size, kernel_size, 0, 0);
+        embeddings = ggml_reshape_3d(ctx0, embeddings, embeddings->ne[0] * embeddings->ne[0], hidden_size, batch_size);
+        embeddings = ggml_cont(ctx0, ggml_transpose(ctx0, embeddings));
+
+        // apply norm before projection
+        embeddings = ggml_rms_norm(ctx0, embeddings, eps);
+        embeddings = ggml_mul(ctx0, embeddings, model.mm_soft_emb_norm_w);
+
+        // apply projection
+        embeddings = ggml_mul_mat(ctx0,
+            ggml_cont(ctx0, ggml_transpose(ctx0, model.mm_input_proj_w)),
+            embeddings);
+    }
+
+    // build the graph
+    ggml_build_forward_expand(gf, embeddings);
+
+    ggml_free(ctx0);
+
+    return gf;
+}
+
+static ggml_cgraph * clip_image_build_graph_legacy(clip_ctx * ctx, const clip_image_f32_batch * imgs, struct clip_image_size * load_image_size, bool is_inf = false) {
     if (!ctx->has_vision_encoder) {
         LOG_ERR("This gguf file seems to have no vision encoder\n");
         return nullptr;
@@ -1160,7 +1337,8 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
         } else {
             GGML_ABORT("fatel error");
         }
-    } else if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
+    }
+    else if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
         embeddings = ggml_reshape_3d(ctx0, embeddings, hidden_size * 4, num_positions / 4, batch_size);
 
         embeddings = ggml_mul_mat(ctx0, model.mm_0_w, embeddings);
@@ -1182,8 +1360,25 @@ static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32
     return gf;
 }
 
+static ggml_cgraph * clip_image_build_graph(clip_ctx * ctx, const clip_image_f32_batch * imgs, struct clip_image_size * load_image_size, bool is_inf = false) {
+    if (ctx->proj_type == PROJECTOR_TYPE_GEMMA3) {
+        return clip_image_build_graph_siglip(ctx, imgs);
+    } else {
+        // TODO: we should have one build_* function per model
+        return clip_image_build_graph_legacy(ctx, imgs, load_image_size, is_inf);
+    }
+}
+
 // read and create ggml_context containing the tensors and their data
 struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
+    return clip_init(fname, clip_context_params{
+        /* use_gpu */   true,
+        /* verbosity */ verbosity,
+    });
+}
+
+struct clip_ctx * clip_init(const char * fname, struct clip_context_params ctx_params) {
+    int verbosity = ctx_params.verbosity;
     struct ggml_context * meta = NULL;
 
     struct gguf_init_params params = {
@@ -1277,7 +1472,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
         }
     }
 
-    clip_ctx * new_clip = new clip_ctx{};
+    clip_ctx * new_clip = new clip_ctx(ctx_params);
 
     // update projector type
     {
@@ -1296,36 +1491,6 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
         }
     }
 
-//#ifdef GGML_USE_CUDA
-//    new_clip->backend = ggml_backend_cuda_init(0);
-//    LOG_INF("%s: CLIP using CUDA backend\n", __func__);
-//#endif
-//
-//#ifdef GGML_USE_METAL
-//    new_clip->backend = ggml_backend_metal_init();
-//    LOG_INF("%s: CLIP using Metal backend\n", __func__);
-//#endif
-//
-//#ifdef GGML_USE_CANN
-//    new_clip->backend = ggml_backend_cann_init(0);
-//    LOG_INF("%s: CLIP using CANN backend\n", __func__);
-//#endif
-//
-//#ifdef GGML_USE_VULKAN
-//    new_clip->backend = ggml_backend_vk_init(0);
-//    LOG_INF("%s: CLIP using Vulkan backend\n", __func__);
-//#endif
-//
-//#ifdef GGML_USE_SYCL
-//    new_clip->backend = ggml_backend_sycl_init(0);
-//    LOG_INF("%s: CLIP using SYCL backend\n", __func__);
-//#endif
-
-    if (!new_clip->backend) {
-        new_clip->backend = ggml_backend_cpu_init();
-        LOG_INF("%s: CLIP using CPU backend\n", __func__);
-    }
-
     // model size and capabilities
     {
         int idx = get_key_idx(ctx, KEY_HAS_TEXT_ENC);
@@ -1363,8 +1528,12 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
         GGML_ASSERT(new_clip->has_vision_encoder);
         GGML_ASSERT(!new_clip->has_text_encoder);
 
-        idx = get_key_idx(ctx, KEY_USE_GELU);
-        new_clip->use_gelu = gguf_get_val_bool(ctx, idx);
+        try {
+            idx = get_key_idx(ctx, KEY_USE_GELU);
+            new_clip->use_gelu = gguf_get_val_bool(ctx, idx);
+        } catch (std::runtime_error & /*e*/) {
+            new_clip->use_gelu = false;
+        }
 
         try {
             idx = get_key_idx(ctx, KEY_USE_SILU);
@@ -1378,6 +1547,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
             LOG_INF("%s: vision_encoder: %d\n", __func__, new_clip->has_vision_encoder);
             LOG_INF("%s: llava_projector:  %d\n", __func__, new_clip->has_llava_projector);
             LOG_INF("%s: minicpmv_projector:  %d\n", __func__, new_clip->has_minicpmv_projector);
+            LOG_INF("%s: minicpmv_version:  %d\n", __func__, new_clip->minicpmv_version);
             LOG_INF("%s: glm_projector:  %d\n", __func__, new_clip->has_glm_projector);
             LOG_INF("%s: model size:     %.2f MB\n", __func__, model_size / 1024.0 / 1024.0);
             LOG_INF("%s: metadata size:  %.2f MB\n", __func__, ggml_get_mem_size(meta) / 1024.0 / 1024.0);
@@ -1420,7 +1590,9 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
         }
 
         // alloc memory and offload data
-        new_clip->params_buffer = ggml_backend_alloc_ctx_tensors(new_clip->ctx_data, new_clip->backend);
+        ggml_backend_buffer_type_t buft = ggml_backend_get_default_buffer_type(new_clip->backend);
+        new_clip->buf = ggml_backend_alloc_ctx_tensors_from_buft(new_clip->ctx_data, buft);
+        ggml_backend_buffer_set_usage(new_clip->buf, GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
         for (int i = 0; i < n_tensors; ++i) {
             const char * name = gguf_get_tensor_name(ctx, i);
             struct ggml_tensor * cur = ggml_get_tensor(new_clip->ctx_data, name);
@@ -1433,7 +1605,7 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
                 return nullptr;
             }
             int num_bytes = ggml_nbytes(cur);
-            if (ggml_backend_buffer_is_host(new_clip->params_buffer)) {
+            if (ggml_backend_buft_is_host(buft)) {
                 // for the CPU and Metal backend, we can read directly into the tensor
                 fin.read(reinterpret_cast<char *>(cur->data), num_bytes);
             } else {
@@ -1569,11 +1741,17 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
         }
 
         try {
-            vision_model.patch_embeddings_0    = get_tensor(new_clip->ctx_data, TN_PATCH_EMBD);
+            vision_model.patch_embeddings_0 = get_tensor(new_clip->ctx_data, TN_PATCH_EMBD);
+        } catch(const std::exception& /*e*/) {
+            vision_model.patch_embeddings_0 = nullptr;
+        }
+
+        try {
             vision_model.position_embeddings = get_tensor(new_clip->ctx_data, format(TN_POS_EMBD, "v"));
         } catch(const std::exception& /*e*/) {
-            LOG_ERR("%s: failed to load vision model tensors\n", __func__);
+            vision_model.position_embeddings = nullptr;
         }
+
         try {
             vision_model.patch_embeddings_1    = get_tensor(new_clip->ctx_data, TN_PATCH_EMBD_1);
         } catch(const std::exception& /*e*/) {
@@ -1684,6 +1862,10 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
             vision_model.mm_1_w = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "weight"));
             vision_model.mm_1_b = get_tensor(new_clip->ctx_data, format(TN_LLAVA_PROJ, 2, "bias"));
         }
+        else if (new_clip->proj_type == PROJECTOR_TYPE_GEMMA3) {
+            vision_model.mm_input_proj_w    = get_tensor(new_clip->ctx_data, TN_MM_INP_PROJ);
+            vision_model.mm_soft_emb_norm_w = get_tensor(new_clip->ctx_data, TN_MM_SOFT_EMB_N);
+        }
         else {
             std::string proj_type = PROJECTOR_TYPE_NAMES[new_clip->proj_type];
             throw std::runtime_error(format("%s: don't support projector with: %s currently\n", __func__, proj_type.c_str()));
@@ -1719,14 +1901,21 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
     // measure mem requirement and allocate
     {
         new_clip->buf_compute_meta.resize(GGML_DEFAULT_GRAPH_SIZE * ggml_tensor_overhead() + ggml_graph_overhead());
-        new_clip->compute_alloc = ggml_gallocr_new(ggml_backend_get_default_buffer_type(new_clip->backend));
         clip_image_f32_batch batch;
         batch.size = 1;
         batch.data = nullptr;
         ggml_cgraph * gf = clip_image_build_graph(new_clip, &batch, nullptr, false);
-        ggml_gallocr_reserve(new_clip->compute_alloc, gf);
-        size_t compute_memory_buffer_size = ggml_gallocr_get_buffer_size(new_clip->compute_alloc, 0);
-        LOG_INF("%s: compute allocated memory: %.2f MB\n", __func__, compute_memory_buffer_size /1024.0/1024.0);
+        ggml_backend_sched_reserve(new_clip->sched.get(), gf);
+        for (size_t i = 0; i < new_clip->backend_ptrs.size(); ++i) {
+            ggml_backend_t backend = new_clip->backend_ptrs[i];
+            ggml_backend_buffer_type_t buft = new_clip->backend_buft[i];
+            size_t size = ggml_backend_sched_get_buffer_size(new_clip->sched.get(), backend);
+            if (size > 1) {
+                LOG_INF("%s: %10s compute buffer size = %8.2f MiB\n", __func__,
+                        ggml_backend_buft_name(buft),
+                        size / 1024.0 / 1024.0);
+            }
+        }
     }
 
     return new_clip;
@@ -2218,7 +2407,7 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, cli
         return true;
     }
 
-    if (ctx->has_glm_projector) {
+    if (ctx->has_glm_projector || ctx->proj_type == PROJECTOR_TYPE_GEMMA3) {
         res_imgs->size = 1;
         res_imgs->data = new clip_image_f32[res_imgs->size];
         clip_image_u8 resized_image;
@@ -2407,12 +2596,6 @@ ggml_tensor * clip_get_newline_tensor(const struct clip_ctx * ctx) {
 }
 
 void clip_free(clip_ctx * ctx) {
-    ggml_free(ctx->ctx_data);
-    gguf_free(ctx->ctx_gguf);
-
-    ggml_backend_buffer_free(ctx->params_buffer);
-    ggml_backend_free(ctx->backend);
-    ggml_gallocr_free(ctx->compute_alloc);
     delete ctx;
 }
 
@@ -2608,8 +2791,9 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
     }
 
     // build the inference graph
+    ggml_backend_sched_reset(ctx->sched.get());
     ggml_cgraph * gf = clip_image_build_graph(ctx, imgs, ctx->load_image_size, true);
-    ggml_gallocr_alloc_graph(ctx->compute_alloc, gf);
+    ggml_backend_sched_alloc_graph(ctx->sched.get(), gf);
 
     // set inputs
     const auto & model = ctx->vision_model;
@@ -2748,6 +2932,9 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
             ggml_backend_tensor_set(positions, positions_data, 0, ggml_nbytes(positions));
             free(positions_data);
         }
+        else if (ctx->proj_type == PROJECTOR_TYPE_GEMMA3) {
+            // do nothing
+        }
         else {
             struct ggml_tensor * positions = ggml_graph_get_tensor(gf, "positions");
 
@@ -2774,11 +2961,13 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
         }
     }
 
-    if (ggml_backend_is_cpu(ctx->backend)) {
-        ggml_backend_cpu_set_n_threads(ctx->backend, n_threads);
-    }
+    ggml_backend_cpu_set_n_threads(ctx->backend_cpu, n_threads);
 
-    ggml_backend_graph_compute(ctx->backend, gf);
+    auto status = ggml_backend_sched_graph_compute(ctx->sched.get(), gf);
+    if (status != GGML_STATUS_SUCCESS) {
+        LOG_ERR("%s: ggml_backend_sched_graph_compute failed with error %d\n", __func__, status);
+        return false;
+    }
 
     // the last node is the embedding tensor
     struct ggml_tensor * embeddings = ggml_graph_node(gf, -1);
@@ -2958,6 +3147,9 @@ int clip_n_mmproj_embd(const struct clip_ctx * ctx) {
     if (ctx->proj_type == PROJECTOR_TYPE_MERGER) {
         return ctx->vision_model.mm_1_b->ne[0];
     }
+    if (ctx->proj_type == PROJECTOR_TYPE_GEMMA3) {
+        return ctx->vision_model.mm_input_proj_w->ne[0];
+    }
 
     std::string proj_type = PROJECTOR_TYPE_NAMES[ctx->proj_type];
     throw std::runtime_error(format("%s: don't support projector with: %s currently\n", __func__, proj_type.c_str()));

examples/llava/clip.h

@@ -39,8 +39,15 @@ struct clip_image_f32_batch {
     size_t size;
 };
 
-CLIP_API struct clip_ctx * clip_model_load    (const char * fname, int verbosity);
-CLIP_API struct clip_ctx * clip_model_load_cpu(const char * fname, int verbosity);
+struct clip_context_params {
+    bool use_gpu;
+    int verbosity;
+};
+
+// deprecated, use clip_init
+CLIP_API struct clip_ctx * clip_model_load(const char * fname, int verbosity);
+
+CLIP_API struct clip_ctx * clip_init(const char * fname, struct clip_context_params ctx_params);
 
 CLIP_API void clip_free(struct clip_ctx * ctx);
 

examples/llava/convert_image_encoder_to_gguf.py

@@ -89,6 +89,7 @@ def bytes_to_unicode():
 ap = argparse.ArgumentParser()
 ap.add_argument("-m", "--model-dir", help="Path to model directory cloned from HF Hub", required=True)
 ap.add_argument("--use-f32", action="store_true", default=False, help="Use f32 instead of f16")
+ap.add_argument('--bigendian', action="store_true", default=False, help="Model is executed on big-endian machine")
 ap.add_argument("--text-only", action="store_true", required=False,
                 help="Save a text-only model. It can't be used to encode images")
 ap.add_argument("--vision-only", action="store_true", required=False,
@@ -191,7 +192,7 @@ output_dir = args.output_dir if args.output_dir is not None else dir_model
 os.makedirs(output_dir, exist_ok=True)
 output_prefix = os.path.basename(output_dir).replace("ggml_", "")
 fname_out = os.path.join(output_dir, f"{fname_middle}model-{ftype_str[ftype]}.gguf")
-fout = GGUFWriter(path=fname_out, arch="clip")
+fout = GGUFWriter(path=fname_out, arch="clip", endianess=GGUFEndian.LITTLE if not args.bigendian else GGUFEndian.BIG)
 
 fout.add_bool("clip.has_text_encoder", has_text_encoder)
 fout.add_bool("clip.has_vision_encoder", has_vision_encoder)

examples/llava/gemma3-cli.cpp

@@ -0,0 +1,341 @@
+#include "arg.h"
+#include "log.h"
+#include "common.h"
+#include "sampling.h"
+#include "clip.h"
+#include "stb_image.h"
+#include "llama.h"
+#include "ggml.h"
+#include "console.h"
+
+#include <vector>
+#include <limits.h>
+#include <inttypes.h>
+
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+#include <signal.h>
+#include <unistd.h>
+#elif defined (_WIN32)
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#define NOMINMAX
+#endif
+#include <windows.h>
+#include <signal.h>
+#endif
+
+static bool g_is_generating = false;
+
+/**
+ * Please note that this is NOT a production-ready stuff.
+ * It is a playground for trying Gemma 3 vision capabilities.
+ * For contributors: please keep this code simple and easy to understand.
+ */
+
+static void show_additional_info(int /*argc*/, char ** argv) {
+    LOG(
+        "Experimental CLI for using Gemma 3 vision model\n\n"
+        "Usage: %s [options] -m <model> --mmproj <mmproj> --image <image> -p <prompt>\n\n"
+        "  -m and --mmproj are required\n"
+        "  --image and -p are optional, if NOT provided, the CLI will run in chat mode\n",
+        argv[0]
+    );
+}
+
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__)) || defined (_WIN32)
+static void sigint_handler(int signo) {
+    if (signo == SIGINT) {
+        if (g_is_generating) {
+            g_is_generating = false;
+        } else {
+            console::cleanup();
+            LOG("\nInterrupted by user\n");
+            _exit(130);
+        }
+    }
+}
+#endif
+
+struct gemma3_context {
+    struct clip_ctx    * ctx_clip = NULL;
+    common_init_result   llama_init;
+
+    llama_model       * model;
+    llama_context     * lctx;
+    const llama_vocab * vocab;
+    llama_batch         batch;
+
+    int n_threads    = 1;
+    llama_pos n_past = 0;
+
+    gemma3_context(common_params & params) : llama_init(common_init_from_params(params)) {
+        model = llama_init.model.get();
+        lctx = llama_init.context.get();
+        vocab = llama_model_get_vocab(model);
+        n_threads = params.cpuparams.n_threads;
+        batch = llama_batch_init(params.n_batch, 0, 1);
+        init_clip_model(params);
+    }
+
+    void init_clip_model(common_params & params) {
+        const char * clip_path = params.mmproj.c_str();
+        ctx_clip = clip_model_load(clip_path, params.verbosity > 1);
+    }
+
+    ~gemma3_context() {
+        clip_free(ctx_clip);
+    }
+};
+
+struct decode_embd_batch {
+    std::vector<llama_pos>      pos;
+    std::vector<int32_t>        n_seq_id;
+    std::vector<llama_seq_id>   seq_id_0;
+    std::vector<llama_seq_id *> seq_ids;
+    std::vector<int8_t>         logits;
+    llama_batch batch;
+    decode_embd_batch(float * embd, int32_t n_tokens, llama_pos pos_0, llama_seq_id seq_id) {
+        pos     .resize(n_tokens);
+        n_seq_id.resize(n_tokens);
+        seq_ids .resize(n_tokens + 1);
+        logits  .resize(n_tokens);
+        seq_id_0.resize(1);
+        seq_id_0[0] = seq_id;
+        seq_ids [n_tokens] = nullptr;
+        batch = {
+            /*n_tokens       =*/ n_tokens,
+            /*tokens         =*/ nullptr,
+            /*embd           =*/ embd,
+            /*pos            =*/ pos.data(),
+            /*n_seq_id       =*/ n_seq_id.data(),
+            /*seq_id         =*/ seq_ids.data(),
+            /*logits         =*/ logits.data(),
+        };
+        for (int i = 0; i < n_tokens; i++) {
+            batch.pos     [i] = pos_0 + i;
+            batch.n_seq_id[i] = 1;
+            batch.seq_id  [i] = seq_id_0.data();
+            batch.logits  [i] = false;
+        }
+    }
+};
+
+static int eval_text(gemma3_context & ctx, std::string input, bool logits_last = false) {
+    llama_tokens tokens = common_tokenize(ctx.lctx, input, false, true);
+    common_batch_clear(ctx.batch);
+    for (llama_token & t : tokens) {
+        common_batch_add(ctx.batch, t, ctx.n_past++, {0}, false);
+    }
+    if (logits_last) {
+        ctx.batch.logits[ctx.batch.n_tokens - 1] = true;
+    }
+    // LOG("eval_text (n_tokens = %d): %s\n", (int)tokens.size(), input.c_str());
+    if (llama_decode(ctx.lctx, ctx.batch)) {
+        LOG_ERR("Failed to decode text\n");
+        return 1;
+    }
+    return 0;
+}
+
+static int eval_image(gemma3_context & ctx, std::string & fname) {
+    std::vector<float> image_embd_v;
+    int n_embd = llama_model_n_embd(ctx.model);
+    int n_tokens = 256;
+    image_embd_v.resize(n_tokens * n_embd);
+
+    bool ok;
+    struct clip_image_u8 * img_u8 = clip_image_u8_init();
+    ok = clip_image_load_from_file(fname.c_str(), img_u8);
+    if (!ok) {
+        LOG_ERR("Unable to load image %s\n", fname.c_str());
+        clip_image_u8_free(img_u8);
+        return 2; // non-fatal error
+    }
+
+    clip_image_f32_batch batch_f32;
+    ok = clip_image_preprocess(ctx.ctx_clip, img_u8, &batch_f32);
+    if (!ok) {
+        LOG_ERR("Unable to preprocess image\n");
+        clip_image_f32_batch_free(&batch_f32);
+        clip_image_u8_free(img_u8);
+        return 1;
+    }
+
+    int64_t t0 = ggml_time_ms();
+    LOG("Encoding image %s\n", fname.c_str());
+    ok = clip_image_batch_encode(ctx.ctx_clip, ctx.n_threads, &batch_f32, image_embd_v.data());
+    if (!ok) {
+        LOG_ERR("Unable to encode image\n");
+        clip_image_f32_batch_free(&batch_f32);
+        clip_image_u8_free(img_u8);
+        return 1;
+    }
+    LOG("Image encoded in %" PRId64 " ms\n", ggml_time_ms() - t0);
+
+    clip_image_f32_batch_free(&batch_f32);
+    clip_image_u8_free(img_u8);
+
+    // decode image embeddings
+    int64_t t1 = ggml_time_ms();
+    eval_text(ctx, "<start_of_image>");
+    llama_set_causal_attn(ctx.lctx, false);
+    decode_embd_batch batch_img(image_embd_v.data(), n_tokens, ctx.n_past, 0);
+    if (llama_decode(ctx.lctx, batch_img.batch)) {
+        LOG_ERR("failed to decode image\n");
+        return 1;
+    }
+    ctx.n_past += n_tokens;
+    llama_set_causal_attn(ctx.lctx, true);
+    eval_text(ctx, "<end_of_image>");
+    LOG("Image decoded in %" PRId64 " ms\n", ggml_time_ms() - t1);
+    return 0;
+}
+
+static int generate_response(gemma3_context & ctx, common_sampler * smpl, int n_predict) {
+    for (int i = 0; i < n_predict; i++) {
+        if (i > n_predict || !g_is_generating) {
+            printf("\n");
+            break;
+        }
+
+        llama_token token_id = common_sampler_sample(smpl, ctx.lctx, -1);
+        common_sampler_accept(smpl, token_id, true);
+
+        if (llama_vocab_is_eog(ctx.vocab, token_id)) {
+            printf("\n");
+            break; // end of generation
+        }
+
+        printf("%s", common_token_to_piece(ctx.lctx, token_id).c_str());
+        fflush(stdout);
+
+        // eval the token
+        common_batch_clear(ctx.batch);
+        common_batch_add(ctx.batch, token_id, ctx.n_past++, {0}, true);
+        if (llama_decode(ctx.lctx, ctx.batch)) {
+            LOG_ERR("failed to decode token\n");
+            return 1;
+        }
+    }
+    return 0;
+}
+
+int main(int argc, char ** argv) {
+    ggml_time_init();
+
+    common_params params;
+    params.sampling.temp = 0.2; // lower temp by default for better quality
+
+    if (!common_params_parse(argc, argv, params, LLAMA_EXAMPLE_LLAVA, show_additional_info)) {
+        return 1;
+    }
+
+    common_init();
+
+    if (params.mmproj.empty()) {
+        show_additional_info(argc, argv);
+        return 1;
+    }
+
+    gemma3_context ctx(params);
+    printf("%s: %s\n", __func__, params.model.c_str());
+
+    bool is_single_turn = !params.prompt.empty() && !params.image.empty();
+
+    struct common_sampler * smpl = common_sampler_init(ctx.model, params.sampling);
+    int n_predict = params.n_predict < 0 ? INT_MAX : params.n_predict;
+
+    // ctrl+C handling
+    {
+#if defined (__unix__) || (defined (__APPLE__) && defined (__MACH__))
+        struct sigaction sigint_action;
+        sigint_action.sa_handler = sigint_handler;
+        sigemptyset (&sigint_action.sa_mask);
+        sigint_action.sa_flags = 0;
+        sigaction(SIGINT, &sigint_action, NULL);
+#elif defined (_WIN32)
+        auto console_ctrl_handler = +[](DWORD ctrl_type) -> BOOL {
+            return (ctrl_type == CTRL_C_EVENT) ? (sigint_handler(SIGINT), true) : false;
+        };
+        SetConsoleCtrlHandler(reinterpret_cast<PHANDLER_ROUTINE>(console_ctrl_handler), true);
+#endif
+    }
+
+    if (eval_text(ctx, "<bos>")) {
+        return 1;
+    }
+
+    if (is_single_turn) {
+        g_is_generating = true;
+        if (eval_text(ctx, "<start_of_turn>user\n")) {
+            return 1;
+        }
+        for (auto & fname : params.image) {
+            if (eval_image(ctx, fname)) {
+                return 1;
+            }
+        }
+        if (eval_text(ctx, params.prompt + "<end_of_turn><start_of_turn>model\n", true)) {
+            return 1;
+        }
+        if (generate_response(ctx, smpl, n_predict)) {
+            return 1;
+        }
+
+    } else {
+        LOG("\n Running in chat mode, available commands:");
+        LOG("\n   /image <path>    load an image");
+        LOG("\n   /clear           clear the chat history");
+        LOG("\n   /quit or /exit   exit the program");
+        LOG("\n");
+
+        if (eval_text(ctx, "<start_of_turn>user\n")) {
+            return 1;
+        }
+
+        while (true) {
+            g_is_generating = false;
+            LOG("\n> ");
+            console::set_display(console::user_input);
+            std::string line;
+            console::readline(line, false);
+            console::set_display(console::reset);
+            line = string_strip(line);
+            if (line.empty()) {
+                continue;
+            }
+            if (line == "/quit" || line == "/exit") {
+                break;
+            }
+            if (line == "/clear") {
+                ctx.n_past = 0;
+                llama_kv_self_seq_rm(ctx.lctx, 0, 1, -1); // keep BOS
+                LOG("Chat history cleared\n\n");
+                continue;
+            }
+            g_is_generating = true;
+            if (line.find("/image") == 0) {
+                std::string image = line.substr(7);
+                int res = eval_image(ctx, image);
+                if (res == 2) {
+                    continue; // image not found
+                }
+                if (res) {
+                    return 1;
+                }
+                continue;
+            }
+            if (eval_text(ctx, line + "<end_of_turn><start_of_turn>model\n", true)) {
+                return 1;
+            }
+            if (generate_response(ctx, smpl, n_predict)) {
+                return 1;
+            }
+            if (eval_text(ctx, "<end_of_turn><start_of_turn>user\n")) {
+                return 1;
+            }
+        }
+    }
+
+    return 0;
+}

examples/llava/gemma3_convert_encoder_to_gguf.py

@@ -0,0 +1,307 @@
+import gguf
+import argparse
+import logging
+import sys
+import torch
+import json
+import os
+import numpy as np
+from typing import cast, ContextManager, Any, Iterator
+from pathlib import Path
+from torch import Tensor
+
+logger = logging.getLogger("gemma3-mmproj")
+
+
+# (copied from convert_hf_to_gguf.py)
+# tree of lazy tensors
+class LazyTorchTensor(gguf.LazyBase):
+    _tensor_type = torch.Tensor
+    # to keep the type-checker happy
+    dtype: torch.dtype
+    shape: torch.Size
+
+    # only used when converting a torch.Tensor to a np.ndarray
+    _dtype_map: dict[torch.dtype, type] = {
+        torch.float16: np.float16,
+        torch.float32: np.float32,
+    }
+
+    # used for safetensors slices
+    # ref: https://github.com/huggingface/safetensors/blob/079781fd0dc455ba0fe851e2b4507c33d0c0d407/bindings/python/src/lib.rs#L1046
+    # TODO: uncomment U64, U32, and U16, ref: https://github.com/pytorch/pytorch/issues/58734
+    _dtype_str_map: dict[str, torch.dtype] = {
+        "F64": torch.float64,
+        "F32": torch.float32,
+        "BF16": torch.bfloat16,
+        "F16": torch.float16,
+        # "U64": torch.uint64,
+        "I64": torch.int64,
+        # "U32": torch.uint32,
+        "I32": torch.int32,
+        # "U16": torch.uint16,
+        "I16": torch.int16,
+        "U8": torch.uint8,
+        "I8": torch.int8,
+        "BOOL": torch.bool,
+        "F8_E4M3": torch.float8_e4m3fn,
+        "F8_E5M2": torch.float8_e5m2,
+    }
+
+    def numpy(self) -> gguf.LazyNumpyTensor:
+        dtype = self._dtype_map[self.dtype]
+        return gguf.LazyNumpyTensor(
+            meta=gguf.LazyNumpyTensor.meta_with_dtype_and_shape(dtype, self.shape),
+            args=(self,),
+            func=(lambda s: s.numpy())
+        )
+
+    @classmethod
+    def meta_with_dtype_and_shape(cls, dtype: torch.dtype, shape: tuple[int, ...]) -> Tensor:
+        return torch.empty(size=shape, dtype=dtype, device="meta")
+
+    @classmethod
+    def from_safetensors_slice(cls, st_slice: Any) -> Tensor:
+        dtype = cls._dtype_str_map[st_slice.get_dtype()]
+        shape: tuple[int, ...] = tuple(st_slice.get_shape())
+        lazy = cls(meta=cls.meta_with_dtype_and_shape(dtype, shape), args=(st_slice,), func=lambda s: s[:])
+        return cast(torch.Tensor, lazy)
+
+    @classmethod
+    def __torch_function__(cls, func, types, args=(), kwargs=None):
+        del types  # unused
+
+        if kwargs is None:
+            kwargs = {}
+
+        if func is torch.Tensor.numpy:
+            return args[0].numpy()
+
+        return cls._wrap_fn(func)(*args, **kwargs)
+
+
+class Gemma3VisionTower:
+    hparams: dict
+    gguf_writer: gguf.GGUFWriter
+    fname_out: Path
+    ftype: gguf.LlamaFileType
+
+    @staticmethod
+    def load_hparams(dir_model: Path):
+        with open(dir_model / "config.json", "r", encoding="utf-8") as f:
+            return json.load(f)
+
+    @staticmethod
+    def get_model_part_names(dir_model: Path, prefix: str, suffix: str) -> list[str]:
+        part_names: list[str] = []
+        for filename in os.listdir(dir_model):
+            if filename.startswith(prefix) and filename.endswith(suffix):
+                part_names.append(filename)
+        part_names.sort()
+        return part_names
+
+    def __init__(self,
+                 dir_model: Path,
+                 fname_out: Path,
+                 ftype: gguf.LlamaFileType,
+                 is_big_endian: bool,):
+        hparams = Gemma3VisionTower.load_hparams(dir_model)
+        self.hparams = hparams
+        self.fname_out = fname_out
+        self.ftype = ftype
+        endianess = gguf.GGUFEndian.BIG if is_big_endian else gguf.GGUFEndian.LITTLE
+        self.gguf_writer = gguf.GGUFWriter(path=None, arch="clip", endianess=endianess)
+
+        text_config = hparams["text_config"]
+        vision_config = hparams["vision_config"]
+
+        assert hparams["architectures"][0] == "Gemma3ForConditionalGeneration"
+        assert text_config is not None
+        assert vision_config is not None
+
+        self.gguf_writer.add_string ("clip.projector_type",              "gemma3")
+        self.gguf_writer.add_bool   ("clip.has_text_encoder",            False)
+        self.gguf_writer.add_bool   ("clip.has_vision_encoder",          True)
+        self.gguf_writer.add_bool   ("clip.has_llava_projector",         False) # legacy
+        self.gguf_writer.add_uint32 ("clip.vision.image_size",           vision_config["image_size"])
+        self.gguf_writer.add_uint32 ("clip.vision.patch_size",           vision_config["patch_size"])
+        self.gguf_writer.add_uint32 ("clip.vision.embedding_length",     vision_config["hidden_size"])
+        self.gguf_writer.add_uint32 ("clip.vision.feed_forward_length",  vision_config["intermediate_size"])
+        self.gguf_writer.add_uint32 ("clip.vision.projection_dim",       text_config["hidden_size"])
+        self.gguf_writer.add_uint32 ("clip.vision.block_count",          vision_config["num_hidden_layers"])
+        self.gguf_writer.add_uint32 ("clip.vision.attention.head_count", vision_config["num_attention_heads"])
+        self.gguf_writer.add_float32("clip.vision.attention.layer_norm_epsilon", vision_config.get("layer_norm_eps", 1e-6))
+        # default values taken from HF tranformers code
+        self.gguf_writer.add_array  ("clip.vision.image_mean", [0.5, 0.5, 0.5])
+        self.gguf_writer.add_array  ("clip.vision.image_std",  [0.5, 0.5, 0.5])
+        self.gguf_writer.add_bool   ("clip.use_gelu", True)
+
+        # load tensors
+        for name, data_torch in self.get_tensors(dir_model):
+            # convert any unsupported data types to float32
+            if data_torch.dtype not in (torch.float16, torch.float32):
+                data_torch = data_torch.to(torch.float32)
+            self.add_tensor(name, data_torch)
+
+    def get_tensors(self, dir_model: Path) -> Iterator[tuple[str, Tensor]]:
+        part_names = Gemma3VisionTower.get_model_part_names(dir_model, "model", ".safetensors")
+        tensor_names_from_parts: set[str] = set()
+        for part_name in part_names:
+            logger.info(f"gguf: loading model part '{part_name}'")
+            from safetensors import safe_open
+            ctx = cast(ContextManager[Any], safe_open(dir_model / part_name, framework="pt", device="cpu"))
+            with ctx as model_part:
+                tensor_names_from_parts.update(model_part.keys())
+
+                for name in model_part.keys():
+                    data = model_part.get_slice(name)
+                    data = LazyTorchTensor.from_safetensors_slice(data)
+                    yield name, data
+
+    def add_tensor(self, name: str, data_torch: Tensor):
+        is_1d = len(data_torch.shape) == 1
+        is_embd = ".embeddings." in name
+        old_dtype = data_torch.dtype
+        can_quantize = not is_1d and not is_embd
+        data_qtype = gguf.GGMLQuantizationType.F32
+
+        # this is to support old checkpoint
+        # TODO: remove this when we have the final model
+        name = name.replace("vision_model.vision_model.", "vision_tower.vision_model.")
+        name = name.replace("multimodal_projector.", "multi_modal_projector.")
+
+        # filter only vision tensors
+        if not name.startswith("vision_tower.vision_model.") and not name.startswith("multi_modal_projector."):
+            return
+        # prefix
+        name = name.replace("vision_tower.vision_model.encoder.layers.", "v.blk.")
+        name = name.replace("vision_tower.vision_model.", "v.")
+        # projector and input embd
+        name = name.replace(".embeddings.patch_embedding.", ".patch_embd.")
+        name = name.replace(".embeddings.position_embedding.", ".position_embd.")
+        name = name.replace(
+            "multi_modal_projector.mm_input_projection_weight",
+            "mm.input_projection.weight"
+        )
+        name = name.replace(
+            "multi_modal_projector.mm_soft_emb_norm.weight",
+            "mm.soft_emb_norm.weight"
+        )
+        name = name.replace("post_layernorm.", "post_ln.")
+        # each block
+        name = name.replace(".self_attn.k_proj.", ".attn_k.")
+        name = name.replace(".self_attn.v_proj.", ".attn_v.")
+        name = name.replace(".self_attn.q_proj.", ".attn_q.")
+        name = name.replace(".self_attn.out_proj.", ".attn_out.")
+        name = name.replace(".layer_norm1.", ".ln1.")
+        name = name.replace(".layer_norm2.", ".ln2.")
+        name = name.replace(".mlp.fc1.", ".ffn_down.")
+        name = name.replace(".mlp.fc2.", ".ffn_up.")
+
+        if can_quantize:
+            if self.ftype == gguf.LlamaFileType.ALL_F32:
+                data_qtype = gguf.GGMLQuantizationType.F32
+            elif self.ftype == gguf.LlamaFileType.MOSTLY_F16:
+                data_qtype = gguf.GGMLQuantizationType.F16
+            elif self.ftype == gguf.LlamaFileType.MOSTLY_BF16:
+                data_qtype = gguf.GGMLQuantizationType.BF16
+            elif self.ftype == gguf.LlamaFileType.MOSTLY_Q8_0:
+                data_qtype = gguf.GGMLQuantizationType.Q8_0
+            else:
+                raise ValueError(f"Unsupported file type: {self.ftype}")
+
+        # corrent norm value ; only this "soft_emb_norm" need to be corrected as it's part of Gemma projector
+        # the other norm values are part of SigLIP model, and they are already correct
+        # ref code: Gemma3RMSNorm
+        if "soft_emb_norm.weight" in name:
+            logger.info(f"Correcting norm value for '{name}'")
+            data_torch = data_torch + 1
+
+        data = data_torch.numpy()
+
+        try:
+            data = gguf.quants.quantize(data, data_qtype)
+        except Exception as e:
+            logger.error(f"Error quantizing tensor '{name}': {e}, fallback to F16")
+            data_qtype = gguf.GGMLQuantizationType.F16
+            data = gguf.quants.quantize(data, data_qtype)
+
+        # reverse shape to make it similar to the internal ggml dimension order
+        shape_str = f"{{{', '.join(str(n) for n in reversed(data_torch.shape))}}}"
+        logger.info(f"{f'%-32s' % f'{name},'} {old_dtype} --> {data_qtype.name}, shape = {shape_str}")
+
+        self.gguf_writer.add_tensor(name, data, raw_dtype=data_qtype)
+
+    def write(self):
+        self.gguf_writer.write_header_to_file(path=self.fname_out)
+        self.gguf_writer.write_kv_data_to_file()
+        self.gguf_writer.write_tensors_to_file(progress=True)
+        self.gguf_writer.close()
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description="Convert Gemma 3 vision tower safetensors to GGUF format",)
+    parser.add_argument(
+        "--outfile", type=Path, default="mmproj.gguf",
+        help="path to write to",
+    )
+    parser.add_argument(
+        "--outtype", type=str, choices=["f32", "f16", "bf16", "q8_0"], default="f16",
+        help="output format",
+    )
+    parser.add_argument(
+        "--bigendian", action="store_true",
+        help="model is executed on big endian machine",
+    )
+    parser.add_argument(
+        "model", type=Path,
+        help="directory containing model file",
+        nargs="?",
+    )
+    parser.add_argument(
+        "--verbose", action="store_true",
+        help="increase output verbosity",
+    )
+
+    args = parser.parse_args()
+    if args.model is None:
+        parser.error("the following arguments are required: model")
+    return args
+
+
+def main() -> None:
+    args = parse_args()
+
+    if args.verbose:
+        logging.basicConfig(level=logging.DEBUG)
+    else:
+        logging.basicConfig(level=logging.INFO)
+
+    dir_model = args.model
+
+    if not dir_model.is_dir():
+        logger.error(f'Error: {args.model} is not a directory')
+        sys.exit(1)
+
+    ftype_map: dict[str, gguf.LlamaFileType] = {
+        "f32": gguf.LlamaFileType.ALL_F32,
+        "f16": gguf.LlamaFileType.MOSTLY_F16,
+        "bf16": gguf.LlamaFileType.MOSTLY_BF16,
+        "q8_0": gguf.LlamaFileType.MOSTLY_Q8_0,
+    }
+
+    logger.info(f"Loading model: {dir_model.name}")
+
+    with torch.inference_mode():
+        gemma3_vision_tower = Gemma3VisionTower(
+            dir_model=dir_model,
+            fname_out=args.outfile,
+            ftype=ftype_map[args.outtype],
+            is_big_endian=args.bigendian,
+        )
+        gemma3_vision_tower.write()
+
+
+if __name__ == '__main__':
+    main()
+

examples/llava/minicpmv-cli.cpp

@@ -86,7 +86,11 @@ static struct clip_ctx * clip_init_context(common_params * params) {
     if (prompt.empty()) {
         prompt = "describe the image in detail.";
     }
-    auto * ctx_clip = clip_model_load(clip_path, /*verbosity=*/ 1);
+    struct clip_context_params clip_params = {
+        /* use_gpu */   params->n_gpu_layers != 0,
+        /* verbosity */ params->verbosity,
+    };
+    auto * ctx_clip = clip_init(clip_path, clip_params);
     return ctx_clip;
 }
 
@@ -148,19 +152,34 @@ static void process_image(struct llava_context * ctx_llava, struct llava_image_e
     process_eval_image_embed(ctx_llava, embeds, params->n_batch, &n_past, idx++);
     eval_string(ctx_llava->ctx_llama, std::string("</image>").c_str(), params->n_batch, &n_past, false);
     if (num_image_embeds > 1) {
-        size_t num_image_embeds_col = clip_uhd_num_image_embeds_col(ctx_llava->ctx_clip);
-        eval_string(ctx_llava->ctx_llama, std::string("<slice>").c_str(), params->n_batch, &n_past, false);
-        for (size_t i = 0; i < (num_image_embeds-1)/num_image_embeds_col; ++i) {
-            for (size_t j = 0; j < num_image_embeds_col; ++j) {
-                eval_string(ctx_llava->ctx_llama, std::string("<image>").c_str(), params->n_batch, &n_past, false);
-                process_eval_image_embed(ctx_llava, embeds, params->n_batch, &n_past, idx++);
-                eval_string(ctx_llava->ctx_llama, std::string("</image>").c_str(), params->n_batch, &n_past, false);
-                if (j == num_image_embeds_col - 1) {
-                    eval_string(ctx_llava->ctx_llama, std::string("\n").c_str(), params->n_batch, &n_past, false);
+        if (has_minicpmv_projector == 2) {
+            size_t num_image_embeds_col = clip_uhd_num_image_embeds_col(ctx_llava->ctx_clip);
+            eval_string(ctx_llava->ctx_llama, std::string("<slice>").c_str(), params->n_batch, &n_past, false);
+            for (size_t i = 0; i < (num_image_embeds-1)/num_image_embeds_col; ++i) {
+                for (size_t j = 0; j < num_image_embeds_col; ++j) {
+                    eval_string(ctx_llava->ctx_llama, std::string("<image>").c_str(), params->n_batch, &n_past, false);
+                    process_eval_image_embed(ctx_llava, embeds, params->n_batch, &n_past, idx++);
+                    eval_string(ctx_llava->ctx_llama, std::string("</image>").c_str(), params->n_batch, &n_past, false);
+                    if (j == num_image_embeds_col - 1) {
+                        eval_string(ctx_llava->ctx_llama, std::string("\n").c_str(), params->n_batch, &n_past, false);
+                    }
+                }
+            }
+            eval_string(ctx_llava->ctx_llama, std::string("</slice>").c_str(), params->n_batch, &n_past, false);
+        }
+        else if (has_minicpmv_projector == 3 || has_minicpmv_projector == 4) {
+            size_t num_image_embeds_col = clip_uhd_num_image_embeds_col(ctx_llava->ctx_clip);
+            for (size_t i = 0; i < (num_image_embeds-1)/num_image_embeds_col; ++i) {
+                for (size_t j = 0; j < num_image_embeds_col; ++j) {
+                    eval_string(ctx_llava->ctx_llama, std::string("<slice>").c_str(), params->n_batch, &n_past, false);
+                    process_eval_image_embed(ctx_llava, embeds, params->n_batch, &n_past, idx++);
+                    eval_string(ctx_llava->ctx_llama, std::string("</slice>").c_str(), params->n_batch, &n_past, false);
+                    if (j == num_image_embeds_col - 1) {
+                        eval_string(ctx_llava->ctx_llama, std::string("\n").c_str(), params->n_batch, &n_past, false);
+                    }
                 }
             }
         }
-        eval_string(ctx_llava->ctx_llama, std::string("</slice>").c_str(), params->n_batch, &n_past, false);
     }
     LOG_INF("%s: image token past: %d\n", __func__, n_past);
 }

examples/llava/minicpmv-convert-image-encoder-to-gguf.py

@@ -597,7 +597,6 @@ elif args.minicpmv_projector is not None:
     fname_middle = "mmproj-"
     has_text_encoder = False
     has_minicpmv_projector = True
-    minicpmv_version = 4
 elif args.vision_only:
     fname_middle = "vision-"
     has_text_encoder = False

examples/lookahead/lookahead.cpp

@@ -7,6 +7,7 @@
 #include <cstdio>
 #include <string>
 #include <vector>
+#include <algorithm>
 
 struct ngram_data {
     bool active = false;
@@ -95,7 +96,7 @@ int main(int argc, char ** argv) {
     llama_decode(ctx, llama_batch_get_one(&inp.back(),           1));
 
     for (int s = 1; s < W + G + 1; ++s) {
-        llama_kv_cache_seq_cp(ctx, 0, s, -1, -1);
+        llama_kv_self_seq_cp(ctx, 0, s, -1, -1);
     }
 
     const auto t_enc_end = ggml_time_us();
@@ -437,17 +438,17 @@ int main(int argc, char ** argv) {
 
         // KV cache management
         // if no verification token matched, we simply remove all cells from this batch -> no fragmentation
-        llama_kv_cache_seq_rm(ctx, -1, n_past, -1);
+        llama_kv_self_seq_rm(ctx, -1, n_past, -1);
 
         if (seq_id_best != 0) {
             // if a verification token matched, we keep the best sequence and remove the rest
             // this leads to some KV cache fragmentation
-            llama_kv_cache_seq_keep(ctx, seq_id_best);
-            llama_kv_cache_seq_cp  (ctx, seq_id_best, 0, -1, -1);
-            llama_kv_cache_seq_rm  (ctx, seq_id_best,    -1, -1);
+            llama_kv_self_seq_keep(ctx, seq_id_best);
+            llama_kv_self_seq_cp  (ctx, seq_id_best, 0, -1, -1);
+            llama_kv_self_seq_rm  (ctx, seq_id_best,    -1, -1);
 
             for (int s = 1; s < W + G + 1; ++s) {
-                llama_kv_cache_seq_cp(ctx, 0, s, -1, -1);
+                llama_kv_self_seq_cp(ctx, 0, s, -1, -1);
             }
         }
     }

examples/lookup/lookup.cpp

@@ -192,7 +192,7 @@ int main(int argc, char ** argv){
 
         // KV cache management
         // clean the cache of draft tokens that weren't accepted
-        llama_kv_cache_seq_rm(ctx, 0, n_past, -1);
+        llama_kv_self_seq_rm(ctx, 0, n_past, -1);
 
         common_batch_clear(batch_tgt);
         common_batch_add(batch_tgt, draft[0], n_past, { 0 }, true);

examples/main/main.cpp

@@ -45,8 +45,8 @@ static void print_usage(int argc, char ** argv) {
     (void) argc;
 
     LOG("\nexample usage:\n");
-    LOG("\n  text generation:     %s -m your_model.gguf -p \"I believe the meaning of life is\" -n 128\n", argv[0]);
-    LOG("\n  chat (conversation): %s -m your_model.gguf -p \"You are a helpful assistant\" -cnv\n", argv[0]);
+    LOG("\n  text generation:     %s -m your_model.gguf -p \"I believe the meaning of life is\" -n 128 -no-cnv\n", argv[0]);
+    LOG("\n  chat (conversation): %s -m your_model.gguf -sys \"You are a helpful assistant\"\n", argv[0]);
     LOG("\n");
 }
 
@@ -217,8 +217,8 @@ int main(int argc, char ** argv) {
     // print chat template example in conversation mode
     if (params.conversation_mode) {
         if (params.enable_chat_template) {
-            if (!params.prompt.empty()) {
-                LOG_WRN("*** User-specified prompt in conversation mode will be ignored, did you mean to set --system-prompt (-sys) instead?\n");
+            if (!params.prompt.empty() && params.system_prompt.empty()) {
+                LOG_WRN("*** User-specified prompt will pre-start conversation, did you mean to set --system-prompt (-sys) instead?\n");
             }
 
             LOG_INF("%s: chat template example:\n%s\n", __func__, common_chat_format_example(chat_templates.get(), params.use_jinja).c_str());
@@ -265,7 +265,7 @@ int main(int argc, char ** argv) {
 
     std::vector<llama_token> embd_inp;
 
-    bool waiting_for_first_input = params.conversation_mode && params.enable_chat_template && params.system_prompt.empty();
+    bool waiting_for_first_input = false;
     auto chat_add_and_format = [&chat_msgs, &chat_templates](const std::string & role, const std::string & content) {
         common_chat_msg new_msg;
         new_msg.role = role;
@@ -276,22 +276,34 @@ int main(int argc, char ** argv) {
         return formatted;
     };
 
+    std::string prompt;
     {
-        std::string prompt;
-
         if (params.conversation_mode && params.enable_chat_template) {
-            // format the system prompt in conversation mode (will use template default if empty)
-            prompt = params.system_prompt;
+            if (!params.system_prompt.empty()) {
+                // format the system prompt (will use template default if empty)
+                chat_add_and_format("system", params.system_prompt);
+            }
 
-            if (!prompt.empty()) {
-                prompt = chat_add_and_format("system", prompt);
+            if (!params.prompt.empty()) {
+                // format and append the user prompt
+                chat_add_and_format("user", params.prompt);
+            } else {
+                waiting_for_first_input = true;
+            }
+
+            if (!params.system_prompt.empty() || !params.prompt.empty()) {
+                common_chat_templates_inputs inputs;
+                inputs.messages = chat_msgs;
+                inputs.add_generation_prompt = !params.prompt.empty();
+
+                prompt = common_chat_templates_apply(chat_templates.get(), inputs).prompt;
             }
         } else {
             // otherwise use the prompt as is
             prompt = params.prompt;
         }
 
-        if (params.interactive_first || !params.prompt.empty() || session_tokens.empty()) {
+        if (params.interactive_first || !prompt.empty() || session_tokens.empty()) {
             LOG_DBG("tokenize the prompt\n");
             embd_inp = common_tokenize(ctx, prompt, true, true);
         } else {
@@ -304,7 +316,7 @@ int main(int argc, char ** argv) {
     }
 
     // Should not run without any tokens
-    if (!params.conversation_mode && embd_inp.empty()) {
+    if (!waiting_for_first_input && embd_inp.empty()) {
         if (add_bos) {
             embd_inp.push_back(llama_vocab_bos(vocab));
             LOG_WRN("embd_inp was considered empty and bos was added: %s\n", string_from(ctx, embd_inp).c_str());
@@ -342,7 +354,7 @@ int main(int argc, char ** argv) {
         }
 
         // remove any "future" tokens that we might have inherited from the previous session
-        llama_kv_cache_seq_rm(ctx, -1, n_matching_session_tokens, -1);
+        llama_kv_self_seq_rm(ctx, -1, n_matching_session_tokens, -1);
     }
 
     LOG_DBG("recalculate the cached logits (check): embd_inp.size() %zu, n_matching_session_tokens %zu, embd_inp.size() %zu, session_tokens.size() %zu\n",
@@ -364,7 +376,12 @@ int main(int argc, char ** argv) {
     }
 
     if (params.conversation_mode) {
-        params.interactive_first = true;
+        if (params.single_turn && !params.prompt.empty()) {
+            params.interactive = false;
+            params.interactive_first = false;
+        } else {
+            params.interactive_first = true;
+        }
     }
 
     // enable interactive mode if interactive start is specified
@@ -585,8 +602,8 @@ int main(int argc, char ** argv) {
                     LOG_DBG("context full, swapping: n_past = %d, n_left = %d, n_ctx = %d, n_keep = %d, n_discard = %d\n",
                             n_past, n_left, n_ctx, params.n_keep, n_discard);
 
-                    llama_kv_cache_seq_rm (ctx, 0, params.n_keep            , params.n_keep + n_discard);
-                    llama_kv_cache_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
+                    llama_kv_self_seq_rm (ctx, 0, params.n_keep            , params.n_keep + n_discard);
+                    llama_kv_self_seq_add(ctx, 0, params.n_keep + n_discard, n_past, -n_discard);
 
                     n_past -= n_discard;
 
@@ -609,9 +626,9 @@ int main(int argc, char ** argv) {
                     LOG_DBG("div:   [%6d, %6d] / %6d -> [%6d, %6d]\n", ga_i + ib*bd, ga_i + ib*bd + ga_w, ga_n, (ga_i + ib*bd)/ga_n, (ga_i + ib*bd + ga_w)/ga_n);
                     LOG_DBG("shift: [%6d, %6d] + %6d -> [%6d, %6d]\n", ga_i + ib*bd + ga_w, n_past + ib*bd, dd, ga_i + ib*bd + ga_w + dd, n_past + ib*bd + dd);
 
-                    llama_kv_cache_seq_add(ctx, 0, ga_i,                n_past,              ib*bd);
-                    llama_kv_cache_seq_div(ctx, 0, ga_i + ib*bd,        ga_i + ib*bd + ga_w, ga_n);
-                    llama_kv_cache_seq_add(ctx, 0, ga_i + ib*bd + ga_w, n_past + ib*bd,      dd);
+                    llama_kv_self_seq_add(ctx, 0, ga_i,                n_past,              ib*bd);
+                    llama_kv_self_seq_div(ctx, 0, ga_i + ib*bd,        ga_i + ib*bd + ga_w, ga_n);
+                    llama_kv_self_seq_add(ctx, 0, ga_i + ib*bd + ga_w, n_past + ib*bd,      dd);
 
                     n_past -= bd;
 
@@ -808,6 +825,11 @@ int main(int argc, char ** argv) {
             if (params.conversation_mode && !waiting_for_first_input) {
                 const auto id = common_sampler_last(smpl);
                 assistant_ss << common_token_to_piece(ctx, id, false);
+
+                if (!prompt.empty()) {
+                    prompt.clear();
+                    is_interacting = false;
+                }
             }
 
             if ((n_past > 0 || waiting_for_first_input) && is_interacting) {
@@ -905,6 +927,11 @@ int main(int argc, char ** argv) {
                     common_sampler_reset(smpl);
                 }
                 is_interacting = false;
+
+                if (waiting_for_first_input && params.single_turn) {
+                    params.interactive = false;
+                    params.interactive_first = false;
+                }
                 waiting_for_first_input = false;
             }
         }

examples/parallel/parallel.cpp

@@ -12,6 +12,7 @@
 #include <string>
 #include <vector>
 #include <ctime>
+#include <algorithm>
 
 // trim whitespace from the beginning and end of a string
 static std::string trim(const std::string & str) {
@@ -201,7 +202,7 @@ int main(int argc, char ** argv) {
 
         // assign the system KV cache to all parallel sequences
         for (int32_t i = 1; i <= n_clients; ++i) {
-            llama_kv_cache_seq_cp(ctx, 0, i, -1, -1);
+            llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
         }
 
         LOG_INF("\n");
@@ -233,9 +234,9 @@ int main(int argc, char ** argv) {
         if (batch.n_tokens == 0) {
             // all sequences have ended - clear the entire KV cache
             for (int i = 1; i <= n_clients; ++i) {
-                llama_kv_cache_seq_rm(ctx, i, -1, -1);
+                llama_kv_self_seq_rm(ctx, i, -1, -1);
                 // but keep the system prompt
-                llama_kv_cache_seq_cp(ctx, 0, i, -1, -1);
+                llama_kv_self_seq_cp(ctx, 0, i, -1, -1);
             }
 
             LOG_INF("%s: clearing the KV cache\n", __func__);
@@ -371,8 +372,8 @@ int main(int argc, char ** argv) {
                     }
 
                     // delete only the generated part of the sequence, i.e. keep the system prompt in the cache
-                    llama_kv_cache_seq_rm(ctx,    client.id + 1, -1, -1);
-                    llama_kv_cache_seq_cp(ctx, 0, client.id + 1, -1, -1);
+                    llama_kv_self_seq_rm(ctx,    client.id + 1, -1, -1);
+                    llama_kv_self_seq_cp(ctx, 0, client.id + 1, -1, -1);
 
                     const auto t_main_end = ggml_time_us();
 

examples/passkey/passkey.cpp

@@ -7,6 +7,7 @@
 #include <cstdio>
 #include <string>
 #include <vector>
+#include <algorithm>
 
 static void print_usage(int, char ** argv) {
     LOG("\nexample usage:\n");
@@ -132,11 +133,11 @@ int main(int argc, char ** argv) {
             const int ib = i/n_batch - 1;
             const int bd = n_batch_grp*(n_grp - 1);
 
-            llama_kv_cache_seq_add (ctx, 0, n_past - n_batch,         n_past,         ib*bd);
-            llama_kv_cache_seq_div (ctx, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
-            llama_kv_cache_update  (ctx);
+            llama_kv_self_seq_add (ctx, 0, n_past - n_batch,         n_past,         ib*bd);
+            llama_kv_self_seq_div (ctx, 0, n_past - n_batch + ib*bd, n_past + ib*bd, n_grp);
+            llama_kv_self_update  (ctx);
 
-            n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1;
+            n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
         }
 
         common_batch_clear(batch);
@@ -166,12 +167,12 @@ int main(int argc, char ** argv) {
 
         LOG_INF("%s: shifting KV cache with %d\n", __func__, n_discard);
 
-        llama_kv_cache_seq_rm (ctx, 0, n_keep            , n_keep + n_discard);
-        llama_kv_cache_seq_add(ctx, 0, n_keep + n_discard, n_ctx,  -n_discard);
-      //llama_kv_cache_defrag (ctx);
-        llama_kv_cache_update (ctx);
+        llama_kv_self_seq_rm (ctx, 0, n_keep            , n_keep + n_discard);
+        llama_kv_self_seq_add(ctx, 0, n_keep + n_discard, n_ctx,  -n_discard);
+      //llama_kv_self_defrag (ctx);
+        llama_kv_self_update (ctx);
 
-        n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1;
+        n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
 
         common_batch_clear(batch);
 
@@ -197,12 +198,12 @@ int main(int argc, char ** argv) {
         if (n_discard > 0) {
             LOG_INF("%s: shifting KV cache with %d to free space for the answer\n", __func__, n_discard);
 
-            llama_kv_cache_seq_rm (ctx, 0, n_keep            , n_keep + n_discard);
-            llama_kv_cache_seq_add(ctx, 0, n_keep + n_discard, n_ctx,  -n_discard);
-          //llama_kv_cache_defrag (ctx);
-            llama_kv_cache_update (ctx);
+            llama_kv_self_seq_rm (ctx, 0, n_keep            , n_keep + n_discard);
+            llama_kv_self_seq_add(ctx, 0, n_keep + n_discard, n_ctx,  -n_discard);
+          //llama_kv_self_defrag (ctx);
+            llama_kv_self_update (ctx);
 
-            n_past = llama_kv_cache_seq_pos_max(ctx, 0) + 1;
+            n_past = llama_kv_self_seq_pos_max(ctx, 0) + 1;
         }
     }
 

examples/perplexity/perplexity.cpp

@@ -361,7 +361,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const common_params
         const auto t_start = std::chrono::high_resolution_clock::now();
 
         // clear the KV cache
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         llama_batch batch = llama_batch_init(n_batch, 0, 1);
 
@@ -547,7 +547,7 @@ static results_perplexity perplexity(llama_context * ctx, const common_params &
         const auto t_start = std::chrono::high_resolution_clock::now();
 
         // clear the KV cache
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         for (int j = 0; j < num_batches; ++j) {
             const int batch_start = start + j * n_batch;
@@ -924,7 +924,7 @@ static void hellaswag_score(llama_context * ctx, const common_params & params) {
             return;
         }
 
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         // decode all tasks [i0, i1)
         if (!decode_helper(ctx, batch, batch_logits, n_batch, n_vocab)) {
@@ -1203,7 +1203,7 @@ static void winogrande_score(llama_context * ctx, const common_params & params)
             return;
         }
 
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         // decode all tasks [i0, i1)
         if (!decode_helper(ctx, batch, batch_logits, n_batch, n_vocab)) {
@@ -1575,7 +1575,7 @@ static void multiple_choice_score(llama_context * ctx, const common_params & par
             return;
         }
 
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         // decode all tasks [i0, i1)
         if (!decode_helper(ctx, batch, batch_logits, n_batch, n_vocab)) {
@@ -1765,7 +1765,7 @@ static void kl_divergence(llama_context * ctx, const common_params & params) {
         }
 
         // clear the KV cache
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
 
         llama_batch batch = llama_batch_init(n_batch, 0, 1);
 

examples/quantize-stats/quantize-stats.cpp

@@ -1,6 +1,6 @@
 #include "ggml.h"
 #include "llama.h"
-#include "llama-context.h"
+#include "llama-model.h"
 #include "common.h"
 
 #include <algorithm>
@@ -328,7 +328,7 @@ int main(int argc, char ** argv) {
         }
     }
 
-    const auto & tensors = llama_internal_get_tensor_map(ctx);
+    const auto & tensors = llama_internal_get_tensor_map(model);
 
     // check layer tensors
     int included_layers = 0;

examples/quantize/quantize.cpp

@@ -8,6 +8,7 @@
 #include <unordered_map>
 #include <fstream>
 #include <cmath>
+#include <cctype>
 
 struct quant_option {
     std::string name;

examples/retrieval/retrieval.cpp

@@ -83,7 +83,7 @@ static void batch_add_seq(llama_batch & batch, const std::vector<int32_t> & toke
 
 static void batch_decode(llama_context * ctx, llama_batch & batch, float * output, int n_seq, int n_embd) {
     // clear previous kv_cache values (irrelevant for embeddings)
-    llama_kv_cache_clear(ctx);
+    llama_kv_self_clear(ctx);
 
     // run model
     LOG_INF("%s: n_tokens = %d, n_seq = %d\n", __func__, batch.n_tokens, n_seq);

examples/run/run.cpp

@@ -79,6 +79,7 @@ class Opt {
         ctx_params           = llama_context_default_params();
         model_params         = llama_model_default_params();
         context_size_default = ctx_params.n_batch;
+        n_threads_default    = ctx_params.n_threads;
         ngl_default          = model_params.n_gpu_layers;
         common_params_sampling sampling;
         temperature_default = sampling.temp;
@@ -104,6 +105,7 @@ class Opt {
 
         ctx_params.n_batch        = context_size >= 0 ? context_size : context_size_default;
         ctx_params.n_ctx          = ctx_params.n_batch;
+        ctx_params.n_threads = ctx_params.n_threads_batch = n_threads >= 0 ? n_threads : n_threads_default;
         model_params.n_gpu_layers = ngl >= 0 ? ngl : ngl_default;
         temperature               = temperature >= 0 ? temperature : temperature_default;
 
@@ -116,12 +118,12 @@ class Opt {
     std::string chat_template_file;
     std::string          user;
     bool                 use_jinja   = false;
-    int                  context_size = -1, ngl = -1;
+    int                  context_size = -1, ngl = -1, n_threads = -1;
     float                temperature = -1;
     bool                 verbose     = false;
 
   private:
-    int   context_size_default = -1, ngl_default = -1;
+    int   context_size_default = -1, ngl_default = -1, n_threads_default = -1;
     float temperature_default = -1;
     bool  help                = false;
 
@@ -159,53 +161,94 @@ class Opt {
         return 0;
     }
 
+    int parse_options_with_value(int argc, const char ** argv, int & i, bool & options_parsing) {
+        if (options_parsing && (strcmp(argv[i], "-c") == 0 || strcmp(argv[i], "--context-size") == 0)) {
+            if (handle_option_with_value(argc, argv, i, context_size) == 1) {
+                return 1;
+            }
+        } else if (options_parsing &&
+                   (strcmp(argv[i], "-n") == 0 || strcmp(argv[i], "-ngl") == 0 || strcmp(argv[i], "--ngl") == 0)) {
+            if (handle_option_with_value(argc, argv, i, ngl) == 1) {
+                return 1;
+            }
+        } else if (options_parsing && (strcmp(argv[i], "-t") == 0 || strcmp(argv[i], "--threads") == 0)) {
+            if (handle_option_with_value(argc, argv, i, n_threads) == 1) {
+                return 1;
+            }
+        } else if (options_parsing && strcmp(argv[i], "--temp") == 0) {
+            if (handle_option_with_value(argc, argv, i, temperature) == 1) {
+                return 1;
+            }
+        } else if (options_parsing && strcmp(argv[i], "--chat-template-file") == 0) {
+            if (handle_option_with_value(argc, argv, i, chat_template_file) == 1) {
+                return 1;
+            }
+            use_jinja = true;
+        } else {
+            return 2;
+        }
+
+        return 0;
+    }
+
+    int parse_options(const char ** argv, int & i, bool & options_parsing) {
+        if (options_parsing && (parse_flag(argv, i, "-v", "--verbose") || parse_flag(argv, i, "-v", "--log-verbose"))) {
+            verbose = true;
+        } else if (options_parsing && strcmp(argv[i], "--jinja") == 0) {
+            use_jinja = true;
+        } else if (options_parsing && parse_flag(argv, i, "-h", "--help")) {
+            help = true;
+            return 0;
+        } else if (options_parsing && strcmp(argv[i], "--") == 0) {
+            options_parsing = false;
+        } else {
+            return 2;
+        }
+
+        return 0;
+    }
+
+    int parse_positional_args(const char ** argv, int & i, int & positional_args_i) {
+        if (positional_args_i == 0) {
+            if (!argv[i][0] || argv[i][0] == '-') {
+                return 1;
+            }
+
+            ++positional_args_i;
+            model_ = argv[i];
+        } else if (positional_args_i == 1) {
+            ++positional_args_i;
+            user = argv[i];
+        } else {
+            user += " " + std::string(argv[i]);
+        }
+
+        return 0;
+    }
+
     int parse(int argc, const char ** argv) {
         bool options_parsing   = true;
         for (int i = 1, positional_args_i = 0; i < argc; ++i) {
-            if (options_parsing && (strcmp(argv[i], "-c") == 0 || strcmp(argv[i], "--context-size") == 0)) {
-                if (handle_option_with_value(argc, argv, i, context_size) == 1) {
-                    return 1;
-                }
-            } else if (options_parsing &&
-                       (strcmp(argv[i], "-n") == 0 || strcmp(argv[i], "-ngl") == 0 || strcmp(argv[i], "--ngl") == 0)) {
-                if (handle_option_with_value(argc, argv, i, ngl) == 1) {
-                    return 1;
-                }
-            } else if (options_parsing && strcmp(argv[i], "--temp") == 0) {
-                if (handle_option_with_value(argc, argv, i, temperature) == 1) {
-                    return 1;
-                }
-            } else if (options_parsing &&
-                       (parse_flag(argv, i, "-v", "--verbose") || parse_flag(argv, i, "-v", "--log-verbose"))) {
-                verbose = true;
-            } else if (options_parsing && strcmp(argv[i], "--jinja") == 0) {
-                use_jinja = true;
-            } else if (options_parsing && strcmp(argv[i], "--chat-template-file") == 0){
-                if (handle_option_with_value(argc, argv, i, chat_template_file) == 1) {
-                    return 1;
-                }
-                use_jinja = true;
-            } else if (options_parsing && parse_flag(argv, i, "-h", "--help")) {
-                help = true;
-                return 0;
-            } else if (options_parsing && strcmp(argv[i], "--") == 0) {
-                options_parsing = false;
-            } else if (positional_args_i == 0) {
-                if (!argv[i][0] || argv[i][0] == '-') {
-                    return 1;
-                }
+            int ret = parse_options_with_value(argc, argv, i, options_parsing);
+            if (ret == 0) {
+                continue;
+            } else if (ret == 1) {
+                return ret;
+            }
 
-                ++positional_args_i;
-                model_ = argv[i];
-            } else if (positional_args_i == 1) {
-                ++positional_args_i;
-                user = argv[i];
-            } else {
-                user += " " + std::string(argv[i]);
+            ret = parse_options(argv, i, options_parsing);
+            if (ret == 0) {
+                continue;
+            } else if (ret == 1) {
+                return ret;
+            }
+
+            if (parse_positional_args(argv, i, positional_args_i)) {
+                return 1;
             }
         }
 
-        if (model_.empty()){
+        if (model_.empty()) {
             return 1;
         }
 
@@ -232,6 +275,8 @@ class Opt {
             "      Number of GPU layers (default: %d)\n"
             "  --temp <value>\n"
             "      Temperature (default: %.1f)\n"
+            "  -t, --threads <value>\n"
+            "      Number of threads to use during generation (default: %d)\n"
             "  -v, --verbose, --log-verbose\n"
             "      Set verbosity level to infinity (i.e. log all messages, useful for debugging)\n"
             "  -h, --help\n"
@@ -260,7 +305,7 @@ class Opt {
             "  llama-run file://some-file3.gguf\n"
             "  llama-run --ngl 999 some-file4.gguf\n"
             "  llama-run --ngl 999 some-file5.gguf Hello World\n",
-            context_size_default, ngl_default, temperature_default);
+            context_size_default, ngl_default, temperature_default, n_threads_default);
     }
 };
 
@@ -891,7 +936,7 @@ static int apply_chat_template(const struct common_chat_templates * tmpls, Llama
 // Function to tokenize the prompt
 static int tokenize_prompt(const llama_vocab * vocab, const std::string & prompt,
                            std::vector<llama_token> & prompt_tokens, const LlamaData & llama_data) {
-    const bool is_first = llama_get_kv_cache_used_cells(llama_data.context.get()) == 0;
+    const bool is_first = llama_kv_self_used_cells(llama_data.context.get()) == 0;
 
     const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true);
     prompt_tokens.resize(n_prompt_tokens);
@@ -907,7 +952,7 @@ static int tokenize_prompt(const llama_vocab * vocab, const std::string & prompt
 // Check if we have enough space in the context to evaluate this batch
 static int check_context_size(const llama_context_ptr & ctx, const llama_batch & batch) {
     const int n_ctx      = llama_n_ctx(ctx.get());
-    const int n_ctx_used = llama_get_kv_cache_used_cells(ctx.get());
+    const int n_ctx_used = llama_kv_self_used_cells(ctx.get());
     if (n_ctx_used + batch.n_tokens > n_ctx) {
         printf(LOG_COL_DEFAULT "\n");
         printe("context size exceeded\n");

examples/save-load-state/save-load-state.cpp

@@ -15,7 +15,7 @@ int main(int argc, char ** argv) {
         return 1;
     }
 
-    print_build_info();
+    common_init();
 
     if (params.n_predict < 0) {
         params.n_predict = 16;
@@ -196,7 +196,7 @@ int main(int argc, char ** argv) {
         fprintf(stderr, "%s : seq 0 copied, %zd bytes\n", __func__, ncopy);
 
         // erase whole kv
-        llama_kv_cache_clear(ctx3);
+        llama_kv_self_clear(ctx3);
         fprintf(stderr, "%s : kv cache cleared\n", __func__);
 
         // restore kv into seq 1

examples/server/public_legacy/json-schema-to-grammar.mjs

@@ -1,5 +1,5 @@
 // WARNING: This file was ported from json_schema_to_grammar.py, please fix bugs / add features there first.
-const SPACE_RULE = '| " " | "\\n" [ \\t]{0,20}';
+const SPACE_RULE = '| " " | "\\n"{1,2} [ \\t]{0,20}';
 
 function _buildRepetition(itemRule, minItems, maxItems, opts={}) {
   if (minItems === 0 && maxItems === 1) {

examples/server/server.cpp

@@ -131,9 +131,9 @@ struct slot_params {
             lora.push_back({{"id", i}, {"scale", this->lora[i].scale}});
         }
 
-        std::vector<std::string> grammar_trigger_words;
-        for (const auto & trigger : sampling.grammar_trigger_words) {
-            grammar_trigger_words.push_back(trigger.word);
+        auto grammar_triggers = json::array();
+        for (const auto & trigger : sampling.grammar_triggers) {
+            grammar_triggers.push_back(trigger.to_json<json>());
         }
 
         return json {
@@ -170,8 +170,8 @@ struct slot_params {
             {"n_probs",                   sampling.n_probs},
             {"min_keep",                  sampling.min_keep},
             {"grammar",                   sampling.grammar},
-            {"grammar_trigger_words",     grammar_trigger_words},
-            {"grammar_trigger_tokens",    sampling.grammar_trigger_tokens},
+            {"grammar_lazy",              sampling.grammar_lazy},
+            {"grammar_triggers",          grammar_triggers},
             {"preserved_tokens",          sampling.preserved_tokens},
             {"chat_format",               common_chat_format_name(oaicompat_chat_format)},
             {"samplers",                  samplers},
@@ -356,24 +356,6 @@ struct server_task {
         }
 
         {
-            const auto grammar_triggers = data.find("grammar_triggers");
-            if (grammar_triggers != data.end()) {
-                for (const auto & t : *grammar_triggers) {
-                    common_grammar_trigger trigger;
-                    trigger.word = t.at("word");
-                    trigger.at_start = t.at("at_start");
-
-                    auto ids = common_tokenize(vocab, trigger.word, /* add_special= */ false, /* parse_special= */ true);
-                    if (ids.size() == 1) {
-                        SRV_DBG("Grammar trigger token: %d (`%s`)\n", ids[0], trigger.word.c_str());
-                        params.sampling.grammar_trigger_tokens.push_back(ids[0]);
-                        params.sampling.preserved_tokens.insert(ids[0]);
-                        continue;
-                    }
-                    SRV_DBG("Grammar trigger word: `%s`\n", trigger.word.c_str());
-                    params.sampling.grammar_trigger_words.push_back(trigger);
-                }
-            }
             const auto preserved_tokens = data.find("preserved_tokens");
             if (preserved_tokens != data.end()) {
                 for (const auto & t : *preserved_tokens) {
@@ -383,12 +365,39 @@ struct server_task {
                         params.sampling.preserved_tokens.insert(ids[0]);
                     } else {
                         // This may happen when using a tool call style meant for a model with special tokens to preserve on a model without said tokens.
-                        SRV_WRN("Not preserved because more than 1 token (wrong chat template override?): %s\n", t.get<std::string>().c_str());
+                        SRV_DBG("Not preserved because more than 1 token: %s\n", t.get<std::string>().c_str());
                     }
                 }
             }
-            if (params.sampling.grammar_lazy) {
-                GGML_ASSERT(params.sampling.grammar_trigger_tokens.size() > 0 || params.sampling.grammar_trigger_words.size() > 0);
+            const auto grammar_triggers = data.find("grammar_triggers");
+            if (grammar_triggers != data.end()) {
+                for (const auto & t : *grammar_triggers) {
+                    auto ct = common_grammar_trigger::from_json(t);
+                    if (ct.type == COMMON_GRAMMAR_TRIGGER_TYPE_WORD) {
+                        const auto & word = ct.value;
+                        auto ids = common_tokenize(vocab, word, /* add_special= */ false, /* parse_special= */ true);
+                        if (ids.size() == 1) {
+                            auto token = ids[0];
+                            if (std::find(params.sampling.preserved_tokens.begin(), params.sampling.preserved_tokens.end(), (llama_token) token) == params.sampling.preserved_tokens.end()) {
+                                throw std::runtime_error("Grammar trigger word should be marked as preserved token: " + word);
+                            }
+                            SRV_DBG("Grammar trigger token: %d (`%s`)\n", token, word.c_str());
+                            common_grammar_trigger trigger;
+                            trigger.type = COMMON_GRAMMAR_TRIGGER_TYPE_TOKEN;
+                            trigger.value = word;
+                            trigger.token = token;
+                            params.sampling.grammar_triggers.push_back(std::move(trigger));
+                        } else {
+                            SRV_DBG("Grammar trigger word: `%s`\n", word.c_str());
+                            params.sampling.grammar_triggers.push_back({COMMON_GRAMMAR_TRIGGER_TYPE_WORD, word});
+                        }
+                    } else {
+                        params.sampling.grammar_triggers.push_back(ct);
+                    }
+                }
+            }
+            if (params.sampling.grammar_lazy && params.sampling.grammar_triggers.empty()) {
+                throw std::runtime_error("Error: no triggers set for lazy grammar!");
             }
         }
 
@@ -742,7 +751,10 @@ struct server_task_result_cmpl_final : server_task_result {
                         {"name", tc.name},
                         {"arguments", tc.arguments},
                     }},
-                    {"id", tc.id},
+                    // Some templates generate and require an id (sometimes in a very specific format, e.g. Mistral Nemo).
+                    // We only generate a random id for the ones that don't generate one by themselves
+                    // (they also won't get to see it as their template likely doesn't use it, so it's all for the client)
+                    {"id", tc.id.empty() ? gen_tool_call_id() : tc.id},
                 });
             }
             message["tool_calls"] = tool_calls;
@@ -1304,7 +1316,7 @@ struct server_slot {
         return task_type == SERVER_TASK_TYPE_EMBEDDING || task_type == SERVER_TASK_TYPE_RERANK;
     }
 
-    bool can_batch_with(server_slot & other_slot) {
+    bool can_batch_with(server_slot & other_slot) const {
         return is_non_causal() == other_slot.is_non_causal()
             && are_lora_equal(lora, other_slot.lora);
     }
@@ -1892,6 +1904,7 @@ struct server_context {
         try {
             common_chat_format_example(chat_templates.get(), params.use_jinja);
         } catch (const std::exception & e) {
+            SRV_WRN("%s: Chat template parsing error: %s\n", __func__, e.what());
             SRV_WRN("%s: The chat template that comes with this model is not yet supported, falling back to chatml. This may cause the model to output suboptimal responses\n", __func__);
             chat_templates = common_chat_templates_init(model, "chatml");
         }
@@ -2027,6 +2040,18 @@ struct server_context {
         return ret;
     }
 
+    bool can_be_detokenized(const struct llama_context * ctx, const std::vector<llama_token> & tokens) {
+        const llama_model * model = llama_get_model(ctx);
+        const llama_vocab * vocab = llama_model_get_vocab(model);
+        const int32_t n_vocab = llama_vocab_n_tokens(vocab);
+        for (const auto & token : tokens) {
+            if (token < 0 || token >= n_vocab) {
+                return false;
+            }
+        }
+        return true;
+    }
+
     bool launch_slot_with_task(server_slot & slot, const server_task & task) {
         slot.reset();
         slot.id_task       = task.id;
@@ -2041,11 +2066,16 @@ struct server_context {
             slot.lora = task.params.lora;
         }
 
+        bool can_detokenize = can_be_detokenized(ctx, slot.prompt_tokens);
+        if (!can_detokenize) {
+            send_error(task, "Prompt contains invalid tokens", ERROR_TYPE_INVALID_REQUEST);
+            return false;
+        }
         SLT_DBG(slot, "launching slot : %s\n", safe_json_to_str(slot.to_json()).c_str());
 
         if (slot.n_predict > 0 && slot.params.n_predict > slot.n_predict) {
             // Might be better to reject the request with a 400 ?
-            SLT_WRN(slot, "n_predict = %d exceeds server configuration, setting to %d", slot.params.n_predict, slot.n_predict);
+            SLT_WRN(slot, "n_predict = %d exceeds server configuration, setting to %d\n", slot.params.n_predict, slot.n_predict);
             slot.params.n_predict = slot.n_predict;
         }
 
@@ -2083,7 +2113,7 @@ struct server_context {
         SRV_DBG("%s", "clearing KV cache\n");
 
         // clear the entire KV cache
-        llama_kv_cache_clear(ctx);
+        llama_kv_self_clear(ctx);
         clean_kv_cache = false;
     }
 
@@ -2148,14 +2178,6 @@ struct server_context {
         }
 
         if (slot.has_new_line) {
-            // if we have already seen a new line, we stop after a certain time limit
-            if (slot.params.t_max_predict_ms > 0 && (ggml_time_us() - slot.t_start_generation > 1000.0f*slot.params.t_max_predict_ms)) {
-                slot.stop           = STOP_TYPE_LIMIT;
-                slot.has_next_token = false;
-
-                SLT_DBG(slot, "stopped by time limit, n_decoded = %d, t_max_predict_ms = %d ms\n", slot.n_decoded, (int) slot.params.t_max_predict_ms);
-            }
-
             // require that each new line has a whitespace prefix (i.e. indentation) of at least slot.params.n_indent
             if (slot.params.n_indent > 0) {
                 // check the current indentation
@@ -2194,6 +2216,14 @@ struct server_context {
         // check if there is a new line in the generated text
         if (result.text_to_send.find('\n') != std::string::npos) {
             slot.has_new_line = true;
+
+            // if we have seen a new line, we stop after a certain time limit, but only upon another new line
+            if (slot.params.t_max_predict_ms > 0 && (ggml_time_us() - slot.t_start_generation > 1000.0f*slot.params.t_max_predict_ms)) {
+                slot.stop           = STOP_TYPE_LIMIT;
+                slot.has_next_token = false;
+
+                SLT_DBG(slot, "stopped by time limit, n_decoded = %d, t_max_predict_ms = %d ms\n", slot.n_decoded, (int) slot.params.t_max_predict_ms);
+            }
         }
 
         // if context shift is disabled, we stop when it reaches the context limit
@@ -2625,8 +2655,8 @@ struct server_context {
                     res->n_tasks_deferred    = queue_tasks.queue_tasks_deferred.size();
                     res->t_start             = metrics.t_start;
 
-                    res->kv_cache_tokens_count = llama_get_kv_cache_token_count(ctx);
-                    res->kv_cache_used_cells   = llama_get_kv_cache_used_cells(ctx);
+                    res->kv_cache_tokens_count = llama_kv_self_n_tokens(ctx);
+                    res->kv_cache_used_cells   = llama_kv_self_used_cells(ctx);
 
                     res->n_prompt_tokens_processed_total = metrics.n_prompt_tokens_processed_total;
                     res->t_prompt_processing_total       = metrics.t_prompt_processing_total;
@@ -2742,7 +2772,7 @@ struct server_context {
 
                     // Erase token cache
                     const size_t n_erased = slot->cache_tokens.size();
-                    llama_kv_cache_seq_rm(ctx, slot->id, -1, -1);
+                    llama_kv_self_seq_rm(ctx, slot->id, -1, -1);
                     slot->cache_tokens.clear();
 
                     auto res = std::make_unique<server_task_result_slot_erase>();
@@ -2810,8 +2840,8 @@ struct server_context {
 
                 SLT_WRN(slot, "slot context shift, n_keep = %d, n_left = %d, n_discard = %d\n", n_keep, n_left, n_discard);
 
-                llama_kv_cache_seq_rm (ctx, slot.id, n_keep            , n_keep + n_discard);
-                llama_kv_cache_seq_add(ctx, slot.id, n_keep + n_discard, slot.n_past,        -n_discard);
+                llama_kv_self_seq_rm (ctx, slot.id, n_keep            , n_keep + n_discard);
+                llama_kv_self_seq_add(ctx, slot.id, n_keep + n_discard, slot.n_past,        -n_discard);
 
                 if (slot.params.cache_prompt) {
                     for (size_t i = n_keep + n_discard; i < slot.cache_tokens.size(); i++) {
@@ -3002,8 +3032,8 @@ struct server_context {
 
                                             const int64_t kv_shift = (int64_t) head_p - (int64_t) head_c;
 
-                                            llama_kv_cache_seq_rm (ctx, slot.id, head_p, head_c);
-                                            llama_kv_cache_seq_add(ctx, slot.id, head_c, -1,     kv_shift);
+                                            llama_kv_self_seq_rm (ctx, slot.id, head_p, head_c);
+                                            llama_kv_self_seq_add(ctx, slot.id, head_c, head_c + n_match, kv_shift);
 
                                             for (size_t i = 0; i < n_match; i++) {
                                                 slot.cache_tokens[head_p + i] = slot.cache_tokens[head_c + i];
@@ -3041,9 +3071,9 @@ struct server_context {
                     }
 
                     // keep only the common part
-                    if (!llama_kv_cache_seq_rm(ctx, slot.id, slot.n_past, -1)) {
+                    if (!llama_kv_self_seq_rm(ctx, slot.id, slot.n_past, -1)) {
                         // could not partially delete (likely using a non-Transformer model)
-                        llama_kv_cache_seq_rm(ctx, slot.id, -1, -1);
+                        llama_kv_self_seq_rm(ctx, slot.id, -1, -1);
 
                         // there is no common part left
                         slot.n_past = 0;
@@ -3283,7 +3313,7 @@ struct server_context {
                 slot.cache_tokens.push_back(id);
                 slot.cache_tokens.insert(slot.cache_tokens.end(), ids.begin(), ids.end() - 1);
 
-                llama_kv_cache_seq_rm(ctx, slot.id, slot.n_past, -1);
+                llama_kv_self_seq_rm(ctx, slot.id, slot.n_past, -1);
 
                 for (size_t i = 0; i < ids.size(); ++i) {
                     completion_token_output result;

examples/server/tests/unit/test_chat_completion.py

@@ -144,6 +144,7 @@ def test_apply_chat_template():
 @pytest.mark.parametrize("response_format,n_predicted,re_content", [
     ({"type": "json_object", "schema": {"const": "42"}}, 6, "\"42\""),
     ({"type": "json_object", "schema": {"items": [{"type": "integer"}]}}, 10, "[ -3000 ]"),
+    ({"type": "json_schema", "json_schema": {"schema": {"const": "foooooo"}}}, 10, "\"foooooo\""),
     ({"type": "json_object"}, 10, "(\\{|John)+"),
     ({"type": "sound"}, 0, None),
     # invalid response format (expected to fail)

examples/server/tests/unit/test_tool_call.py

@@ -1,4 +1,12 @@
+#!/usr/bin/env python
 import pytest
+
+# ensure grandparent path is in sys.path
+from pathlib import Path
+import sys
+path = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(path))
+
 from utils import *
 
 server: ServerProcess
@@ -66,15 +74,8 @@ WEATHER_TOOL = {
 }
 
 
-def do_test_completion_with_required_tool_tiny(template_name: str, tool: dict, argument_key: str | None):
-    global server
-    n_predict = 512
-    # server = ServerPreset.stories15m_moe()
-    server.jinja = True
-    server.n_predict = n_predict
-    server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
-    server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
+def do_test_completion_with_required_tool_tiny(server: ServerProcess, tool: dict, argument_key: str | None, n_predict, **kwargs):
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "max_tokens": n_predict,
         "messages": [
             {"role": "system", "content": "You are a coding assistant."},
@@ -83,16 +84,15 @@ def do_test_completion_with_required_tool_tiny(template_name: str, tool: dict, a
         "tool_choice": "required",
         "tools": [tool],
         "parallel_tool_calls": False,
-        "temperature": 0.0,
-        "top_k": 1,
-        "top_p": 1.0,
+        **kwargs,
     })
     assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
     choice = res.body["choices"][0]
     tool_calls = choice["message"].get("tool_calls")
     assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
     tool_call = tool_calls[0]
-    assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
+    assert choice["message"].get("content") in (None, ""), f'Expected no content in {choice["message"]}'
+    assert len(tool_call.get("id", "")) > 0, f'Expected non empty tool call id in {tool_call}'
     expected_function_name = "python" if tool["type"] == "code_interpreter" else tool["function"]["name"]
     assert expected_function_name == tool_call["function"]["name"]
     actual_arguments = tool_call["function"]["arguments"]
@@ -108,7 +108,14 @@ def do_test_completion_with_required_tool_tiny(template_name: str, tool: dict, a
     ("meta-llama-Llama-3.3-70B-Instruct",             PYTHON_TOOL,          "code"),
 ])
 def test_completion_with_required_tool_tiny_fast(template_name: str, tool: dict, argument_key: str | None):
-    do_test_completion_with_required_tool_tiny(template_name, tool, argument_key)
+    global server
+    n_predict = 512
+    # server = ServerPreset.stories15m_moe()
+    server.jinja = True
+    server.n_predict = n_predict
+    server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
+    server.start(timeout_seconds=TIMEOUT_SERVER_START)
+    do_test_completion_with_required_tool_tiny(server, tool, argument_key, n_predict, temperature=0.0, top_k=1, top_p=1.0)
 
 
 @pytest.mark.slow
@@ -130,10 +137,17 @@ def test_completion_with_required_tool_tiny_fast(template_name: str, tool: dict,
     ("deepseek-ai-DeepSeek-R1-Distill-Llama-8B",      TEST_TOOL,            "success"),
     ("deepseek-ai-DeepSeek-R1-Distill-Llama-8B",      PYTHON_TOOL,          "code"),
     ("fireworks-ai-llama-3-firefunction-v2",          TEST_TOOL,            "success"),
-    ("fireworks-ai-llama-3-firefunction-v2",          PYTHON_TOOL,          "code"),
+    # ("fireworks-ai-llama-3-firefunction-v2",          PYTHON_TOOL,          "code"),
 ])
 def test_completion_with_required_tool_tiny_slow(template_name: str, tool: dict, argument_key: str | None):
-    do_test_completion_with_required_tool_tiny(template_name, tool, argument_key)
+    global server
+    n_predict = 512
+    # server = ServerPreset.stories15m_moe()
+    server.jinja = True
+    server.n_predict = n_predict
+    server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
+    server.start(timeout_seconds=TIMEOUT_SERVER_START)
+    do_test_completion_with_required_tool_tiny(server, tool, argument_key, n_predict)
 
 
 @pytest.mark.slow
@@ -142,25 +156,33 @@ def test_completion_with_required_tool_tiny_slow(template_name: str, tool: dict,
     (PYTHON_TOOL,  "code",     "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
     (PYTHON_TOOL,  "code",     "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", "chatml"),
 
-    # Note: gemma-2-2b-it knows itself as "model", not "assistant", so we don't test the ill-suited chatml on it.
     (TEST_TOOL,    "success",  "bartowski/gemma-2-2b-it-GGUF:Q4_K_M",              None),
     (PYTHON_TOOL,  "code",     "bartowski/gemma-2-2b-it-GGUF:Q4_K_M",              None),
+    (PYTHON_TOOL,  "code",     "bartowski/gemma-2-2b-it-GGUF:Q4_K_M",              "chatml"),
 
     (TEST_TOOL,    "success",  "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      None),
     (PYTHON_TOOL,  "code",     "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      None),
     (PYTHON_TOOL,  "code",     "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      "chatml"),
 
+    (TEST_TOOL,    "success",  "bartowski/Qwen2.5-1.5B-Instruct-GGUF:Q4_K_M",      None),
+    (PYTHON_TOOL,  "code",     "bartowski/Qwen2.5-1.5B-Instruct-GGUF:Q4_K_M",      None),
+    (PYTHON_TOOL,  "code",     "bartowski/Qwen2.5-1.5B-Instruct-GGUF:Q4_K_M",      "chatml"),
+
+    (TEST_TOOL,    "success",  "bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M",      None),
+    (PYTHON_TOOL,  "code",     "bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M",      None),
+    (PYTHON_TOOL,  "code",     "bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M",      "chatml"),
+
     (TEST_TOOL,    "success",  "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        None),
     (PYTHON_TOOL,  "code",     "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        None),
     (PYTHON_TOOL,  "code",     "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        "chatml"),
 
     (TEST_TOOL,    "success",  "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
     (PYTHON_TOOL,  "code",     "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
-    # (PYTHON_TOOL,  "code",     "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", "chatml"),
+    (PYTHON_TOOL,  "code",     "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M", "chatml"),
 
     (TEST_TOOL,    "success",  "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",   ("NousResearch/Hermes-3-Llama-3.1-8B", "tool_use")),
     (PYTHON_TOOL,  "code",     "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",   ("NousResearch/Hermes-3-Llama-3.1-8B", "tool_use")),
-    # (PYTHON_TOOL,  "code",     "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",   "chatml"),
+    (PYTHON_TOOL,  "code",     "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",   "chatml"),
 
     (TEST_TOOL,    "success",  "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
     (PYTHON_TOOL,  "code",     "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
@@ -176,10 +198,10 @@ def test_completion_with_required_tool_tiny_slow(template_name: str, tool: dict,
 
     (TEST_TOOL,    "success",  "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      ("meta-llama/Llama-3.2-3B-Instruct", None)),
     (PYTHON_TOOL,  "code",     "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      ("meta-llama/Llama-3.2-3B-Instruct", None)),
-    # (PYTHON_TOOL,  "code",     "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      "chatml"),
-    # TODO: fix these
-    # (TEST_TOOL,    "success",  "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
-    # (PYTHON_TOOL,  "code",     "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
+    (PYTHON_TOOL,  "code",     "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      "chatml"),
+
+    (TEST_TOOL,    "success",  "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
+    (PYTHON_TOOL,  "code",     "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
 ])
 def test_completion_with_required_tool_real_model(tool: dict, argument_key: str | None, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
     global server
@@ -197,7 +219,7 @@ def test_completion_with_required_tool_real_model(tool: dict, argument_key: str
     elif isinstance(template_override, str):
         server.chat_template = template_override
     server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "max_tokens": n_predict,
         "messages": [
             {"role": "system", "content": "You are a coding assistant."},
@@ -215,7 +237,7 @@ def test_completion_with_required_tool_real_model(tool: dict, argument_key: str
     tool_calls = choice["message"].get("tool_calls")
     assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
     tool_call = tool_calls[0]
-    assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
+    # assert choice["message"].get("content") in (None, ""), f'Expected no content in {choice["message"]}'
     expected_function_name = "python" if tool["type"] == "code_interpreter" else tool["function"]["name"]
     assert expected_function_name == tool_call["function"]["name"]
     actual_arguments = tool_call["function"]["arguments"]
@@ -225,13 +247,8 @@ def test_completion_with_required_tool_real_model(tool: dict, argument_key: str
         assert argument_key in actual_arguments, f"tool arguments: {json.dumps(actual_arguments)}, expected: {argument_key}"
 
 
-def do_test_completion_without_tool_call(template_name: str, n_predict: int, tools: list[dict], tool_choice: str | None):
-    global server
-    server.jinja = True
-    server.n_predict = n_predict
-    server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
-    server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
+def do_test_completion_without_tool_call(server: ServerProcess, n_predict: int, tools: list[dict], tool_choice: str | None, **kwargs):
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "max_tokens": n_predict,
         "messages": [
             {"role": "system", "content": "You are a coding assistant."},
@@ -239,9 +256,7 @@ def do_test_completion_without_tool_call(template_name: str, n_predict: int, too
         ],
         "tools": tools if tools else None,
         "tool_choice": tool_choice,
-        "temperature": 0.0,
-        "top_k": 1,
-        "top_p": 1.0,
+        **kwargs,
     }, timeout=TIMEOUT_HTTP_REQUEST)
     assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
     choice = res.body["choices"][0]
@@ -254,7 +269,12 @@ def do_test_completion_without_tool_call(template_name: str, n_predict: int, too
     ("meta-llama-Llama-3.3-70B-Instruct",         128, [PYTHON_TOOL], 'none'),
 ])
 def test_completion_without_tool_call_fast(template_name: str, n_predict: int, tools: list[dict], tool_choice: str | None):
-    do_test_completion_without_tool_call(template_name, n_predict, tools, tool_choice)
+    global server
+    server.jinja = True
+    server.n_predict = n_predict
+    server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
+    server.start(timeout_seconds=TIMEOUT_SERVER_START)
+    do_test_completion_without_tool_call(server, n_predict, tools, tool_choice)
 
 
 @pytest.mark.slow
@@ -270,7 +290,12 @@ def test_completion_without_tool_call_fast(template_name: str, n_predict: int, t
     ("meta-llama-Llama-3.2-3B-Instruct",              256, [PYTHON_TOOL], 'none'),
 ])
 def test_completion_without_tool_call_slow(template_name: str, n_predict: int, tools: list[dict], tool_choice: str | None):
-    do_test_completion_without_tool_call(template_name, n_predict, tools, tool_choice)
+    global server
+    server.jinja = True
+    server.n_predict = n_predict
+    server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
+    server.start(timeout_seconds=TIMEOUT_SERVER_START)
+    do_test_completion_without_tool_call(server, n_predict, tools, tool_choice)
 
 
 @pytest.mark.slow
@@ -281,6 +306,12 @@ def test_completion_without_tool_call_slow(template_name: str, n_predict: int, t
     ("bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      None),
     ("bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      "chatml"),
 
+    ("bartowski/Qwen2.5-1.5B-Instruct-GGUF:Q4_K_M",      None),
+    ("bartowski/Qwen2.5-1.5B-Instruct-GGUF:Q4_K_M",      "chatml"),
+
+    ("bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M",      None),
+    ("bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M",      "chatml"),
+
     ("bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        None),
     ("bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        "chatml"),
 
@@ -324,48 +355,53 @@ def test_weather(hf_repo: str, template_override: str | Tuple[str, str | None] |
     elif isinstance(template_override, str):
         server.chat_template = template_override
     server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
-        "max_tokens": n_predict,
+    do_test_weather(server, max_tokens=n_predict)
+
+
+def do_test_weather(server: ServerProcess, **kwargs):
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "messages": [
             {"role": "system", "content": "You are a chatbot that uses tools/functions. Dont overthink things."},
             {"role": "user", "content": "What is the weather in Istanbul?"},
         ],
         "tools": [WEATHER_TOOL],
+        **kwargs,
     }, timeout=TIMEOUT_HTTP_REQUEST)
     assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
     choice = res.body["choices"][0]
     tool_calls = choice["message"].get("tool_calls")
     assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
     tool_call = tool_calls[0]
-    assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
-    assert tool_call["function"]["name"] == WEATHER_TOOL["function"]["name"]
+    # assert choice["message"].get("content") in (None, ""), f'Expected no content in {choice["message"]}'
+    assert tool_call["function"]["name"] == WEATHER_TOOL["function"]["name"], f'Expected weather tool call, got {tool_call["function"]["name"]}'
+    assert len(tool_call.get("id", "")) > 0, f'Expected non empty tool call id in {tool_call}'
     actual_arguments = json.loads(tool_call["function"]["arguments"])
     assert 'location' in actual_arguments, f"location not found in {json.dumps(actual_arguments)}"
     location = actual_arguments["location"]
     assert isinstance(location, str), f"Expected location to be a string, got {type(location)}: {json.dumps(location)}"
-    assert re.match('^Istanbul(, (TR|Turkey|Türkiye))?$', location), f'Expected Istanbul for location, got {location}'
+    assert re.match('^Istanbul(( |, ?)(TR|Turkey|Türkiye))?$', location), f'Expected Istanbul for location, got {location}'
 
 
 @pytest.mark.slow
 @pytest.mark.parametrize("result_override,n_predict,hf_repo,template_override", [
     (None,                                           128,  "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",       "chatml"),
-    (None,                                           128,  "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",         None),
+    (None,                                           128,  "bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M", None),
+    (None,                                           128,  "bartowski/Qwen2.5-Coder-3B-Instruct-GGUF:Q4_K_M", "chatml"),
     (None,                                           128,  "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",         "chatml"),
     (None,                                           128,  "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M",     ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
     (None,                                           128,  "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",       ("NousResearch/Hermes-3-Llama-3.1-8B", "tool_use")),
     (None,                                           128,  "bartowski/functionary-small-v3.2-GGUF:Q8_0",        ("meetkai/functionary-medium-v3.2", None)),
-    (None,                                           128,  "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M",  None),
     (None,                                           128,  "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M",  None),
     (None,                                           128,  "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M",  "chatml"),
     (None,                                           128,  "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",       None),
+    ("[\\s\\S]*?\\*\\*\\s*0.5($|\\*\\*)",            8192, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", ("llama-cpp-deepseek-r1", None)),
 
     # TODO: fix these (wrong results, either didn't respect decimal instruction or got wrong value)
-    ("[\\s\\S]*?\\*\\*\\s*0.5($|\\*\\*)",            8192, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
-    # ("[\\s\\S]*?\\*\\*\\s*0.5($|\\*\\*)",            8192, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", ("llama-cpp-deepseek-r1", None)),
+    # (None,                                           128,  "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M",  None),
+    # ("[\\s\\S]*?\\*\\*\\s*0.5($|\\*\\*)",            8192, "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
 ])
 def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
     global server
-    # n_predict = 512
     server.n_slots = 1
     server.jinja = True
     server.n_ctx = 8192 * 2
@@ -379,10 +415,14 @@ def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str,
     elif isinstance(template_override, str):
         server.chat_template = template_override
     server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
+    do_test_calc_result(server, result_override, n_predict)
+
+
+def do_test_calc_result(server: ServerProcess, result_override: str | None, n_predict: int, **kwargs):
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "max_tokens": n_predict,
         "messages": [
-            {"role": "system", "content": "You are a chatbot that uses tools/functions. Dont overthink things, and provide very concise answers. Do not explain your reasoning to the user. Provide any numerical values back to the user with at most two decimals."},
+            {"role": "system", "content": "You are a tools-calling assistant. You express numerical values with at most two decimals."},
             {"role": "user", "content": "What's the y coordinate of a point on the unit sphere at angle 30 degrees?"},
             {
                 "role": "assistant",
@@ -423,7 +463,8 @@ def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str,
                     }
                 }
             }
-        ]
+        ],
+        **kwargs,
     }, timeout=TIMEOUT_HTTP_REQUEST)
     assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
     choice = res.body["choices"][0]
@@ -434,19 +475,19 @@ def test_calc_result(result_override: str | None, n_predict: int, hf_repo: str,
     if result_override is not None:
         assert re.match(result_override, content), f'Expected {result_override}, got {content}'
     else:
-        assert re.match('^[\\s\\S]*?The (y[ -])?coordinate [\\s\\S]*?is (approximately )?0\\.56\\b|^0\\.56$', content), \
+        assert re.match('^[\\s\\S]*?((That\'s|\\bis) (approximately )?)?\\b0\\.(5\\b|56\\b|556)', content), \
             f'Expected something like "The y coordinate is 0.56.", got {content}'
 
 
 @pytest.mark.slow
 @pytest.mark.parametrize("n_predict,reasoning_format,expect_content,expect_reasoning_content,hf_repo,template_override", [
-    (128, 'deepseek',  "^The sum of 102 and 7 is 109.*",                        None,                                          "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",       None),
-    (128,  None,        "^The sum of 102 and 7 is 109.*",                       None,                                          "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",       None),
+    (128, 'deepseek',  "^The sum of 102 and 7 is 109[\\s\\S]*",                        None,                                          "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",       None),
+    (128,  None,        "^The sum of 102 and 7 is 109[\\s\\S]*",                       None,                                          "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",       None),
 
-    (1024, 'deepseek',  "To find the sum of.*",                                 "I need to calculate the sum of 102 and 7.*",  "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
-    (1024, 'none',      "^I need[\\s\\S]*?</think>\n?To find.*",                None,                                          "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
+    (1024, 'deepseek',  "To find the sum of[\\s\\S]*",                                 "I need to calculate the sum of 102 and 7[\\s\\S]*",  "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
+    (1024, 'none',      "^(<think>\\s*)?I need[\\s\\S]*?</think>\\s*To find[\\s\\S]*",                None,                                          "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
 
-    (1024, 'deepseek',  "To find the sum of.*",                                 "First, I [\\s\\S]*",                          "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", ("llama-cpp-deepseek-r1", None)),
+    (1024, 'deepseek',  "To find the sum of[\\s\\S]*",                                 "First, I [\\s\\S]*",                          "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", ("llama-cpp-deepseek-r1", None)),
 ])
 def test_thoughts(n_predict: int, reasoning_format: Literal['deepseek', 'none'] | None, expect_content: str | None, expect_reasoning_content: str | None, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
     global server
@@ -464,7 +505,7 @@ def test_thoughts(n_predict: int, reasoning_format: Literal['deepseek', 'none']
     elif isinstance(template_override, str):
         server.chat_template = template_override
     server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "max_tokens": n_predict,
         "messages": [
             {"role": "user", "content": "What's the sum of 102 and 7?"},
@@ -476,7 +517,7 @@ def test_thoughts(n_predict: int, reasoning_format: Literal['deepseek', 'none']
 
     content = choice["message"].get("content")
     if expect_content is None:
-        assert content is None, f'Expected no content in {choice["message"]}'
+        assert choice["message"].get("content") in (None, ""), f'Expected no content in {choice["message"]}'
     else:
         assert re.match(expect_content, content), f'Expected {expect_content}, got {content}'
 
@@ -488,46 +529,46 @@ def test_thoughts(n_predict: int, reasoning_format: Literal['deepseek', 'none']
 
 
 @pytest.mark.slow
-@pytest.mark.parametrize("expected_arguments_override,hf_repo,template_override", [
-    (None,                 "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
-    # (None,                 "bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", "chatml"),
+@pytest.mark.parametrize("hf_repo,template_override", [
+    ("bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q4_K_M", None),
 
-    (None,                 "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      None),
-    (None,                 "bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      "chatml"),
+    ("bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      None),
+    ("bartowski/Phi-3.5-mini-instruct-GGUF:Q4_K_M",      "chatml"),
 
-    (None,                 "bartowski/functionary-small-v3.2-GGUF:Q8_0",       ("meetkai-functionary-medium-v3.2", None)),
-    (None,                 "bartowski/functionary-small-v3.2-GGUF:Q8_0",       "chatml"),
+    ("bartowski/functionary-small-v3.2-GGUF:Q8_0",       ("meetkai-functionary-medium-v3.2", None)),
+    ("bartowski/functionary-small-v3.2-GGUF:Q8_0",       "chatml"),
 
-    ('{"code":"print("}',  "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
-    (None,                 "bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", "chatml"),
+    # ("bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", None),
+    ("bartowski/Meta-Llama-3.1-8B-Instruct-GGUF:Q4_K_M", "chatml"),
 
-    (None,                 "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      ("meta-llama-Llama-3.2-3B-Instruct", None)),
-    (None,                 "bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      "chatml"),
+    ("bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      ("meta-llama-Llama-3.2-3B-Instruct", None)),
+    ("bartowski/Llama-3.2-1B-Instruct-GGUF:Q4_K_M",      None),
 
-    ('{"code":"print("}',  "bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M",      ("meta-llama-Llama-3.2-3B-Instruct", None)),
-    (None,                 "bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M",      "chatml"),
+    ("bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M",      ("meta-llama-Llama-3.2-3B-Instruct", None)),
+    ("bartowski/Llama-3.2-3B-Instruct-GGUF:Q4_K_M",      None),
 
-    (None,                 "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        None),
-    (None,                 "bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        "chatml"),
+    ("bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        None),
+    ("bartowski/Qwen2.5-7B-Instruct-GGUF:Q4_K_M",        "chatml"),
 
-    (None,                 "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M",    ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
-    (None,                 "bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M",    "chatml"),
+    ("bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M",    ("NousResearch/Hermes-2-Pro-Llama-3-8B", "tool_use")),
+    ("bartowski/Hermes-2-Pro-Llama-3-8B-GGUF:Q4_K_M",    "chatml"),
 
-    (None,                 "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",      ("NousResearch-Hermes-3-Llama-3.1-8B", "tool_use")),
-    (None,                 "bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",      "chatml"),
+    ("bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",      ("NousResearch-Hermes-3-Llama-3.1-8B", "tool_use")),
+    ("bartowski/Hermes-3-Llama-3.1-8B-GGUF:Q4_K_M",      "chatml"),
 
-    (None,                 "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
-    (None,                 "bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", "chatml"),
+    ("bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", None),
+    ("bartowski/Mistral-Nemo-Instruct-2407-GGUF:Q4_K_M", "chatml"),
 
-    # Note: gemma-2-2b-it knows itself as "model", not "assistant", so we don't test the ill-suited chatml on it.
-    (None,                 "bartowski/gemma-2-2b-it-GGUF:Q4_K_M",              None),
+    ("bartowski/gemma-2-2b-it-GGUF:Q4_K_M",              None),
+    ("bartowski/gemma-2-2b-it-GGUF:Q4_K_M",              "chatml"),
 ])
-def test_hello_world(expected_arguments_override: str | None, hf_repo: str, template_override: str | Tuple[str, str | None] | None):
+def test_hello_world(hf_repo: str, template_override: str | Tuple[str, str | None] | None):
     global server
+    n_predict = 512 # High because of DeepSeek R1
     server.n_slots = 1
     server.jinja = True
     server.n_ctx = 8192
-    server.n_predict = 512 # High because of DeepSeek R1
+    server.n_predict = n_predict
     server.model_hf_repo = hf_repo
     server.model_hf_file = None
     if isinstance(template_override, tuple):
@@ -537,31 +578,29 @@ def test_hello_world(expected_arguments_override: str | None, hf_repo: str, temp
     elif isinstance(template_override, str):
         server.chat_template = template_override
     server.start(timeout_seconds=TIMEOUT_SERVER_START)
-    res = server.make_request("POST", "/chat/completions", data={
-        "max_tokens": 256,
+
+    do_test_hello_world(server, max_tokens=n_predict)
+
+
+def do_test_hello_world(server: ServerProcess, **kwargs):
+    res = server.make_request("POST", "/v1/chat/completions", data={
         "messages": [
-            {"role": "system", "content": "You are a coding assistant."},
+            {"role": "system", "content": "You are a tool-calling agent."},
             {"role": "user", "content": "say hello world with python"},
         ],
         "tools": [PYTHON_TOOL],
-        # Note: without these greedy params, Functionary v3.2 writes `def hello_world():\n    print("Hello, World!")\nhello_world()` which is correct but a pain to test.
-        "temperature": 0.0,
-        "top_k": 1,
-        "top_p": 1.0,
+        **kwargs,
     }, timeout=TIMEOUT_HTTP_REQUEST)
     assert res.status_code == 200, f"Expected status code 200, got {res.status_code}"
     choice = res.body["choices"][0]
     tool_calls = choice["message"].get("tool_calls")
     assert tool_calls and len(tool_calls) == 1, f'Expected 1 tool call in {choice["message"]}'
     tool_call = tool_calls[0]
-    assert choice["message"].get("content") is None, f'Expected no content in {choice["message"]}'
+    # assert choice["message"].get("content") in (None, ""), f'Expected no content in {choice["message"]}'
     assert tool_call["function"]["name"] == PYTHON_TOOL["function"]["name"]
-    actual_arguments = tool_call["function"]["arguments"]
-    if expected_arguments_override is not None:
-        assert actual_arguments == expected_arguments_override
-    else:
-        actual_arguments = json.loads(actual_arguments)
-        assert 'code' in actual_arguments, f"code not found in {json.dumps(actual_arguments)}"
-        code = actual_arguments["code"]
-        assert isinstance(code, str), f"Expected code to be a string, got {type(code)}: {json.dumps(code)}"
-        assert re.match(r'''print\(("[Hh]ello,? [Ww]orld!?"|'[Hh]ello,? [Ww]orld!?')\)''', code), f'Expected hello world, got {code}'
+    assert len(tool_call.get("id", "")) > 0, f'Expected non empty tool call id in {tool_call}'
+    actual_arguments = json.loads(tool_call["function"]["arguments"])
+    assert 'code' in actual_arguments, f"code not found in {json.dumps(actual_arguments)}"
+    code = actual_arguments["code"]
+    assert isinstance(code, str), f"Expected code to be a string, got {type(code)}: {json.dumps(code)}"
+    assert re.match(r'''print\(("[Hh]ello,? [Ww]orld!?"|'[Hh]ello,? [Ww]orld!?')\)''', code), f'Expected hello world, got {code}'

examples/server/tests/utils.py

@@ -67,6 +67,9 @@ class ServerProcess:
     id_slot: int | None = None
     cache_prompt: bool | None = None
     n_slots: int | None = None
+    ctk: str | None = None
+    ctv: str | None = None
+    fa: bool | None = None
     server_continuous_batching: bool | None = False
     server_embeddings: bool | None = False
     server_reranking: bool | None = False
@@ -84,6 +87,7 @@ class ServerProcess:
     reasoning_format: Literal['deepseek', 'none'] | None = None
     chat_template: str | None = None
     chat_template_file: str | None = None
+    server_path: str | None = None
 
     # session variables
     process: subprocess.Popen | None = None
@@ -97,7 +101,9 @@ class ServerProcess:
             self.server_port = int(os.environ["PORT"])
 
     def start(self, timeout_seconds: int | None = DEFAULT_HTTP_TIMEOUT) -> None:
-        if "LLAMA_SERVER_BIN_PATH" in os.environ:
+        if self.server_path is not None:
+            server_path = self.server_path
+        elif "LLAMA_SERVER_BIN_PATH" in os.environ:
             server_path = os.environ["LLAMA_SERVER_BIN_PATH"]
         elif os.name == "nt":
             server_path = "../../../build/bin/Release/llama-server.exe"
@@ -151,6 +157,12 @@ class ServerProcess:
             server_args.extend(["--ctx-size", self.n_ctx])
         if self.n_slots:
             server_args.extend(["--parallel", self.n_slots])
+        if self.ctk:
+            server_args.extend(["-ctk", self.ctk])
+        if self.ctv:
+            server_args.extend(["-ctv", self.ctv])
+        if self.fa is not None:
+            server_args.append("-fa")
         if self.n_predict:
             server_args.extend(["--n-predict", self.n_predict])
         if self.slot_save_path:
@@ -184,7 +196,7 @@ class ServerProcess:
             server_args.extend(["--chat-template-file", self.chat_template_file])
 
         args = [str(arg) for arg in [server_path, *server_args]]
-        print(f"bench: starting server with: {' '.join(args)}")
+        print(f"tests: starting server with: {' '.join(args)}")
 
         flags = 0
         if "nt" == os.name:
@@ -215,6 +227,10 @@ class ServerProcess:
                     return  # server is ready
             except Exception as e:
                 pass
+            # Check if process died
+            if self.process.poll() is not None:
+                raise RuntimeError(f"Server process died with return code {self.process.returncode}")
+
             print(f"Waiting for server to start...")
             time.sleep(0.5)
         raise TimeoutError(f"Server did not start within {timeout_seconds} seconds")
@@ -286,7 +302,7 @@ class ServerPreset:
         server.model_hf_repo = "ggml-org/models"
         server.model_hf_file = "tinyllamas/stories260K.gguf"
         server.model_alias = "tinyllama-2"
-        server.n_ctx = 256
+        server.n_ctx = 512
         server.n_batch = 32
         server.n_slots = 2
         server.n_predict = 64

examples/server/utils.hpp

@@ -435,6 +435,10 @@ static std::string gen_chatcmplid() {
     return "chatcmpl-" + random_string();
 }
 
+static std::string gen_tool_call_id() {
+    return random_string();
+}
+
 //
 // other common utils
 //
@@ -590,8 +594,8 @@ static json oaicompat_completion_params_parse(
         if (response_type == "json_object") {
             json_schema = json_value(response_format, "schema", json::object());
         } else if (response_type == "json_schema") {
-            json json_schema = json_value(response_format, "json_schema", json::object());
-            json_schema = json_value(json_schema, "schema", json::object());
+            auto schema_wrapper = json_value(response_format, "json_schema", json::object());
+            json_schema = json_value(schema_wrapper, "schema", json::object());
         } else if (!response_type.empty() && response_type != "text") {
             throw std::runtime_error("response_format type must be one of \"text\" or \"json_object\", but got: " + response_type);
         }
@@ -607,6 +611,7 @@ static json oaicompat_completion_params_parse(
     inputs.use_jinja             = use_jinja;
     inputs.parallel_tool_calls   = json_value(body, "parallel_tool_calls", false);
     inputs.extract_reasoning     = reasoning_format != COMMON_REASONING_FORMAT_NONE;
+    inputs.add_generation_prompt = json_value(body, "add_generation_prompt", true);
     if (!inputs.tools.empty() && inputs.tool_choice != COMMON_CHAT_TOOL_CHOICE_NONE && body.contains("grammar")) {
         throw std::runtime_error("Cannot use custom grammar constraints with tools.");
     }
@@ -616,14 +621,13 @@ static json oaicompat_completion_params_parse(
 
     llama_params["chat_format"]      = static_cast<int>(chat_params.format);
     llama_params["prompt"]           = chat_params.prompt;
-    llama_params["grammar"]          = chat_params.grammar;
+    if (!chat_params.grammar.empty()) {
+        llama_params["grammar"] = chat_params.grammar;
+    }
     llama_params["grammar_lazy"]     = chat_params.grammar_lazy;
     auto grammar_triggers = json::array();
     for (const auto & trigger : chat_params.grammar_triggers) {
-        grammar_triggers.push_back({
-            {"word", trigger.word},
-            {"at_start", trigger.at_start},
-        });
+        grammar_triggers.push_back(trigger.to_json<json>());
     }
     llama_params["grammar_triggers"] = grammar_triggers;
     llama_params["preserved_tokens"] = chat_params.preserved_tokens;

examples/simple-chat/simple-chat.cpp

@@ -98,7 +98,7 @@ int main(int argc, char ** argv) {
     auto generate = [&](const std::string & prompt) {
         std::string response;
 
-        const bool is_first = llama_get_kv_cache_used_cells(ctx) == 0;
+        const bool is_first = llama_kv_self_used_cells(ctx) == 0;
 
         // tokenize the prompt
         const int n_prompt_tokens = -llama_tokenize(vocab, prompt.c_str(), prompt.size(), NULL, 0, is_first, true);
@@ -113,7 +113,7 @@ int main(int argc, char ** argv) {
         while (true) {
             // check if we have enough space in the context to evaluate this batch
             int n_ctx = llama_n_ctx(ctx);
-            int n_ctx_used = llama_get_kv_cache_used_cells(ctx);
+            int n_ctx_used = llama_kv_self_used_cells(ctx);
             if (n_ctx_used + batch.n_tokens > n_ctx) {
                 printf("\033[0m\n");
                 fprintf(stderr, "context size exceeded\n");

examples/speculative-simple/speculative-simple.cpp

@@ -217,7 +217,7 @@ int main(int argc, char ** argv) {
         {
             LOG_DBG("clear kv cache from any extra tokens, n_past = %d\n", n_past);
 
-            llama_kv_cache_seq_rm(ctx_tgt, 0, n_past, -1);
+            llama_kv_self_seq_rm(ctx_tgt, 0, n_past, -1);
         }
 
         if ((params.n_predict >= 0 && n_predict > params.n_predict) || has_eos) {

examples/speculative/speculative.cpp

@@ -420,14 +420,14 @@ int main(int argc, char ** argv) {
             {
                 LOG_DBG("keeping sequence %d, n_past_tgt = %d, n_past_dft = %d\n", s_keep, n_past_tgt, n_past_dft);
 
-                llama_kv_cache_seq_keep(ctx_dft, s_keep);
-                llama_kv_cache_seq_cp  (ctx_dft, s_keep, 0, -1, -1);
-                llama_kv_cache_seq_keep(ctx_dft, 0);
+                llama_kv_self_seq_keep(ctx_dft, s_keep);
+                llama_kv_self_seq_cp  (ctx_dft, s_keep, 0, -1, -1);
+                llama_kv_self_seq_keep(ctx_dft, 0);
 
-                llama_kv_cache_seq_rm  (ctx_tgt, s_keep, n_past_tgt, -1);
-                llama_kv_cache_seq_keep(ctx_tgt, s_keep);
-                llama_kv_cache_seq_cp  (ctx_tgt, s_keep, 0, -1, -1);
-                llama_kv_cache_seq_keep(ctx_tgt, 0);
+                llama_kv_self_seq_rm  (ctx_tgt, s_keep, n_past_tgt, -1);
+                llama_kv_self_seq_keep(ctx_tgt, s_keep);
+                llama_kv_self_seq_cp  (ctx_tgt, s_keep, 0, -1, -1);
+                llama_kv_self_seq_keep(ctx_tgt, 0);
             }
 
             for (int s = 0; s < n_seq_dft; ++s) {
@@ -444,7 +444,7 @@ int main(int argc, char ** argv) {
             common_batch_clear(batch_dft);
             common_batch_add  (batch_dft, token_id, n_past_dft, { 0 }, true);
 
-            llama_kv_cache_seq_rm(ctx_dft, 0, n_past_dft, -1);
+            llama_kv_self_seq_rm(ctx_dft, 0, n_past_dft, -1);
             // LOG_DBG("dft batch: %s\n", LOG_BATCH_TOSTR_PRETTY(ctx_dft, batch_dft).c_str());
             llama_decode(ctx_dft, batch_dft);
 
@@ -503,8 +503,8 @@ int main(int argc, char ** argv) {
                     if (n_seq_cur < n_seq_dft && cur_p->data[f].p > p_draft_split) {
                         LOG_DBG("splitting seq %3d into %3d\n", s, n_seq_cur);
 
-                        llama_kv_cache_seq_rm(ctx_dft,    n_seq_cur, -1, -1);
-                        llama_kv_cache_seq_cp(ctx_dft, s, n_seq_cur, -1, -1);
+                        llama_kv_self_seq_rm(ctx_dft,    n_seq_cur, -1, -1);
+                        llama_kv_self_seq_cp(ctx_dft, s, n_seq_cur, -1, -1);
 
                         // all previous tokens from this branch are now also part of the new branch
                         for (int t = 0; t < batch_tgt.n_tokens; ++t) {
@@ -585,9 +585,9 @@ int main(int argc, char ** argv) {
 
         // evaluate the target model on the drafted tokens
         {
-            llama_kv_cache_seq_keep(ctx_tgt, 0);
+            llama_kv_self_seq_keep(ctx_tgt, 0);
             for (int s = 1; s < n_seq_dft; ++s) {
-                llama_kv_cache_seq_cp(ctx_tgt, 0, s, -1, -1);
+                llama_kv_self_seq_cp(ctx_tgt, 0, s, -1, -1);
             }
 
             // LOG_DBG("target batch: %s\n", LOG_BATCH_TOSTR_PRETTY(ctx_tgt, batch_tgt).c_str());

examples/tts/tts.cpp

@@ -1,3 +1,5 @@
+#define _USE_MATH_DEFINES // For M_PI on MSVC
+
 #include "arg.h"
 #include "common.h"
 #include "sampling.h"
@@ -5,8 +7,6 @@
 #include "llama.h"
 #include "json.hpp"
 
-#define _USE_MATH_DEFINES // For M_PI on MSVC
-
 #include <algorithm>
 #include <cmath>
 #include <cstdio>
@@ -87,11 +87,11 @@ struct wav_header {
     uint32_t data_size;
 };
 
-static void save_wav16(const std::string & fname, const std::vector<float> & data, int sample_rate) {
+static bool save_wav16(const std::string & fname, const std::vector<float> & data, int sample_rate) {
     std::ofstream file(fname, std::ios::binary);
     if (!file) {
-        LOG_ERR("%s: Failed to open file '%s' for writing", __func__, fname.c_str());
-        return;
+        LOG_ERR("%s: Failed to open file '%s' for writing.\n", __func__, fname.c_str());
+        return false;
     }
 
     wav_header header;
@@ -108,7 +108,7 @@ static void save_wav16(const std::string & fname, const std::vector<float> & dat
         file.write(reinterpret_cast<const char*>(&pcm_sample), sizeof(pcm_sample));
     }
 
-    file.close();
+    return file.good();
 }
 
 static void fill_hann_window(int length, bool periodic, float * output) {
@@ -536,6 +536,7 @@ static std::string audio_data_from_speaker(json speaker, const outetts_version t
 int main(int argc, char ** argv) {
     common_params params;
 
+    params.out_file = "output.wav";
     params.prompt = "";
 
     params.n_predict = 4096;
@@ -1060,8 +1061,6 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
     }
 #endif
 
-    const std::string fname = "output.wav";
-
     const int n_sr = 24000; // sampling rate
 
     // zero out first 0.25 seconds
@@ -1072,11 +1071,15 @@ lovely<|t_0.56|><|code_start|><|634|><|596|><|1766|><|1556|><|1306|><|1285|><|14
     LOG_INF("%s: time for spectral ops: %.3f ms\n", __func__, (ggml_time_us() - t_spec_start) / 1000.0f);
     LOG_INF("%s: total time:            %.3f ms\n", __func__, (ggml_time_us() - t_main_start) / 1000.0f);
 
-    save_wav16(fname, audio, n_sr);
+    int retval = 0;
 
-    LOG_INF("%s: audio written to file '%s'\n", __func__, fname.c_str());
+    if (save_wav16(params.out_file, audio, n_sr)) {
+        LOG_INF("%s: audio written to file '%s'\n", __func__, params.out_file.c_str());
+    } else {
+        retval = ENOENT;
+    }
 
     llama_backend_free();
 
-    return 0;
+    return retval;
 }

ggml/CMakeLists.txt

@@ -106,6 +106,7 @@ option(GGML_CPU_KLEIDIAI     "ggml: use KleidiAI optimized kernels if applicable
 option(GGML_AVX              "ggml: enable AVX"              ${INS_ENB})
 option(GGML_AVX_VNNI         "ggml: enable AVX-VNNI"         OFF)
 option(GGML_AVX2             "ggml: enable AVX2"             ${INS_ENB})
+option(GGML_BMI2             "ggml: enable BMI2"             ${INS_ENB})
 option(GGML_AVX512           "ggml: enable AVX512F"          OFF)
 option(GGML_AVX512_VBMI      "ggml: enable AVX512-VBMI"      OFF)
 option(GGML_AVX512_VNNI      "ggml: enable AVX512-VNNI"      OFF)
@@ -194,6 +195,8 @@ option(GGML_OPENCL                          "ggml: use OpenCL"
 option(GGML_OPENCL_PROFILING                "ggml: use OpenCL profiling (increases overhead)" OFF)
 option(GGML_OPENCL_EMBED_KERNELS            "ggml: embed kernels"                             ON)
 option(GGML_OPENCL_USE_ADRENO_KERNELS       "ggml: use optimized kernels for Adreno"          ON)
+set   (GGML_OPENCL_TARGET_VERSION "300" CACHE STRING
+                                            "gmml: OpenCL API version to target")
 
 # toolchain for vulkan-shaders-gen
 set   (GGML_VULKAN_SHADERS_GEN_TOOLCHAIN "" CACHE FILEPATH "ggml: toolchain file for vulkan-shaders-gen")

ggml/cmake/common.cmake

@@ -0,0 +1,26 @@
+function(ggml_get_flags CCID CCVER)
+    set(C_FLAGS "")
+    set(CXX_FLAGS "")
+
+    if (CCID MATCHES "Clang")
+        set(C_FLAGS   -Wunreachable-code-break -Wunreachable-code-return)
+        set(CXX_FLAGS -Wunreachable-code-break -Wunreachable-code-return -Wmissing-prototypes -Wextra-semi)
+
+        if (
+            (CCID STREQUAL "Clang"      AND CCVER VERSION_GREATER_EQUAL 3.8.0) OR
+            (CCID STREQUAL "AppleClang" AND CCVER VERSION_GREATER_EQUAL 7.3.0)
+        )
+            list(APPEND C_FLAGS -Wdouble-promotion)
+        endif()
+    elseif (CCID STREQUAL "GNU")
+        set(C_FLAGS   -Wdouble-promotion)
+        set(CXX_FLAGS -Wno-array-bounds)
+
+        if (CCVER VERSION_GREATER_EQUAL 8.1.0)
+            list(APPEND CXX_FLAGS -Wextra-semi)
+        endif()
+    endif()
+
+    set(GF_C_FLAGS   ${C_FLAGS}   PARENT_SCOPE)
+    set(GF_CXX_FLAGS ${CXX_FLAGS} PARENT_SCOPE)
+endfunction()

ggml/include/ggml-cpu.h

@@ -80,6 +80,7 @@ extern "C" {
     GGML_BACKEND_API int ggml_cpu_has_avx        (void);
     GGML_BACKEND_API int ggml_cpu_has_avx_vnni   (void);
     GGML_BACKEND_API int ggml_cpu_has_avx2       (void);
+    GGML_BACKEND_API int ggml_cpu_has_bmi2       (void);
     GGML_BACKEND_API int ggml_cpu_has_f16c       (void);
     GGML_BACKEND_API int ggml_cpu_has_fma        (void);
     GGML_BACKEND_API int ggml_cpu_has_avx512     (void);

ggml/include/ggml.h

@@ -2140,7 +2140,11 @@ extern "C" {
 #        define GGML_RESTRICT
 #    endif
 #else
-#    define GGML_RESTRICT restrict
+#    if defined (_MSC_VER) && (__STDC_VERSION__ < 201112L)
+#        define GGML_RESTRICT __restrict
+#    else
+#        define GGML_RESTRICT restrict
+#    endif
 #endif
     typedef void (*ggml_to_float_t)  (const void  * GGML_RESTRICT x, float * GGML_RESTRICT y, int64_t k);
     typedef void (*ggml_from_float_t)(const float * GGML_RESTRICT x, void  * GGML_RESTRICT y, int64_t k);

ggml/src/CMakeLists.txt

@@ -1,4 +1,5 @@
 include(CheckCXXCompilerFlag)
+include("../cmake/common.cmake")
 
 add_compile_definitions(GGML_SCHED_MAX_COPIES=${GGML_SCHED_MAX_COPIES})
 
@@ -24,33 +25,6 @@ if (NOT MSVC)
     endif()
 endif()
 
-function(ggml_get_flags CCID CCVER)
-    set(C_FLAGS "")
-    set(CXX_FLAGS "")
-
-    if (CCID MATCHES "Clang")
-        set(C_FLAGS   -Wunreachable-code-break -Wunreachable-code-return)
-        set(CXX_FLAGS -Wunreachable-code-break -Wunreachable-code-return -Wmissing-prototypes -Wextra-semi)
-
-        if (
-            (CCID STREQUAL "Clang"      AND CCVER VERSION_GREATER_EQUAL 3.8.0) OR
-            (CCID STREQUAL "AppleClang" AND CCVER VERSION_GREATER_EQUAL 7.3.0)
-        )
-            list(APPEND C_FLAGS -Wdouble-promotion)
-        endif()
-    elseif (CCID STREQUAL "GNU")
-        set(C_FLAGS   -Wdouble-promotion)
-        set(CXX_FLAGS -Wno-array-bounds)
-
-        if (CCVER VERSION_GREATER_EQUAL 8.1.0)
-            list(APPEND CXX_FLAGS -Wextra-semi)
-        endif()
-    endif()
-
-    set(GF_C_FLAGS   ${C_FLAGS}   PARENT_SCOPE)
-    set(GF_CXX_FLAGS ${CXX_FLAGS} PARENT_SCOPE)
-endfunction()
-
 if (GGML_FATAL_WARNINGS)
     if (CMAKE_CXX_COMPILER_ID MATCHES "GNU" OR CMAKE_CXX_COMPILER_ID MATCHES "Clang")
         list(APPEND C_FLAGS   -Werror)
@@ -236,7 +210,7 @@ add_library(ggml
 target_link_libraries(ggml PUBLIC ggml-base)
 
 if (CMAKE_SYSTEM_NAME MATCHES "Linux")
-    target_link_libraries(ggml PRIVATE dl)
+    target_link_libraries(ggml PRIVATE dl stdc++fs)
 endif()
 
 function(ggml_add_backend_library backend)
@@ -289,7 +263,7 @@ function(ggml_add_cpu_backend_variant tag_name)
     set(GGML_CPU_TAG_NAME ${tag_name})
     # other: OPENMP LLAMAFILE CPU_HBM
     foreach (feat NATIVE
-                  AVX AVX2 AVX_VNNI FMA F16C
+                  AVX AVX2 BMI2 AVX_VNNI FMA F16C
                   AVX512 AVX512_VBMI AVX512_VNNI AVX512_BF16
                   AMX_TILE AMX_INT8 AMX_BF16)
         set(GGML_${feat} OFF)
@@ -309,13 +283,13 @@ if (GGML_CPU_ALL_VARIANTS)
         message(FATAL_ERROR "GGML_CPU_ALL_VARIANTS requires GGML_BACKEND_DL")
     endif()
     ggml_add_cpu_backend_variant(sandybridge    AVX)
-    ggml_add_cpu_backend_variant(haswell        AVX F16C AVX2 FMA)
-    ggml_add_cpu_backend_variant(skylakex       AVX F16C AVX2 FMA AVX512)
-    ggml_add_cpu_backend_variant(icelake        AVX F16C AVX2 FMA AVX512 AVX512_VBMI AVX512_VNNI)
-    ggml_add_cpu_backend_variant(alderlake      AVX F16C AVX2 FMA AVX_VNNI)
+    ggml_add_cpu_backend_variant(haswell        AVX F16C AVX2 BMI2 FMA)
+    ggml_add_cpu_backend_variant(skylakex       AVX F16C AVX2 BMI2 FMA AVX512)
+    ggml_add_cpu_backend_variant(icelake        AVX F16C AVX2 BMI2 FMA AVX512 AVX512_VBMI AVX512_VNNI)
+    ggml_add_cpu_backend_variant(alderlake      AVX F16C AVX2 BMI2 FMA AVX_VNNI)
     if (NOT MSVC)
         # MSVC doesn't support AMX
-        ggml_add_cpu_backend_variant(sapphirerapids AVX F16C AVX2 FMA AVX512 AVX512_VBMI AVX512_VNNI AVX512_BF16 AMX_TILE AMX_INT8)
+        ggml_add_cpu_backend_variant(sapphirerapids AVX F16C AVX2 BMI2 FMA AVX512 AVX512_VBMI AVX512_VNNI AVX512_BF16 AMX_TILE AMX_INT8)
     endif()
 elseif (GGML_CPU)
     ggml_add_cpu_backend_variant_impl("")

ggml/src/ggml-backend-reg.cpp

@@ -8,6 +8,7 @@
 #include <string>
 #include <type_traits>
 #include <vector>
+#include <cctype>
 
 #ifdef _WIN32
 #    define WIN32_LEAN_AND_MEAN
@@ -75,7 +76,14 @@ namespace fs = std::filesystem;
 static std::string path_str(const fs::path & path) {
     std::string u8path;
     try {
+#if defined(__cpp_lib_char8_t)
+        // C++20 and later: u8string() returns std::u8string
+        std::u8string u8str = path.u8string();
+        u8path = std::string(reinterpret_cast<const char*>(u8str.c_str()));
+#else
+        // C++17: u8string() returns std::string
         u8path = path.u8string();
+#endif
     } catch (...) {
     }
     return u8path;
@@ -489,7 +497,7 @@ static ggml_backend_reg_t ggml_backend_load_best(const char * name, bool silent,
         search_paths.push_back(get_executable_path());
         search_paths.push_back(fs::current_path());
     } else {
-        search_paths.push_back(user_search_path);
+        search_paths.push_back(fs::u8path(user_search_path));
     }
 
     int best_score = 0;
@@ -503,9 +511,9 @@ static ggml_backend_reg_t ggml_backend_load_best(const char * name, bool silent,
         fs::directory_iterator dir_it(search_path, fs::directory_options::skip_permission_denied);
         for (const auto & entry : dir_it) {
             if (entry.is_regular_file()) {
-                auto filename = entry.path().filename().native();
-                auto ext = entry.path().extension().native();
-                if (filename.find(file_prefix) == 0 && ext == file_extension) {
+                auto filename = entry.path().filename();
+                auto ext = entry.path().extension();
+                if (filename.native().find(file_prefix) == 0 && ext == file_extension) {
                     dl_handle_ptr handle { dl_load_library(entry) };
                     if (!handle && !silent) {
                         GGML_LOG_ERROR("%s: failed to load %s\n", __func__, path_str(entry.path()).c_str());
@@ -536,7 +544,7 @@ static ggml_backend_reg_t ggml_backend_load_best(const char * name, bool silent,
         // try to load the base backend
         for (const auto & search_path : search_paths) {
             fs::path filename = backend_filename_prefix().native() + name_path.native() + backend_filename_extension().native();
-            fs::path path = search_path.native() + filename.native();
+            fs::path path = search_path / filename;
             if (fs::exists(path)) {
                 return get_reg().load_backend(path, silent);
             }

ggml/src/ggml-backend.cpp

@@ -21,6 +21,7 @@
 #include <string.h>
 #include <string>
 #include <vector>
+#include <algorithm>
 
 #ifdef __APPLE__
 #include <sys/types.h>

ggml/src/ggml-cann/aclnn_ops.cpp

@@ -2790,10 +2790,14 @@ static void ggml_cann_mul_mat_quant(ggml_backend_cann_context& ctx,
                 (char*)output_buffer + batch1 * output_stride, ACL_FLOAT16,
                 output_elem_size, output_ne, output_nb, 2, ACL_FORMAT_ND,
                 output_ne_offset);
+            int64_t antiquantGroupSize = 0;
+            if (src0->ne[0] > QK8_0) {
+                antiquantGroupSize = QK8_0;
+            }
 
             ACL_CHECK(aclnnWeightQuantBatchMatmulV2GetWorkspaceSize(
                 acl_input_tensor, acl_weight_tensor, acl_scale_tensor, nullptr,
-                nullptr, nullptr, nullptr, QK8_0, acl_output_tensor,
+                nullptr, nullptr, nullptr, antiquantGroupSize, acl_output_tensor,
                 &workspaceSize, &executor));
             if (workspaceAddr == nullptr) {
                 workspaceAddr = workspace_allocator.alloc(workspaceSize);
@@ -2833,7 +2837,7 @@ static void ggml_cann_mul_mat_quant(ggml_backend_cann_context& ctx,
 
                 ACL_CHECK(aclnnWeightQuantBatchMatmulV2GetWorkspaceSize(
                     acl_input_tensor, acl_weight_tensor, acl_scale_tensor,
-                    nullptr, nullptr, nullptr, nullptr, QK8_0,
+                    nullptr, nullptr, nullptr, nullptr, antiquantGroupSize,
                     acl_output_tensor, &workspaceSize, &executor));
                 ACL_CHECK(aclnnWeightQuantBatchMatmulV2(
                     workspaceAddr, workspaceSize, executor, ctx.stream()));

ggml/src/ggml-cann/ggml-cann.cpp

@@ -1689,11 +1689,6 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
         case GGML_OP_MUL_MAT: {
             switch (op->src[0]->type) {
                 case GGML_TYPE_Q8_0:
-                    // Current groupsize should not be greater than k-1 in
-                    // aclnnWeightQuantBatchMatmulV2GetWorkspaceSize
-                    if (op->src[0]->ne[0] <= QK8_0) {
-                        return false;
-                    }
                 case GGML_TYPE_F16:
                 case GGML_TYPE_F32:
                 case GGML_TYPE_Q4_0:

ggml/src/ggml-cpu/CMakeLists.txt

@@ -219,6 +219,10 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
             if (GGML_AVX_VNNI)
                 list(APPEND ARCH_DEFINITIONS __AVXVNNI__ GGML_AVX_VNNI)
             endif()
+            if (GGML_BMI2)
+                # MSVC does not define macro __BMI2__
+                list(APPEND ARCH_DEFINITIONS __BMI2__ GGML_BMI2)
+            endif()
         else ()
             if (GGML_NATIVE)
                 list(APPEND ARCH_FLAGS -march=native)
@@ -233,6 +237,10 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
                     list(APPEND ARCH_FLAGS -mfma)
                     list(APPEND ARCH_DEFINITIONS GGML_FMA)
                 endif()
+                if (GGML_BMI2)
+                    list(APPEND ARCH_FLAGS -mbmi2)
+                    list(APPEND ARCH_DEFINITIONS GGML_BMI2)
+                endif()
                 if (GGML_AVX)
                     list(APPEND ARCH_FLAGS -mavx)
                     list(APPEND ARCH_DEFINITIONS GGML_AVX)

ggml/src/ggml-cpu/cpu-feats-x86.cpp

@@ -278,6 +278,10 @@ static int ggml_backend_cpu_x86_score() {
     if (!is.SSE42()) { return 0; }
     score += 1<<2;
 #endif
+#ifdef GGML_BMI2
+    if (!is.BMI2()) { return 0; }
+    score += 1<<3;
+#endif
 #ifdef GGML_AVX
     if (!is.AVX()) { return 0; }
     score += 1<<4;

ggml/src/ggml-cpu/ggml-cpu.c

@@ -247,9 +247,9 @@ typedef pthread_t ggml_thread_t;
 static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
 
 
-static void ggml_vec_dot_f32(int n, float * restrict s, size_t bs, const float * restrict x, size_t bx, const float * restrict y, size_t by, int nrc);
-static void ggml_vec_dot_f16(int n, float * restrict s, size_t bs, ggml_fp16_t * restrict x, size_t bx, ggml_fp16_t * restrict y, size_t by, int nrc);
-static void ggml_vec_dot_bf16(int n, float * restrict s, size_t bs, ggml_bf16_t * restrict x, size_t bx, ggml_bf16_t * restrict y, size_t by, int nrc);
+static void ggml_vec_dot_f32(int n, float * GGML_RESTRICT s, size_t bs, const float * GGML_RESTRICT x, size_t bx, const float * GGML_RESTRICT y, size_t by, int nrc);
+static void ggml_vec_dot_f16(int n, float * GGML_RESTRICT s, size_t bs, ggml_fp16_t * GGML_RESTRICT x, size_t bx, ggml_fp16_t * GGML_RESTRICT y, size_t by, int nrc);
+static void ggml_vec_dot_bf16(int n, float * GGML_RESTRICT s, size_t bs, ggml_bf16_t * GGML_RESTRICT x, size_t bx, ggml_bf16_t * GGML_RESTRICT y, size_t by, int nrc);
 
 static const struct ggml_type_traits_cpu type_traits_cpu[GGML_TYPE_COUNT] = {
     [GGML_TYPE_F32] = {
@@ -1451,7 +1451,7 @@ inline static void ggml_vec_div_f16 (const int n, ggml_fp16_t * z, const ggml_fp
     }
 }
 
-static void ggml_vec_dot_f32(int n, float * restrict s, size_t bs, const float * restrict x, size_t bx, const float * restrict y, size_t by, int nrc) {
+static void ggml_vec_dot_f32(int n, float * GGML_RESTRICT s, size_t bs, const float * GGML_RESTRICT x, size_t bx, const float * GGML_RESTRICT y, size_t by, int nrc) {
    assert(nrc == 1);
    UNUSED(nrc);
    UNUSED(bx);
@@ -1494,7 +1494,7 @@ static void ggml_vec_dot_f32(int n, float * restrict s, size_t bs, const float *
     *s = sumf;
 }
 
-static void ggml_vec_dot_bf16(int n, float * restrict s, size_t bs, ggml_bf16_t * restrict x, size_t bx, ggml_bf16_t * restrict y, size_t by, int nrc) {
+static void ggml_vec_dot_bf16(int n, float * GGML_RESTRICT s, size_t bs, ggml_bf16_t * GGML_RESTRICT x, size_t bx, ggml_bf16_t * GGML_RESTRICT y, size_t by, int nrc) {
     assert(nrc == 1);
     UNUSED(nrc);
     UNUSED(bx);
@@ -1562,7 +1562,7 @@ static void ggml_vec_dot_bf16(int n, float * restrict s, size_t bs, ggml_bf16_t
     *s = sumf;
 }
 
-static void ggml_vec_dot_f16(int n, float * restrict s, size_t bs, ggml_fp16_t * restrict x, size_t bx, ggml_fp16_t * restrict y, size_t by, int nrc) {
+static void ggml_vec_dot_f16(int n, float * GGML_RESTRICT s, size_t bs, ggml_fp16_t * GGML_RESTRICT x, size_t bx, ggml_fp16_t * GGML_RESTRICT y, size_t by, int nrc) {
     assert(nrc == 1);
     UNUSED(nrc);
     UNUSED(bx);
@@ -1606,10 +1606,10 @@ static void ggml_vec_dot_f16(int n, float * restrict s, size_t bs, ggml_fp16_t *
 
 // compute GGML_VEC_DOT_UNROLL dot products at once
 // xs - x row stride in bytes
-inline static void ggml_vec_dot_f16_unroll(const int n, const int xs, float * restrict s, void * restrict xv, ggml_fp16_t * restrict y) {
+inline static void ggml_vec_dot_f16_unroll(const int n, const int xs, float * GGML_RESTRICT s, void * GGML_RESTRICT xv, ggml_fp16_t * GGML_RESTRICT y) {
     ggml_float sumf[GGML_VEC_DOT_UNROLL] = { 0.0 };
 
-    ggml_fp16_t * restrict x[GGML_VEC_DOT_UNROLL];
+    ggml_fp16_t * GGML_RESTRICT x[GGML_VEC_DOT_UNROLL];
 
     for (int i = 0; i < GGML_VEC_DOT_UNROLL; ++i) {
         x[i] = (ggml_fp16_t *) ((char *) xv + i*xs);
@@ -1659,7 +1659,7 @@ inline static void ggml_vec_dot_f16_unroll(const int n, const int xs, float * re
     }
 }
 
-inline static void ggml_vec_mad_f32(const int n, float * restrict y, const float * restrict x, const float v) {
+inline static void ggml_vec_mad_f32(const int n, float * GGML_RESTRICT y, const float * GGML_RESTRICT x, const float v) {
 #if defined(GGML_SIMD)
     const int np = (n & ~(GGML_F32_STEP - 1));
 
@@ -1690,7 +1690,7 @@ inline static void ggml_vec_mad_f32(const int n, float * restrict y, const float
 #endif
 }
 
-inline static void ggml_vec_mad_f16(const int n, ggml_fp16_t * restrict y, const ggml_fp16_t * restrict x, const float v) {
+inline static void ggml_vec_mad_f16(const int n, ggml_fp16_t * GGML_RESTRICT y, const ggml_fp16_t * GGML_RESTRICT x, const float v) {
 #if defined(GGML_SIMD)
     const int np = (n & ~(GGML_F16_STEP - 1));
 
@@ -1722,10 +1722,10 @@ inline static void ggml_vec_mad_f16(const int n, ggml_fp16_t * restrict y, const
 }
 
 // xs and vs are byte strides of x and v
-inline static void ggml_vec_mad_f32_unroll(const int n, const int xs, const int vs, float * restrict y, const float * restrict xv, const float * restrict vv) {
+inline static void ggml_vec_mad_f32_unroll(const int n, const int xs, const int vs, float * GGML_RESTRICT y, const float * GGML_RESTRICT xv, const float * GGML_RESTRICT vv) {
 
-    const float * restrict x[GGML_VEC_MAD_UNROLL];
-    const float * restrict v[GGML_VEC_MAD_UNROLL];
+    const float * GGML_RESTRICT x[GGML_VEC_MAD_UNROLL];
+    const float * GGML_RESTRICT v[GGML_VEC_MAD_UNROLL];
 
     for (int i = 0; i < GGML_VEC_MAD_UNROLL; ++i) {
         x[i] = (const float *) ((const char *) xv + i*xs);
@@ -6648,6 +6648,135 @@ static void ggml_compute_forward_repeat_back(
 
 // ggml_compute_forward_concat
 
+static void ggml_compute_forward_concat_any(
+    const struct ggml_compute_params * params,
+    struct ggml_tensor * dst) {
+
+    const struct ggml_tensor * src0 = dst->src[0];
+    const struct ggml_tensor * src1 = dst->src[1];
+
+    const size_t len = ggml_type_size(src0->type);
+
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    GGML_TENSOR_BINARY_OP_LOCALS
+
+    const int32_t dim = ggml_get_op_params_i32(dst, 0);
+
+    GGML_ASSERT(dim >= 0 && dim < 4);
+
+    int64_t o[4] = {0, 0, 0, 0};
+    o[dim] = src0->ne[dim];
+
+    const char * x;
+
+    // TODO: smarter multi-theading
+    for (int i3 = 0; i3 < ne3; i3++) {
+        for (int i2 = ith; i2 < ne2; i2 += nth) {
+            for (int i1 = 0; i1 < ne1; i1++) {
+                for (int i0 = 0; i0 < ne0; i0++) {
+                    if (i0 < ne00 && i1 < ne01 && i2 < ne02 && i3 < ne03) {
+                        x = (const char *)src0->data + (i0       )*nb00 + (i1       )*nb01 + (i2       )*nb02 + (i3       )*nb03;
+                    } else {
+                        x = (const char *)src1->data + (i0 - o[0])*nb10 + (i1 - o[1])*nb11 + (i2 - o[2])*nb12 + (i3 - o[3])*nb13;
+                    }
+
+                    char * y = (char *)dst->data + i0*nb0 + i1*nb1 + i2*nb2 + i3*nb3;
+
+                    memcpy(y, x, len);
+                }
+            }
+        }
+    }
+}
+
+static void ggml_compute_forward_concat_i8(
+    const struct ggml_compute_params * params,
+    struct ggml_tensor * dst) {
+
+    const struct ggml_tensor * src0 = dst->src[0];
+    const struct ggml_tensor * src1 = dst->src[1];
+
+    GGML_ASSERT(ggml_type_size(src0->type) == sizeof(int8_t));
+
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    GGML_TENSOR_BINARY_OP_LOCALS
+
+    const int32_t dim = ggml_get_op_params_i32(dst, 0);
+
+    GGML_ASSERT(dim >= 0 && dim < 4);
+
+    int64_t o[4] = {0, 0, 0, 0};
+    o[dim] = src0->ne[dim];
+
+    const int8_t * x;
+
+    // TODO: smarter multi-theading
+    for (int i3 = 0; i3 < ne3; i3++) {
+        for (int i2 = ith; i2 < ne2; i2 += nth) {
+            for (int i1 = 0; i1 < ne1; i1++) {
+                for (int i0 = 0; i0 < ne0; i0++) {
+                    if (i0 < ne00 && i1 < ne01 && i2 < ne02 && i3 < ne03) {
+                        x = (const int8_t *) ((const char *)src0->data + (i0       )*nb00 + (i1       )*nb01 + (i2       )*nb02 + (i3       )*nb03);
+                    } else {
+                        x = (const int8_t *) ((const char *)src1->data + (i0 - o[0])*nb10 + (i1 - o[1])*nb11 + (i2 - o[2])*nb12 + (i3 - o[3])*nb13);
+                    }
+
+                    int8_t * y = (int8_t *)((char *)dst->data + i0*nb0 + i1*nb1 + i2*nb2 + i3*nb3);
+
+                    *y = *x;
+                }
+            }
+        }
+    }
+}
+
+static void ggml_compute_forward_concat_f16(
+    const struct ggml_compute_params * params,
+    struct ggml_tensor * dst) {
+
+    const struct ggml_tensor * src0 = dst->src[0];
+    const struct ggml_tensor * src1 = dst->src[1];
+
+    GGML_ASSERT(ggml_type_size(src0->type) == sizeof(ggml_fp16_t));
+
+    const int ith = params->ith;
+    const int nth = params->nth;
+
+    GGML_TENSOR_BINARY_OP_LOCALS
+
+    const int32_t dim = ggml_get_op_params_i32(dst, 0);
+
+    GGML_ASSERT(dim >= 0 && dim < 4);
+
+    int64_t o[4] = {0, 0, 0, 0};
+    o[dim] = src0->ne[dim];
+
+    const ggml_fp16_t * x;
+
+    // TODO: smarter multi-theading
+    for (int i3 = 0; i3 < ne3; i3++) {
+        for (int i2 = ith; i2 < ne2; i2 += nth) {
+            for (int i1 = 0; i1 < ne1; i1++) {
+                for (int i0 = 0; i0 < ne0; i0++) {
+                    if (i0 < ne00 && i1 < ne01 && i2 < ne02 && i3 < ne03) {
+                        x = (const ggml_fp16_t *) ((const char *)src0->data + (i0       )*nb00 + (i1       )*nb01 + (i2       )*nb02 + (i3       )*nb03);
+                    } else {
+                        x = (const ggml_fp16_t *) ((const char *)src1->data + (i0 - o[0])*nb10 + (i1 - o[1])*nb11 + (i2 - o[2])*nb12 + (i3 - o[3])*nb13);
+                    }
+
+                    ggml_fp16_t * y = (ggml_fp16_t *)((char *)dst->data + i0*nb0 + i1*nb1 + i2*nb2 + i3*nb3);
+
+                    *y = *x;
+                }
+            }
+        }
+    }
+}
+
 static void ggml_compute_forward_concat_f32(
     const struct ggml_compute_params * params,
     struct ggml_tensor * dst) {
@@ -6655,7 +6784,7 @@ static void ggml_compute_forward_concat_f32(
     const struct ggml_tensor * src0 = dst->src[0];
     const struct ggml_tensor * src1 = dst->src[1];
 
-    GGML_ASSERT(src0->nb[0] == sizeof(float));
+    GGML_ASSERT(ggml_type_size(src0->type) == sizeof(float));
 
     const int ith = params->ith;
     const int nth = params->nth;
@@ -6698,6 +6827,16 @@ static void ggml_compute_forward_concat(
     const struct ggml_tensor * src0 = dst->src[0];
 
     switch (src0->type) {
+        case GGML_TYPE_F16:
+        case GGML_TYPE_BF16:
+        case GGML_TYPE_I16:
+            {
+                ggml_compute_forward_concat_f16(params, dst);
+            } break;
+        case GGML_TYPE_I8:
+            {
+                ggml_compute_forward_concat_i8(params, dst);
+            } break;
         case GGML_TYPE_F32:
         case GGML_TYPE_I32:
             {
@@ -6705,7 +6844,7 @@ static void ggml_compute_forward_concat(
             } break;
         default:
             {
-                GGML_ABORT("fatal error");
+                ggml_compute_forward_concat_any(params, dst);
             }
     }
 }
@@ -15440,6 +15579,14 @@ int ggml_cpu_has_amx_int8(void) {
 #endif
 }
 
+int ggml_cpu_has_bmi2(void) {
+#if defined(__BMI2__)
+    return 1;
+#else
+    return 0;
+#endif
+}
+
 int ggml_cpu_has_fma(void) {
 #if defined(__FMA__)
     return 1;

ggml/src/ggml-cpu/ggml-cpu.cpp

@@ -511,6 +511,9 @@ static ggml_backend_feature * ggml_backend_cpu_get_features(ggml_backend_reg_t r
         if (ggml_cpu_has_fma()) {
             features.push_back({ "FMA", "1" });
         }
+        if (ggml_cpu_has_bmi2()) {
+            features.push_back({ "BMI2", "1" });
+        }
         if (ggml_cpu_has_avx512()) {
             features.push_back({ "AVX512", "1" });
         }

ggml/src/ggml-cuda/common.cuh

@@ -395,11 +395,11 @@ static __device__ __forceinline__ uint32_t __hgt2_mask(const half2 a, const half
 
 static __device__ __forceinline__ int ggml_cuda_dp4a(const int a, const int b, int c) {
 #if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)
-#if defined(__gfx906__) || defined(__gfx908__) || defined(__gfx90a__) || defined(RDNA2)
+#if defined(CDNA) || defined(RDNA2) || defined(__gfx906__)
     c = __builtin_amdgcn_sdot4(a, b, c, false);
 #elif defined(RDNA3)
     c = __builtin_amdgcn_sudot4( true, a, true, b, c, false);
-#elif defined(__gfx1010__) || defined(__gfx900__)
+#elif defined(RDNA1) || defined(__gfx900__)
     int tmp1;
     int tmp2;
     asm("\n \

ggml/src/ggml-cuda/fattn-common.cuh

@@ -52,12 +52,11 @@ typedef half (*vec_dot_KQ_f16_t)(
 typedef float (*vec_dot_KQ_f32_t)(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8 , const void * __restrict__ Q_ds);
 
-template<typename T, int D>
+template<typename T, int D, int warp_size>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_0(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
 
     const block_q4_0 * K_q4_0 = (const block_q4_0 *) K_c;
-    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
     GGML_UNUSED(Q_v);
 
     T sum = 0.0f;
@@ -93,12 +92,11 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_0(
     return sum;
 }
 
-template<typename T, int D>
+template<typename T, int D, int warp_size>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_1(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
 
     const block_q4_1 * K_q4_1 = (const block_q4_1 *) K_c;
-    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
     GGML_UNUSED(Q_v);
 
     T sum = 0.0f;
@@ -138,12 +136,11 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q4_1(
     return sum;
 }
 
-template<typename T, int D>
+template<typename T, int D, int warp_size>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_0(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
 
     const block_q5_0 * K_q5_0 = (const block_q5_0 *) K_c;
-    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
     GGML_UNUSED(Q_v);
 
     T sum = 0.0f;
@@ -186,12 +183,11 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_0(
     return sum;
 }
 
-template<typename T, int D>
+template<typename T, int D, int warp_size>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
 
     const block_q5_1 * K_q5_1 = (const block_q5_1 *) K_c;
-    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
     GGML_UNUSED(Q_v);
 
     T sum = 0.0f;
@@ -238,12 +234,11 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(
     return sum;
 }
 
-template <typename T, int D>
+template <typename T, int D, int warp_size>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q8_0(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
 
     const block_q8_0 * K_q8_0 = (const block_q8_0 *) K_c;
-    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
     GGML_UNUSED(Q_v);
 
     T sum = 0.0f;
@@ -272,12 +267,11 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q8_0(
     return sum;
 }
 
-template <typename T, int D>
+template <typename T, int D, int warp_size>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_f16(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8 , const void * __restrict__ Q_ds_v) {
 
     const half2 * K_h2 = (const half2 *) K_c;
-    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
     GGML_UNUSED(Q_q8);
     GGML_UNUSED(Q_ds_v);
 
@@ -480,25 +474,25 @@ static __device__ __forceinline__ T dequantize_1_f16(const void * __restrict__ v
     return x[i];
 }
 
-template <int D>
+template <int D, int warp_size = WARP_SIZE>
 constexpr __device__ vec_dot_KQ_f16_t get_vec_dot_KQ_f16(ggml_type type_K) {
-    return type_K == GGML_TYPE_Q4_0 ? vec_dot_fattn_vec_KQ_q4_0<half, D> :
-        type_K == GGML_TYPE_Q4_1 ? vec_dot_fattn_vec_KQ_q4_1<half, D> :
-        type_K == GGML_TYPE_Q5_0 ? vec_dot_fattn_vec_KQ_q5_0<half, D> :
-        type_K == GGML_TYPE_Q5_1 ? vec_dot_fattn_vec_KQ_q5_1<half, D> :
-        type_K == GGML_TYPE_Q8_0 ? vec_dot_fattn_vec_KQ_q8_0<half, D> :
-        type_K == GGML_TYPE_F16 ? vec_dot_fattn_vec_KQ_f16<half, D> :
+    return type_K == GGML_TYPE_Q4_0 ? vec_dot_fattn_vec_KQ_q4_0<half, D, warp_size> :
+        type_K == GGML_TYPE_Q4_1 ? vec_dot_fattn_vec_KQ_q4_1<half, D, warp_size> :
+        type_K == GGML_TYPE_Q5_0 ? vec_dot_fattn_vec_KQ_q5_0<half, D, warp_size> :
+        type_K == GGML_TYPE_Q5_1 ? vec_dot_fattn_vec_KQ_q5_1<half, D, warp_size> :
+        type_K == GGML_TYPE_Q8_0 ? vec_dot_fattn_vec_KQ_q8_0<half, D, warp_size> :
+        type_K == GGML_TYPE_F16 ? vec_dot_fattn_vec_KQ_f16<half, D, warp_size> :
         nullptr;
 }
 
-template <int D>
+template <int D, int warp_size = WARP_SIZE>
 constexpr __device__ vec_dot_KQ_f32_t get_vec_dot_KQ_f32(ggml_type type_K) {
-    return type_K == GGML_TYPE_Q4_0 ? vec_dot_fattn_vec_KQ_q4_0<float, D> :
-        type_K == GGML_TYPE_Q4_1 ? vec_dot_fattn_vec_KQ_q4_1<float, D> :
-        type_K == GGML_TYPE_Q5_0 ? vec_dot_fattn_vec_KQ_q5_0<float, D> :
-        type_K == GGML_TYPE_Q5_1 ? vec_dot_fattn_vec_KQ_q5_1<float, D> :
-        type_K == GGML_TYPE_Q8_0 ? vec_dot_fattn_vec_KQ_q8_0<float, D> :
-        type_K == GGML_TYPE_F16 ? vec_dot_fattn_vec_KQ_f16<float, D> :
+    return type_K == GGML_TYPE_Q4_0 ? vec_dot_fattn_vec_KQ_q4_0<float, D, warp_size> :
+        type_K == GGML_TYPE_Q4_1 ? vec_dot_fattn_vec_KQ_q4_1<float, D, warp_size> :
+        type_K == GGML_TYPE_Q5_0 ? vec_dot_fattn_vec_KQ_q5_0<float, D, warp_size> :
+        type_K == GGML_TYPE_Q5_1 ? vec_dot_fattn_vec_KQ_q5_1<float, D, warp_size> :
+        type_K == GGML_TYPE_Q8_0 ? vec_dot_fattn_vec_KQ_q8_0<float, D, warp_size> :
+        type_K == GGML_TYPE_F16 ? vec_dot_fattn_vec_KQ_f16<float, D, warp_size> :
         nullptr;
 }
 
@@ -681,7 +675,8 @@ static void on_no_fattn_vec_case(const int D) {
 template <int D, int ncols1, int ncols2, int parallel_blocks, int KQ_stride>
 void launch_fattn(
     ggml_backend_cuda_context & ctx, ggml_tensor * dst, fattn_kernel_t fattn_kernel,
-    const int nwarps, const size_t nbytes_shared, const bool need_f16_K, const bool need_f16_V
+    const int nwarps, const size_t nbytes_shared, const bool need_f16_K, const bool need_f16_V,
+    const int warp_size = WARP_SIZE
 ) {
     constexpr int ncols = ncols1 * ncols2;
 
@@ -704,8 +699,6 @@ void launch_fattn(
 
     GGML_ASSERT(Q->ne[3] == 1);
 
-    const int warp_size = ggml_cuda_info().devices[ctx.device].warp_size;
-
     ggml_cuda_pool & pool = ctx.pool();
     cudaStream_t main_stream = ctx.stream();
     const int id  = ggml_cuda_get_device();
@@ -805,7 +798,6 @@ void launch_fattn(
     const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
 
     GGML_ASSERT(block_dim.x % warp_size == 0);
-    GGML_ASSERT(!GGML_CUDA_CC_IS_AMD(cc) || block_dim.x * block_dim.y <= 4 * (unsigned int)warp_size);
     fattn_kernel<<<blocks_num, block_dim, nbytes_shared, main_stream>>>(
         (const char *) Q->data,
         K_data,

ggml/src/ggml-cuda/fattn-wmma-f16.cu

@@ -469,6 +469,7 @@ void ggml_cuda_flash_attn_ext_wmma_f16_case(ggml_backend_cuda_context & ctx, ggm
     constexpr int frag_m = cols_per_block == 8 && D % 32 == 0 ? 32 : 16;
     const int blocks_num_pb1 = ((Q->ne[1] + cols_per_block - 1) / cols_per_block)*Q->ne[2]*Q->ne[3];
     const int nsm = ggml_cuda_info().devices[ggml_cuda_get_device()].nsm;
+    const int warp_size = ggml_cuda_info().devices[ggml_cuda_get_device()].warp_size;
 
     float logit_softcap;
     memcpy(&logit_softcap, (const float *) KQV->op_params + 2, sizeof(float));
@@ -485,7 +486,7 @@ void ggml_cuda_flash_attn_ext_wmma_f16_case(ggml_backend_cuda_context & ctx, ggm
             fattn_kernel = flash_attn_ext_f16<
                 D, cols_per_block, nwarps, get_VKQ_stride(D, nwarps, frag_m), parallel_blocks, KQ_acc_t, use_logit_softcap>;
         }
-        launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, 0, true, true);
+        launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, 0, true, true, warp_size);
         return;
     }
     if (2*blocks_num_pb1 < 2*nsm) {
@@ -500,7 +501,7 @@ void ggml_cuda_flash_attn_ext_wmma_f16_case(ggml_backend_cuda_context & ctx, ggm
             fattn_kernel = flash_attn_ext_f16<
                 D, cols_per_block, nwarps, get_VKQ_stride(D, nwarps, frag_m), parallel_blocks, KQ_acc_t, use_logit_softcap>;
         }
-        launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, 0, true, true);
+        launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, 0, true, true, warp_size);
         return;
     }
     constexpr int parallel_blocks = 1;
@@ -514,7 +515,7 @@ void ggml_cuda_flash_attn_ext_wmma_f16_case(ggml_backend_cuda_context & ctx, ggm
         fattn_kernel = flash_attn_ext_f16<
             D, cols_per_block, nwarps, get_VKQ_stride(D, nwarps, frag_m), parallel_blocks, KQ_acc_t, use_logit_softcap>;
     }
-    launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, 0, true, true);
+    launch_fattn<D, cols_per_block, 1, parallel_blocks, -1>(ctx, dst, fattn_kernel, nwarps, 0, true, true, warp_size);
 }
 
 void ggml_cuda_flash_attn_ext_wmma_f16(ggml_backend_cuda_context & ctx, ggml_tensor * dst) {

ggml/src/ggml-cuda/fattn.cu

@@ -310,7 +310,7 @@ void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst
     }
 
     // The MMA implementation needs Turing or newer, use the old WMMA code for Volta:
-    if (cc == GGML_CUDA_CC_VOLTA) {
+    if (fp16_mma_available(cc) && !new_mma_available(cc)) {
         ggml_cuda_flash_attn_ext_wmma_f16(ctx, dst);
         return;
     }

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -2571,7 +2571,7 @@ static void maintain_cuda_graph(ggml_backend_cuda_context * cuda_ctx, std::vecto
         for (size_t i = 0; i < cuda_ctx->cuda_graph->num_nodes; i++) {
             if(count(ggml_cuda_cpy_fn_ptrs.begin(), ggml_cuda_cpy_fn_ptrs.end(), cuda_ctx->cuda_graph->params[i].func) > 0) {
                 char ** updated_kernel_arg_ptr = cuda_ctx->cuda_graph->updated_kernel_arg.at(k++);
-                cuda_ctx->cuda_graph->params[i].kernelParams[1] = updated_kernel_arg_ptr;
+                *(void**)cuda_ctx->cuda_graph->params[i].kernelParams[1] = *(void**)updated_kernel_arg_ptr;
                 CUDA_CHECK(cudaGraphKernelNodeSetParams(cuda_ctx->cuda_graph->nodes[i], &cuda_ctx->cuda_graph->params[i]));
             }
         }

ggml/src/ggml-cuda/mmvq.cu

@@ -47,11 +47,89 @@ static constexpr __device__ int get_vdr_mmvq(ggml_type type) {
         1;
 }
 
+enum mmvq_parameter_table_id {
+    MMVQ_PARAMETERS_GENERIC = 0,
+    MMVQ_PARAMETERS_GCN,
+    MMVQ_PARAMETERS_RDNA2
+};
+
+static constexpr __device__ mmvq_parameter_table_id get_device_table_id() {
+#if defined(RDNA2) || defined(RDNA3)
+    return MMVQ_PARAMETERS_RDNA2;
+#elif defined(GCN) || defined(CDNA)
+    return MMVQ_PARAMETERS_GCN;
+#else
+    return MMVQ_PARAMETERS_GENERIC;
+#endif
+}
+
+static __host__ mmvq_parameter_table_id get_device_table_id(int cc) {
+    if (GGML_CUDA_CC_IS_RDNA2(cc) || GGML_CUDA_CC_IS_RDNA3(cc)) {
+        return MMVQ_PARAMETERS_RDNA2;
+    }
+    if (GGML_CUDA_CC_IS_GCN(cc) || GGML_CUDA_CC_IS_CDNA(cc)) {
+        return MMVQ_PARAMETERS_GCN;
+    }
+    return MMVQ_PARAMETERS_GENERIC;
+}
+
+static constexpr __host__ __device__ int calc_nwarps(int ncols_y,  mmvq_parameter_table_id table_id) {
+    if (table_id == MMVQ_PARAMETERS_GENERIC) {
+        switch (ncols_y) {
+            case 1:
+            case 2:
+            case 3:
+            case 4:
+                return 4;
+            case 5:
+            case 6:
+            case 7:
+            case 8:
+                return 2;
+            default:
+                return 1;
+        }
+    } else if (table_id == MMVQ_PARAMETERS_GCN) {
+        switch (ncols_y) {
+            case 1:
+            case 2:
+            case 3:
+            case 4:
+                return 2;
+            case 5:
+            case 6:
+            case 7:
+            case 8:
+            default:
+                return 1;
+        }
+    }
+    return 1;
+}
+
+static constexpr __host__ __device__ int calc_rows_per_block(int ncols_y, int table_id) {
+    if (table_id == MMVQ_PARAMETERS_GENERIC || table_id == MMVQ_PARAMETERS_GCN) {
+        switch (ncols_y) {
+            case 1:
+                return 1;
+            case 2:
+            case 3:
+            case 4:
+            case 5:
+            case 6:
+            case 7:
+            case 8:
+                return 2;
+            default:
+                return 1;
+        }
+    }
+    return 1;
+}
+
 template <ggml_type type, int ncols_y>
-#if !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))
 // tell the compiler to use as many registers as it wants, see nwarps definition below
-__launch_bounds__((ncols_y <= 4 ? 4 : 2)*WARP_SIZE, 1)
-#endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))
+__launch_bounds__(calc_nwarps(ncols_y, get_device_table_id())*ggml_cuda_get_physical_warp_size(), 1)
 static __global__ void mul_mat_vec_q(
     const void * __restrict__ vx, const void * __restrict__ vy, float * __restrict__ dst,
     const int ncols_x, const int nrows_x, const int nrows_y, const int nrows_dst) {
@@ -59,24 +137,20 @@ static __global__ void mul_mat_vec_q(
     constexpr int qk  = ggml_cuda_type_traits<type>::qk;
     constexpr int qi  = ggml_cuda_type_traits<type>::qi;
     constexpr int vdr = get_vdr_mmvq(type);
+    constexpr mmvq_parameter_table_id table_id = get_device_table_id();
+    constexpr int nwarps = calc_nwarps(ncols_y, table_id);
+    constexpr int rows_per_cuda_block = calc_rows_per_block(ncols_y, table_id);
+    constexpr int warp_size = ggml_cuda_get_physical_warp_size();
 
     constexpr vec_dot_q_cuda_t vec_dot_q_cuda = get_vec_dot_q_cuda(type);
 
-#if defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && (defined(RDNA2) || defined(RDNA3))
-    constexpr int nwarps              = 1;
-    constexpr int rows_per_cuda_block = 1;
-#else
-    constexpr int nwarps              = ncols_y <= 4 ? 4 : 2;
-    constexpr int rows_per_cuda_block = ncols_y == 1 ? 1 : 2;
-#endif // defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__) && !defined(RDNA2) && !defined(RDNA3)
-
-    const     int tid = WARP_SIZE*threadIdx.y + threadIdx.x;
+    const     int tid = warp_size*threadIdx.y + threadIdx.x;
     const     int row0 = rows_per_cuda_block*blockIdx.x;
     const     int blocks_per_row_x = ncols_x / qk;
     const     int blocks_per_col_y = nrows_y / QK8_1;
-    constexpr int blocks_per_iter = vdr * nwarps*WARP_SIZE / qi;
+    constexpr int blocks_per_iter = vdr * nwarps*warp_size / qi;
 
-// partial sum for each thread
+    // partial sum for each thread
     float tmp[ncols_y][rows_per_cuda_block] = {0.0f};
 
     const block_q8_1 * y = (const block_q8_1 *) vy;
@@ -96,7 +170,7 @@ static __global__ void mul_mat_vec_q(
         }
     }
 
-    __shared__ float tmp_shared[nwarps-1 > 0 ? nwarps-1 : 1][ncols_y][rows_per_cuda_block][WARP_SIZE];
+    __shared__ float tmp_shared[nwarps-1 > 0 ? nwarps-1 : 1][ncols_y][rows_per_cuda_block][warp_size];
     if (threadIdx.y > 0) {
 #pragma unroll
         for (int j = 0; j < ncols_y; ++j) {
@@ -120,7 +194,7 @@ static __global__ void mul_mat_vec_q(
             for (int l = 0; l < nwarps-1; ++l) {
                 tmp[j][i] += tmp_shared[l][j][i][threadIdx.x];
             }
-            tmp[j][i] = warp_reduce_sum(tmp[j][i]);
+            tmp[j][i] = warp_reduce_sum<warp_size>(tmp[j][i]);
         }
 
         if (threadIdx.x < rows_per_cuda_block && (rows_per_cuda_block == 1 || row0 + threadIdx.x < nrows_dst)) {
@@ -129,6 +203,13 @@ static __global__ void mul_mat_vec_q(
     }
 }
 
+static std::pair<dim3, dim3> calc_launch_params(const int ncols_y, const int nrows_x, const int warp_size, const mmvq_parameter_table_id table_id) {
+    const int64_t nblocks = (nrows_x + calc_rows_per_block(ncols_y, table_id) - 1) / calc_rows_per_block(ncols_y, table_id);
+    const dim3 block_nums(nblocks, 1, 1);
+    const dim3 block_dims(warp_size, calc_nwarps(ncols_y, table_id), 1);
+    return {block_nums, block_dims};
+}
+
 template <ggml_type type>
 static void mul_mat_vec_q_cuda(
     const void * vx, const void * vy, float * dst,
@@ -137,65 +218,67 @@ static void mul_mat_vec_q_cuda(
     GGML_ASSERT(ncols_x % ggml_blck_size(type) == 0);
     GGML_ASSERT(ncols_y <= MMVQ_MAX_BATCH_SIZE);
 
-    int id = ggml_cuda_get_device();
-
-    int64_t nwarps = 1;
-    int64_t rows_per_cuda_block = 1;
-
-    if (ggml_cuda_info().devices[id].cc < GGML_CUDA_CC_RDNA2) { // NVIDIA and AMD older than RDNA2
-        switch(ncols_y) {
-            case 1:
-                nwarps = 4;
-                rows_per_cuda_block = 1;
-                break;
-            case 2:
-            case 3:
-            case 4:
-                nwarps = 4;
-                rows_per_cuda_block = 2;
-                break;
-            case 5:
-            case 6:
-            case 7:
-            case 8:
-                nwarps = 2;
-                rows_per_cuda_block = 2;
-                break;
-            default:
-                GGML_ABORT("fatal error");
-                break;
-        }
-    }
-
-    const int64_t nblocks = (nrows_x + rows_per_cuda_block - 1) / rows_per_cuda_block;
-    const dim3 block_nums(nblocks, 1, 1);
-    const dim3 block_dims(WARP_SIZE, nwarps, 1);
+    const int device = ggml_cuda_get_device();
+    const int warp_size = ggml_cuda_info().devices[device].warp_size;
+    const mmvq_parameter_table_id table_id = get_device_table_id(ggml_cuda_info().devices[device].cc);
 
     switch (ncols_y) {
         case 1:
-            mul_mat_vec_q<type, 1><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 1;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 2:
-            mul_mat_vec_q<type, 2><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 2;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 3:
-            mul_mat_vec_q<type, 3><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 3;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 4:
-            mul_mat_vec_q<type, 4><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 4;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 5:
-            mul_mat_vec_q<type, 5><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 5;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 6:
-            mul_mat_vec_q<type, 6><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 6;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 7:
-            mul_mat_vec_q<type, 7><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 7;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         case 8:
-            mul_mat_vec_q<type, 8><<<block_nums, block_dims, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
+        {
+            constexpr int c_ncols_y = 8;
+            std::pair<dim3, dim3> dims = calc_launch_params(c_ncols_y, nrows_x, warp_size, table_id);
+            mul_mat_vec_q<type, c_ncols_y><<<dims.first, dims.second, 0, stream>>>(vx, vy, dst, ncols_x, nrows_x, nrows_y, nrows_dst);
             break;
+        }
         default:
             GGML_ABORT("fatal error");
             break;

ggml/src/ggml-metal/CMakeLists.txt

@@ -27,12 +27,12 @@ configure_file(../ggml-common.h  ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h
 configure_file(ggml-metal.metal  ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal  COPYONLY)
 configure_file(ggml-metal-impl.h ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal-impl.h COPYONLY)
 
+set(METALLIB_COMMON "${CMAKE_CURRENT_SOURCE_DIR}/../ggml-common.h")
 if (GGML_METAL_EMBED_LIBRARY)
     enable_language(ASM)
 
     add_compile_definitions(GGML_METAL_EMBED_LIBRARY)
 
-    set(METALLIB_COMMON "${CMAKE_CURRENT_SOURCE_DIR}/../ggml-common.h")
     set(METALLIB_SOURCE "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal.metal")
     set(METALLIB_IMPL   "${CMAKE_CURRENT_SOURCE_DIR}/ggml-metal-impl.h")
 
@@ -88,12 +88,11 @@ else()
 
     add_custom_command(
         OUTPUT ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
-        COMMAND xcrun -sdk macosx metal    ${XC_FLAGS} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
-        COMMAND xcrun -sdk macosx metallib                ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air   -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
-        COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.air
+        COMMAND xcrun -sdk macosx metal ${XC_FLAGS} -c ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal -o - |
+            xcrun -sdk macosx metallib - -o ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/default.metallib
         COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-common.h
         COMMAND rm -f ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/ggml-metal.metal
-        DEPENDS ggml-metal.metal ggml-common.h
+        DEPENDS ggml-metal.metal ${METALLIB_COMMON}
         COMMENT "Compiling Metal kernels"
         )
 

ggml/src/ggml-metal/ggml-metal-impl.h

@@ -285,4 +285,239 @@ typedef struct {
     float    eps;
 } ggml_metal_kargs_rms_norm;
 
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    int32_t  n_groups;
+    float    eps;
+} ggml_metal_kargs_group_norm;
+
+typedef struct {
+    int32_t  IC;
+    int32_t  IL;
+    int32_t  K;
+    int32_t  s0;
+    uint64_t nb0;
+    uint64_t nb1;
+} ggml_metal_kargs_conv_transpose_1d;
+
+typedef struct {
+    uint64_t  ofs0;
+    uint64_t  ofs1;
+    int32_t  IW;
+    int32_t  IH;
+    int32_t  CHW;
+    int32_t  s0;
+    int32_t  s1;
+    int32_t  p0;
+    int32_t  p1;
+    int32_t  d0;
+    int32_t  d1;
+    int32_t  N;
+    int32_t  KH;
+    int32_t  KW;
+    int32_t  KHW; // KH * KW, pre-computed on CPU to save GPU resources
+} ggml_metal_kargs_im2col;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    int64_t  ne03;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    uint64_t nb03;
+    int64_t  ne10;
+    int64_t  ne11;
+    int64_t  ne12;
+    int64_t  ne13;
+    uint64_t nb10;
+    uint64_t nb11;
+    uint64_t nb12;
+    uint64_t nb13;
+    int64_t  ne0;
+    int64_t  ne1;
+    int64_t  ne2;
+    int64_t  ne3;
+    uint64_t nb0;
+    uint64_t nb1;
+    uint64_t nb2;
+    uint64_t nb3;
+} ggml_metal_kargs_sum_rows;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    float    scale;
+    float    max_bias;
+    float    m0;
+    float    m1;
+    uint32_t n_head_log2;
+} ggml_metal_kargs_soft_max;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int      n_past;
+} ggml_metal_kargs_diag_mask_inf;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    int64_t  ne10;
+    int64_t  ne11;
+    uint64_t nb10;
+    uint64_t nb11;
+    int64_t  ne0;
+    int64_t  ne1;
+    int64_t  ne2;
+    uint64_t nb0;
+    uint64_t nb1;
+    uint64_t nb2;
+} ggml_metal_kargs_ssm_conv;
+
+typedef struct {
+    int64_t  d_state;
+    int64_t  d_inner;
+    int64_t  n_seq_tokens;
+    int64_t  n_seqs;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    uint64_t nb10;
+    uint64_t nb11;
+    uint64_t nb12;
+    uint64_t nb13;
+    uint64_t nb20;
+    uint64_t nb21;
+    uint64_t nb22;
+    uint64_t nb30;
+    uint64_t nb31;
+    uint64_t nb40;
+    uint64_t nb41;
+    uint64_t nb42;
+    uint64_t nb50;
+    uint64_t nb51;
+    uint64_t nb52;
+} ggml_metal_kargs_ssm_scan;
+
+typedef struct {
+    int64_t  ne00;
+    uint64_t nb01;
+    uint64_t nb02;
+    int64_t  ne10;
+    uint64_t nb10;
+    uint64_t nb11;
+    uint64_t nb1;
+    uint64_t nb2;
+} ggml_metal_kargs_get_rows;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    int64_t  ne03;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    uint64_t nb03;
+    int64_t  ne0;
+    int64_t  ne1;
+    int64_t  ne2;
+    int64_t  ne3;
+    uint64_t nb0;
+    uint64_t nb1;
+    uint64_t nb2;
+    uint64_t nb3;
+    float    sf0;
+    float    sf1;
+    float    sf2;
+    float    sf3;
+} ggml_metal_kargs_upscale;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    int64_t  ne03;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    uint64_t nb03;
+    int64_t  ne0;
+    int64_t  ne1;
+    int64_t  ne2;
+    int64_t  ne3;
+    uint64_t nb0;
+    uint64_t nb1;
+    uint64_t nb2;
+    uint64_t nb3;
+} ggml_metal_kargs_pad;
+
+typedef struct {
+    int64_t  ne00;
+    int64_t  ne01;
+    int64_t  ne02;
+    int64_t  ne03;
+    uint64_t nb00;
+    uint64_t nb01;
+    uint64_t nb02;
+    uint64_t nb03;
+    int64_t  ne0;
+    int64_t  ne1;
+    int64_t  ne2;
+    int64_t  ne3;
+    uint64_t nb0;
+    uint64_t nb1;
+    uint64_t nb2;
+    uint64_t nb3;
+    int32_t  p0;
+    int32_t  p1;
+} ggml_metal_kargs_pad_reflect_1d;
+
+typedef struct {
+    uint64_t nb1;
+    int      dim;
+    int      max_period;
+} ggml_metal_kargs_timestep_embedding;
+
+typedef struct {
+    float    slope;
+} ggml_metal_kargs_leaky_relu;
+
+typedef struct {
+    int64_t  ncols;
+    int64_t  ncols_pad;
+} ggml_metal_kargs_argsort;
+
+typedef struct {
+    int64_t  ne0;
+    float    start;
+    float    step;
+} ggml_metal_kargs_arange;
+
+typedef struct {
+    int32_t  k0;
+    int32_t  k1;
+    int32_t  s0;
+    int32_t  s1;
+    int32_t  p0;
+    int32_t  p1;
+    int64_t  IH;
+    int64_t  IW;
+    int64_t  OH;
+    int64_t  OW;
+    int64_t  parallel_elements;
+} ggml_metal_kargs_pool_2d;
+
 #endif // GGML_METAL_IMPL

ggml/src/ggml-metal/ggml-metal.m

@@ -46,6 +46,7 @@ static struct ggml_backend_device g_ggml_backend_metal_device;
 static struct ggml_backend_metal_device_context {
     id<MTLDevice> mtl_device;
     int           mtl_device_ref_count;
+    id<MTLLibrary> mtl_library;
 
     bool has_simdgroup_reduction;
     bool has_simdgroup_mm;
@@ -57,6 +58,7 @@ static struct ggml_backend_metal_device_context {
 } g_ggml_ctx_dev_main = {
     /*.mtl_device              =*/ nil,
     /*.mtl_device_ref_count    =*/ 0,
+    /*.mtl_library             =*/ nil,
     /*.has_simdgroup_reduction =*/ false,
     /*.has_simdgroup_mm        =*/ false,
     /*.has_residency_sets      =*/ false,
@@ -108,6 +110,11 @@ static void ggml_backend_metal_device_rel(struct ggml_backend_metal_device_conte
     ctx->mtl_device_ref_count--;
 
     if (ctx->mtl_device_ref_count == 0) {
+        if (ctx->mtl_library) {
+            [ctx->mtl_library release];
+            ctx->mtl_library = nil;
+        }
+
         if (ctx->mtl_device) {
             [ctx->mtl_device release];
             ctx->mtl_device = nil;
@@ -467,11 +474,13 @@ struct ggml_backend_metal_context {
 //       for now it is easier to work in a separate file
 // static NSString * const msl_library_source = @"see metal.metal";
 
+#if !GGML_METAL_EMBED_LIBRARY
 // Here to assist with NSBundle Path Hack
 @interface GGMLMetalClass : NSObject
 @end
 @implementation GGMLMetalClass
 @end
+#endif
 
 static void * ggml_metal_host_malloc(size_t n) {
     void * data = NULL;
@@ -493,6 +502,139 @@ static void * ggml_metal_host_malloc(size_t n) {
     return data;
 }
 
+// load library
+//
+// - first check if the library is embedded
+// - then check if the library is in the bundle
+// - if not found, load the source and compile it
+// - if that fails, return NULL
+static id<MTLLibrary> ggml_metal_load_library(id<MTLDevice> device, bool use_bfloat) {
+    id<MTLLibrary> metal_library = nil;
+    NSError * error = nil;
+    NSString * src = nil;
+
+#if GGML_METAL_EMBED_LIBRARY
+    GGML_LOG_INFO("%s: using embedded metal library\n", __func__);
+
+    extern const char ggml_metallib_start[];
+    extern const char ggml_metallib_end[];
+
+    src = [[NSString alloc] initWithBytes:ggml_metallib_start length:(ggml_metallib_end-ggml_metallib_start) encoding:NSUTF8StringEncoding];
+
+#else
+
+#ifdef SWIFT_PACKAGE
+    NSBundle * bundle = SWIFTPM_MODULE_BUNDLE;
+#else
+    NSBundle * bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
+#endif
+
+    NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
+    if (path_lib == nil) {
+        // Try to find the resource in the directory where the current binary located.
+        NSString * current_binary = [[NSProcessInfo processInfo] arguments][0];
+        NSString * bin_dir = [current_binary stringByDeletingLastPathComponent];
+        NSString * default_metallib_path = [NSString pathWithComponents:@[bin_dir, @"default.metallib"]];
+        if ([[NSFileManager defaultManager] isReadableFileAtPath:default_metallib_path]) {
+            GGML_LOG_INFO("%s: found '%s'\n", __func__, [default_metallib_path UTF8String]);
+            NSDictionary * atts = [[NSFileManager defaultManager] attributesOfItemAtPath:default_metallib_path error:&error];
+            if (atts && atts[NSFileType] == NSFileTypeSymbolicLink) {
+                // Optionally, if this is a symlink, try to resolve it.
+                default_metallib_path = [[NSFileManager defaultManager] destinationOfSymbolicLinkAtPath:default_metallib_path error:&error];
+                if (default_metallib_path && [default_metallib_path length] > 0 && ![[default_metallib_path substringToIndex:1] isEqualToString:@"/"]) {
+                    // It is a relative path, adding the binary directory as directory prefix.
+                    default_metallib_path = [NSString pathWithComponents:@[bin_dir, default_metallib_path]];
+                }
+                if (!default_metallib_path || ![[NSFileManager defaultManager] isReadableFileAtPath:default_metallib_path]) {
+                    // Link to the resource could not be resolved.
+                    default_metallib_path = nil;
+                } else {
+                    GGML_LOG_INFO("%s: symlink resolved '%s'\n", __func__, [default_metallib_path UTF8String]);
+                }
+            }
+        } else {
+            // The resource couldn't be found in the binary's directory.
+            default_metallib_path = nil;
+        }
+        path_lib = default_metallib_path;
+    }
+
+    if (path_lib != nil) {
+        // pre-compiled library found
+        NSURL * libURL = [NSURL fileURLWithPath:path_lib];
+        GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_lib UTF8String]);
+
+        metal_library = [device newLibraryWithURL:libURL error:&error];
+        if (error) {
+            GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
+            return NULL;
+        }
+    } else {
+        GGML_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
+
+        NSString * path_source;
+        NSString * path_resource = [[NSProcessInfo processInfo].environment objectForKey:@"GGML_METAL_PATH_RESOURCES"];
+
+        GGML_LOG_INFO("%s: GGML_METAL_PATH_RESOURCES = %s\n", __func__, path_resource ? [path_resource UTF8String] : "nil");
+
+        if (path_resource) {
+            path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal.metal"];
+        } else {
+            path_source = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
+        }
+
+        if (path_source == nil) {
+            GGML_LOG_WARN("%s: error: could not use bundle path to find ggml-metal.metal, falling back to trying cwd\n", __func__);
+            path_source = @"ggml-metal.metal";
+        }
+
+        GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_source UTF8String]);
+
+        src = [NSString stringWithContentsOfFile:path_source encoding:NSUTF8StringEncoding error:&error];
+        if (error) {
+            GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
+            return NULL;
+        }
+    }
+#endif
+
+    if (!metal_library) {
+        @autoreleasepool {
+            // dictionary of preprocessor macros
+            NSMutableDictionary * prep = [NSMutableDictionary dictionary];
+
+            if (use_bfloat) {
+                [prep setObject:@"1" forKey:@"GGML_METAL_USE_BF16"];
+            }
+
+#if GGML_METAL_EMBED_LIBRARY
+            [prep setObject:@"1" forKey:@"GGML_METAL_EMBED_LIBRARY"];
+#endif
+
+            MTLCompileOptions * options = [MTLCompileOptions new];
+            options.preprocessorMacros = prep;
+
+            //[options setFastMathEnabled:false];
+
+            metal_library = [device newLibraryWithSource:src options:options error:&error];
+            if (error) {
+                GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
+                return NULL;
+            }
+
+#if !__has_feature(objc_arc)
+            [options release];
+#endif
+        }
+    }
+
+#if GGML_METAL_EMBED_LIBRARY
+    [src release];
+#endif // GGML_METAL_EMBED_LIBRARY
+
+    return metal_library;
+}
+
 static struct ggml_backend_metal_context * ggml_metal_init(ggml_backend_dev_t dev) {
     GGML_LOG_INFO("%s: allocating\n", __func__);
 
@@ -520,137 +662,14 @@ static struct ggml_backend_metal_context * ggml_metal_init(ggml_backend_dev_t de
 
     ctx->d_queue = dispatch_queue_create("ggml-metal", DISPATCH_QUEUE_CONCURRENT);
 
-    id<MTLLibrary> metal_library;
-
     // load library
-    //
-    // - first check if the library is embedded
-    // - then check if the library is in the bundle
-    // - if not found, load the source and compile it
-    // - if that fails, return NULL
-    {
-        NSBundle * bundle = nil;
-#ifdef SWIFT_PACKAGE
-        bundle = SWIFTPM_MODULE_BUNDLE;
-#else
-        bundle = [NSBundle bundleForClass:[GGMLMetalClass class]];
-#endif
-
-        NSError * error = nil;
-
-#if GGML_METAL_EMBED_LIBRARY
-        const bool try_metallib = false;
-#else
-        const bool try_metallib = true;
-#endif
-
-        NSString * path_lib = [bundle pathForResource:@"default" ofType:@"metallib"];
-        if (path_lib == nil) {
-            // Try to find the resource in the directory where the current binary located.
-            NSString * current_binary = [[NSProcessInfo processInfo] arguments][0];
-            NSString * bin_dir = [current_binary stringByDeletingLastPathComponent];
-            NSString * default_metallib_path = [NSString pathWithComponents:@[bin_dir, @"default.metallib"]];
-            if ([[NSFileManager defaultManager] isReadableFileAtPath:default_metallib_path]) {
-                GGML_LOG_INFO("%s: found '%s'\n", __func__, [default_metallib_path UTF8String]);
-                NSDictionary * atts = [[NSFileManager defaultManager] attributesOfItemAtPath:default_metallib_path error:&error];
-                if (atts && atts[NSFileType] == NSFileTypeSymbolicLink) {
-                    // Optionally, if this is a symlink, try to resolve it.
-                    default_metallib_path = [[NSFileManager defaultManager] destinationOfSymbolicLinkAtPath:default_metallib_path error:&error];
-                    if (default_metallib_path && [default_metallib_path length] > 0 && ![[default_metallib_path substringToIndex:1] isEqualToString:@"/"]) {
-                        // It is a relative path, adding the binary directory as directory prefix.
-                        default_metallib_path = [NSString pathWithComponents:@[bin_dir, default_metallib_path]];
-                    }
-                    if (!default_metallib_path || ![[NSFileManager defaultManager] isReadableFileAtPath:default_metallib_path]) {
-                        // Link to the resource could not be resolved.
-                        default_metallib_path = nil;
-                    } else {
-                        GGML_LOG_INFO("%s: symlink resolved '%s'\n", __func__, [default_metallib_path UTF8String]);
-                    }
-                }
-            } else {
-                // The resource couldn't be found in the binary's directory.
-                default_metallib_path = nil;
-            }
-            path_lib = default_metallib_path;
-        }
-
-        if (try_metallib && path_lib != nil) {
-            // pre-compiled library found
-            NSURL * libURL = [NSURL fileURLWithPath:path_lib];
-            GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_lib UTF8String]);
-
-            metal_library = [device newLibraryWithURL:libURL error:&error];
-            if (error) {
-                GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
-                return NULL;
-            }
-        } else {
-#if GGML_METAL_EMBED_LIBRARY
-            GGML_LOG_INFO("%s: using embedded metal library\n", __func__);
-
-            extern const char ggml_metallib_start[];
-            extern const char ggml_metallib_end[];
-
-            NSString * src = [[NSString alloc] initWithBytes:ggml_metallib_start length:(ggml_metallib_end-ggml_metallib_start) encoding:NSUTF8StringEncoding];
-#else
-            GGML_LOG_INFO("%s: default.metallib not found, loading from source\n", __func__);
-
-            NSString * path_source;
-            NSString * path_resource = [[NSProcessInfo processInfo].environment objectForKey:@"GGML_METAL_PATH_RESOURCES"];
-
-            GGML_LOG_INFO("%s: GGML_METAL_PATH_RESOURCES = %s\n", __func__, path_resource ? [path_resource UTF8String] : "nil");
-
-            if (path_resource) {
-                path_source = [path_resource stringByAppendingPathComponent:@"ggml-metal.metal"];
-            } else {
-                path_source = [bundle pathForResource:@"ggml-metal" ofType:@"metal"];
-            }
-
-            if (path_source == nil) {
-                GGML_LOG_WARN("%s: error: could not use bundle path to find ggml-metal.metal, falling back to trying cwd\n", __func__);
-                path_source = @"ggml-metal.metal";
-            }
-
-            GGML_LOG_INFO("%s: loading '%s'\n", __func__, [path_source UTF8String]);
-
-            NSString * src = [NSString stringWithContentsOfFile:path_source encoding:NSUTF8StringEncoding error:&error];
-            if (error) {
-                GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
-                return NULL;
-            }
-#endif // GGML_METAL_EMBED_LIBRARY
-
-            @autoreleasepool {
-                // dictionary of preprocessor macros
-                NSMutableDictionary * prep = [NSMutableDictionary dictionary];
-
-                if (ctx_dev->use_bfloat) {
-                    [prep setObject:@"1" forKey:@"GGML_METAL_USE_BF16"];
-                }
-
-#if GGML_METAL_EMBED_LIBRARY
-                [prep setObject:@"1" forKey:@"GGML_METAL_EMBED_LIBRARY"];
-#endif
-
-                MTLCompileOptions * options = [MTLCompileOptions new];
-                options.preprocessorMacros = prep;
-
-                //[options setFastMathEnabled:false];
-
-                metal_library = [device newLibraryWithSource:src options:options error:&error];
-                if (error) {
-                    GGML_LOG_ERROR("%s: error: %s\n", __func__, [[error description] UTF8String]);
-                    return NULL;
-                }
-
-#if !__has_feature(objc_arc)
-                [options release];
-#endif
-            }
-#if GGML_METAL_EMBED_LIBRARY
-            [src release];
-#endif // GGML_METAL_EMBED_LIBRARY
-        }
+    if (ctx_dev->mtl_library == nil) {
+        ctx_dev->mtl_library = ggml_metal_load_library(device, ctx_dev->use_bfloat);
+    }
+    id<MTLLibrary> metal_library = ctx_dev->mtl_library;
+    if (metal_library == nil) {
+        GGML_LOG_ERROR("%s: error: metal library is nil\n", __func__);
+        return NULL;
     }
 
     // print MTL GPU family:
@@ -724,7 +743,6 @@ static struct ggml_backend_metal_context * ggml_metal_init(ggml_backend_dev_t de
             [metal_function release]; \
             if (error) { \
                 GGML_LOG_ERROR("%s: error: load pipeline error: %s\n", __func__, [[error description] UTF8String]); \
-                [metal_library release]; \
                 return NULL; \
             } \
         } else { \
@@ -1043,8 +1061,6 @@ static struct ggml_backend_metal_context * ggml_metal_init(ggml_backend_dev_t de
         GGML_METAL_ADD_KERNEL(GGML_METAL_KERNEL_TYPE_POOL_2D_MAX_F32,               pool_2d_max_f32,                true);
     }
 
-    [metal_library release];
-
     return ctx;
 }
 
@@ -1944,34 +1960,38 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SUM_ROWS].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+
+                ggml_metal_kargs_sum_rows args = {
+                   /*.ne00 =*/ ne00,
+                   /*.ne01 =*/ ne01,
+                   /*.ne02 =*/ ne02,
+                   /*.ne03 =*/ ne03,
+                   /*.nb00 =*/ nb00,
+                   /*.nb01 =*/ nb01,
+                   /*.nb02 =*/ nb02,
+                   /*.nb03 =*/ nb03,
+                   /*.ne10 =*/ ne10,
+                   /*.ne11 =*/ ne11,
+                   /*.ne12 =*/ ne12,
+                   /*.ne13 =*/ ne13,
+                   /*.nb10 =*/ nb10,
+                   /*.nb11 =*/ nb11,
+                   /*.nb12 =*/ nb12,
+                   /*.nb13 =*/ nb13,
+                   /*.ne0  =*/ ne0,
+                   /*.ne1  =*/ ne1,
+                   /*.ne2  =*/ ne2,
+                   /*.ne3  =*/ ne3,
+                   /*.nb0  =*/ nb0,
+                   /*.nb1  =*/ nb1,
+                   /*.nb2  =*/ nb2,
+                   /*.nb3  =*/ nb3,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
-                [encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
-                [encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
-                [encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
-                [encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
-                [encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
-                [encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
-                [encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
-                [encoder setBytes:&nb03 length:sizeof(nb03) atIndex:9];
-                [encoder setBytes:&ne10 length:sizeof(ne10) atIndex:10];
-                [encoder setBytes:&ne11 length:sizeof(ne11) atIndex:11];
-                [encoder setBytes:&ne12 length:sizeof(ne12) atIndex:12];
-                [encoder setBytes:&ne13 length:sizeof(ne13) atIndex:13];
-                [encoder setBytes:&nb10 length:sizeof(nb10) atIndex:14];
-                [encoder setBytes:&nb11 length:sizeof(nb11) atIndex:15];
-                [encoder setBytes:&nb12 length:sizeof(nb12) atIndex:16];
-                [encoder setBytes:&nb13 length:sizeof(nb13) atIndex:17];
-                [encoder setBytes:&ne0  length:sizeof(ne0)  atIndex:18];
-                [encoder setBytes:&ne1  length:sizeof(ne1)  atIndex:19];
-                [encoder setBytes:&ne2  length:sizeof(ne2)  atIndex:20];
-                [encoder setBytes:&ne3  length:sizeof(ne3)  atIndex:21];
-                [encoder setBytes:&nb0  length:sizeof(nb0)  atIndex:22];
-                [encoder setBytes:&nb1  length:sizeof(nb1)  atIndex:23];
-                [encoder setBytes:&nb2  length:sizeof(nb2)  atIndex:24];
-                [encoder setBytes:&nb3  length:sizeof(nb3)  atIndex:25];
+                [encoder setBytes:&args length:sizeof(args) atIndex:2];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(ne01, ne02, ne03) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
             } break;
@@ -2020,8 +2040,17 @@ static void ggml_metal_encode_node(
                 const float m0 = powf(2.0f, -(max_bias       ) / n_head_log2);
                 const float m1 = powf(2.0f, -(max_bias / 2.0f) / n_head_log2);
 
-                // TODO: add ggml_metal_kargs struct
-                // TODO: optimize (see https://github.com/ggml-org/llama.cpp/pull/10238/commits/7941b6b9ec29a2866fec6fa6c51612515ca509f6)
+                ggml_metal_kargs_soft_max args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.ne02 =*/ ne02,
+                    /*.scale =*/ scale,
+                    /*.max_bias =*/ max_bias,
+                    /*.m0 =*/ m0,
+                    /*.m1 =*/ m1,
+                    /*.n_head_log2 =*/ n_head_log2,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0   atIndex:0];
                 if (id_src1) {
@@ -2030,14 +2059,7 @@ static void ggml_metal_encode_node(
                     [encoder setBuffer:id_src0 offset:offs_src0   atIndex:1];
                 }
                 [encoder setBuffer:id_dst      offset:offs_dst            atIndex:2];
-                [encoder setBytes:&ne00        length:sizeof(ne00)        atIndex:3];
-                [encoder setBytes:&ne01        length:sizeof(ne01)        atIndex:4];
-                [encoder setBytes:&ne02        length:sizeof(ne02)        atIndex:5];
-                [encoder setBytes:&scale       length:sizeof(scale)       atIndex:6];
-                [encoder setBytes:&max_bias    length:sizeof(max_bias)    atIndex:7];
-                [encoder setBytes:&m0          length:sizeof(m0)          atIndex:8];
-                [encoder setBytes:&m1          length:sizeof(m1)          atIndex:9];
-                [encoder setBytes:&n_head_log2 length:sizeof(n_head_log2) atIndex:10];
+                [encoder setBytes:&args        length:sizeof(args)        atIndex:3];
 
                 [encoder setThreadgroupMemoryLength:32*sizeof(float) atIndex:0];
 
@@ -2055,13 +2077,16 @@ static void ggml_metal_encode_node(
                     pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_DIAG_MASK_INF].pipeline;
                 }
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_diag_mask_inf args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.n_past =*/ n_past,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
-                [encoder setBytes:&ne00   length:sizeof(ne00) atIndex:2];
-                [encoder setBytes:&ne01   length:sizeof(ne01) atIndex:3];
-                [encoder setBytes:&n_past length:sizeof(int)  atIndex:4];
+                [encoder setBytes:&args  length:sizeof(args) atIndex:2];
 
                 if (ne00%8 == 0) {
                     [encoder dispatchThreadgroups:MTLSizeMake(ne00*ne01*ne02/8, 1, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
@@ -2080,27 +2105,30 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SSM_CONV_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_ssm_conv args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.ne02 =*/ ne02,
+                    /*.nb00 =*/ nb00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.ne10 =*/ ne10,
+                    /*.ne11 =*/ ne11,
+                    /*.nb10 =*/ nb10,
+                    /*.nb11 =*/ nb11,
+                    /*.ne0  =*/ ne0,
+                    /*.ne1  =*/ ne1,
+                    /*.ne2  =*/ ne2,
+                    /*.nb0  =*/ nb0,
+                    /*.nb1  =*/ nb1,
+                    /*.nb2  =*/ nb2,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0    atIndex:0];
                 [encoder setBuffer:id_src1 offset:offs_src1    atIndex:1];
                 [encoder setBuffer:id_dst  offset:offs_dst     atIndex:2];
-                [encoder setBytes:&ne00    length:sizeof(ne00) atIndex:3];
-                [encoder setBytes:&ne01    length:sizeof(ne01) atIndex:4];
-                [encoder setBytes:&ne02    length:sizeof(ne02) atIndex:5];
-                [encoder setBytes:&nb00    length:sizeof(nb00) atIndex:6];
-                [encoder setBytes:&nb01    length:sizeof(nb01) atIndex:7];
-                [encoder setBytes:&nb02    length:sizeof(nb02) atIndex:8];
-                [encoder setBytes:&ne10    length:sizeof(ne10) atIndex:9];
-                [encoder setBytes:&ne11    length:sizeof(ne11) atIndex:10];
-                [encoder setBytes:&nb10    length:sizeof(nb10) atIndex:11];
-                [encoder setBytes:&nb11    length:sizeof(nb11) atIndex:12];
-                [encoder setBytes:&ne0     length:sizeof(ne0)  atIndex:13];
-                [encoder setBytes:&ne1     length:sizeof(ne1)  atIndex:14];
-                [encoder setBytes:&ne2     length:sizeof(ne2)  atIndex:15];
-                [encoder setBytes:&nb0     length:sizeof(nb0)  atIndex:16];
-                [encoder setBytes:&nb1     length:sizeof(nb1)  atIndex:17];
-                [encoder setBytes:&nb2     length:sizeof(nb2)  atIndex:18];
+                [encoder setBytes:&args    length:sizeof(args) atIndex:3];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(ne01, ne1, ne02) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
             } break;
@@ -2151,7 +2179,31 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_SSM_SCAN_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_ssm_scan args = {
+                    /*.d_state =*/ d_state,
+                    /*.d_inner =*/ d_inner,
+                    /*.n_seq_tokens =*/ n_seq_tokens,
+                    /*.n_seqs =*/ n_seqs,
+                    /*.nb00 =*/ nb00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.nb10 =*/ nb10,
+                    /*.nb11 =*/ nb11,
+                    /*.nb12 =*/ nb12,
+                    /*.nb13 =*/ nb13,
+                    /*.nb20 =*/ nb20,
+                    /*.nb21 =*/ nb21,
+                    /*.nb22 =*/ nb22,
+                    /*.nb30 =*/ nb30,
+                    /*.nb31 =*/ nb31,
+                    /*.nb40 =*/ nb40,
+                    /*.nb41 =*/ nb41,
+                    /*.nb42 =*/ nb42,
+                    /*.nb50 =*/ nb50,
+                    /*.nb51 =*/ nb51,
+                    /*.nb52 =*/ nb52,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_src1 offset:offs_src1 atIndex:1];
@@ -2160,30 +2212,7 @@ static void ggml_metal_encode_node(
                 [encoder setBuffer:id_src4 offset:offs_src4 atIndex:4];
                 [encoder setBuffer:id_src5 offset:offs_src5 atIndex:5];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:6];
-
-                [encoder setBytes:&d_state      length:sizeof(d_state)      atIndex:7];
-                [encoder setBytes:&d_inner      length:sizeof(d_inner)      atIndex:8];
-                [encoder setBytes:&n_seq_tokens length:sizeof(n_seq_tokens) atIndex:9];
-                [encoder setBytes:&n_seqs       length:sizeof(n_seqs)       atIndex:10];
-
-                [encoder setBytes:&nb00 length:sizeof(nb00) atIndex:11];
-                [encoder setBytes:&nb01 length:sizeof(nb01) atIndex:12];
-                [encoder setBytes:&nb02 length:sizeof(nb02) atIndex:13];
-                [encoder setBytes:&nb10 length:sizeof(nb10) atIndex:14];
-                [encoder setBytes:&nb11 length:sizeof(nb11) atIndex:15];
-                [encoder setBytes:&nb12 length:sizeof(nb12) atIndex:16];
-                [encoder setBytes:&nb13 length:sizeof(nb13) atIndex:17];
-                [encoder setBytes:&nb20 length:sizeof(nb20) atIndex:18];
-                [encoder setBytes:&nb21 length:sizeof(nb21) atIndex:19];
-                [encoder setBytes:&nb22 length:sizeof(nb22) atIndex:20];
-                [encoder setBytes:&nb30 length:sizeof(nb30) atIndex:21];
-                [encoder setBytes:&nb31 length:sizeof(nb31) atIndex:22];
-                [encoder setBytes:&nb40 length:sizeof(nb40) atIndex:23];
-                [encoder setBytes:&nb41 length:sizeof(nb41) atIndex:24];
-                [encoder setBytes:&nb42 length:sizeof(nb42) atIndex:25];
-                [encoder setBytes:&nb50 length:sizeof(nb50) atIndex:26];
-                [encoder setBytes:&nb51 length:sizeof(nb51) atIndex:27];
-                [encoder setBytes:&nb52 length:sizeof(nb52) atIndex:28];
+                [encoder setBytes:&args    length:sizeof(args) atIndex:7];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(d_inner, n_seqs, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
             } break;
@@ -3040,19 +3069,22 @@ static void ggml_metal_encode_node(
                     default: GGML_ABORT("not implemented");
                 }
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_get_rows args = {
+                    /*.ne00 =*/ ne00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.ne10 =*/ ne10,
+                    /*.nb10 =*/ nb10,
+                    /*.nb11 =*/ nb11,
+                    /*.nb1 =*/ nb1,
+                    /*.nb2 =*/ nb2,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0     offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_src1     offset:offs_src1 atIndex:1];
                 [encoder setBuffer:id_dst      offset:offs_dst  atIndex:2];
-                [encoder setBytes:&ne00 length:sizeof( int64_t) atIndex:3];
-                [encoder setBytes:&nb01 length:sizeof(uint64_t) atIndex:4];
-                [encoder setBytes:&nb02 length:sizeof(uint64_t) atIndex:5];
-                [encoder setBytes:&ne10 length:sizeof( int64_t) atIndex:6];
-                [encoder setBytes:&nb10 length:sizeof( int64_t) atIndex:7];
-                [encoder setBytes:&nb11 length:sizeof( int64_t) atIndex:8];
-                [encoder setBytes:&nb1  length:sizeof(uint64_t) atIndex:9];
-                [encoder setBytes:&nb2  length:sizeof(uint64_t) atIndex:10];
+                [encoder setBytes:&args length:sizeof(args) atIndex:3];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(ne10, ne11, 1) threadsPerThreadgroup:MTLSizeMake(32, 1, 1)];
             } break;
@@ -3109,18 +3141,21 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_GROUP_NORM].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_group_norm args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.ne02 =*/ ne02,
+                    /*.nb00 =*/ nb00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.n_groups =*/ n_groups,
+                    /*.eps =*/ eps,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0  offset:offs_src0        atIndex:0];
                 [encoder setBuffer:id_dst   offset:offs_dst         atIndex:1];
-                [encoder setBytes:&ne00     length:sizeof( int64_t) atIndex:2];
-                [encoder setBytes:&ne01     length:sizeof( int64_t) atIndex:3];
-                [encoder setBytes:&ne02     length:sizeof( int64_t) atIndex:4];
-                [encoder setBytes:&nb00     length:sizeof(uint64_t) atIndex:5];
-                [encoder setBytes:&nb01     length:sizeof(uint64_t) atIndex:6];
-                [encoder setBytes:&nb02     length:sizeof(uint64_t) atIndex:7];
-                [encoder setBytes:&n_groups length:sizeof( int32_t) atIndex:8];
-                [encoder setBytes:&eps      length:sizeof(   float) atIndex:9];
+                [encoder setBytes:&args     length:sizeof(args)     atIndex:2];
                 [encoder setThreadgroupMemoryLength:32*sizeof(float) atIndex:0];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(n_groups, 1, 1) threadsPerThreadgroup:MTLSizeMake(nth, 1, 1)];
@@ -3278,8 +3313,8 @@ static void ggml_metal_encode_node(
 
                 const int32_t CHW = IC * KH * KW;
 
-                const int32_t ofs0 = src1->nb[is_2D ? 3 : 2] / 4;
-                const int32_t ofs1 = src1->nb[is_2D ? 2 : 1] / 4;
+                const uint64_t ofs0 = src1->nb[is_2D ? 3 : 2] / 4;
+                const uint64_t ofs1 = src1->nb[is_2D ? 2 : 1] / 4;
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_IM2COL_F32].pipeline;
 
@@ -3301,27 +3336,30 @@ static void ggml_metal_encode_node(
                     default: GGML_ABORT("fatal error");
                 };
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_im2col args = {
+                    /*.ofs0 =*/ ofs0,
+                    /*.ofs1 =*/ ofs1,
+                    /*.IW   =*/ IW,
+                    /*.IH   =*/ IH,
+                    /*.CHW  =*/ CHW,
+                    /*.s0   =*/ s0,
+                    /*.s1   =*/ s1,
+                    /*.p0   =*/ p0,
+                    /*.p1   =*/ p1,
+                    /*.d0   =*/ d0,
+                    /*.d1   =*/ d1,
+                    /*.N    =*/ N,
+                    /*.KH   =*/ KH,
+                    /*.KW   =*/ KW,
+                    /*.KHW  =*/ KH * KW,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src1 offset:offs_src1       atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst        atIndex:1];
-                [encoder setBytes:&ofs0    length:sizeof(int32_t) atIndex:2];
-                [encoder setBytes:&ofs1    length:sizeof(int32_t) atIndex:3];
-                [encoder setBytes:&IW      length:sizeof(int32_t) atIndex:4];
-                [encoder setBytes:&IH      length:sizeof(int32_t) atIndex:5];
-                [encoder setBytes:&CHW     length:sizeof(int32_t) atIndex:6];
-                [encoder setBytes:&s0      length:sizeof(int32_t) atIndex:7];
-                [encoder setBytes:&s1      length:sizeof(int32_t) atIndex:8];
-                [encoder setBytes:&p0      length:sizeof(int32_t) atIndex:9];
-                [encoder setBytes:&p1      length:sizeof(int32_t) atIndex:10];
-                [encoder setBytes:&d0      length:sizeof(int32_t) atIndex:11];
-                [encoder setBytes:&d1      length:sizeof(int32_t) atIndex:12];
+                [encoder setBytes:&args length:sizeof(args)       atIndex:2];
 
                 if (is_gt_mttpt) {
-                    [encoder setBytes:&N   length:sizeof(int32_t) atIndex:13];
-                    [encoder setBytes:&KH  length:sizeof(int32_t) atIndex:14];
-                    [encoder setBytes:&KW  length:sizeof(int32_t) atIndex:15];
-
                     const uint64_t n_threads = MIN(pipeline.maxTotalThreadsPerThreadgroup, (uint64_t)N);
 
                     const int64_t  quotient  = N / n_threads + (N % n_threads > 0 ? 1 : 0);
@@ -3361,16 +3399,20 @@ static void ggml_metal_encode_node(
                     default: GGML_ABORT("fatal error");
                 };
 
+                ggml_metal_kargs_conv_transpose_1d args = {
+                    /*.IC =*/ IC,
+                    /*.IL =*/ IL,
+                    /*.K  =*/ K,
+                    /*.s0 =*/ s0,
+                    /*.nb0 =*/ nb0,
+                    /*.nb1 =*/ nb1,
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0         atIndex:0];
                 [encoder setBuffer:id_src1 offset:offs_src1         atIndex:1];
                 [encoder setBuffer:id_dst  offset:offs_dst          atIndex:2];
-                [encoder setBytes:&IC      length:sizeof( int32_t)  atIndex:3];
-                [encoder setBytes:&IL      length:sizeof( int32_t)  atIndex:4];
-                [encoder setBytes:&K       length:sizeof( int32_t)  atIndex:5];
-                [encoder setBytes:&s0      length:sizeof( int32_t)  atIndex:6];
-                [encoder setBytes:&nb0     length:sizeof(uint64_t)  atIndex:7];
-                [encoder setBytes:&nb1     length:sizeof(uint64_t)  atIndex:8];
+                [encoder setBytes:&args    length:sizeof(args)       atIndex:3];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(OL, OC, 1) threadsPerThreadgroup:MTLSizeMake(1, 1, 1)];
             } break;
@@ -3385,30 +3427,33 @@ static void ggml_metal_encode_node(
 
                 const id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_UPSCALE_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_upscale args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.ne02 =*/ ne02,
+                    /*.ne03 =*/ ne03,
+                    /*.nb00 =*/ nb00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.nb03 =*/ nb03,
+                    /*.ne0 =*/ ne0,
+                    /*.ne1 =*/ ne1,
+                    /*.ne2 =*/ ne2,
+                    /*.ne3 =*/ ne3,
+                    /*.nb0 =*/ nb0,
+                    /*.nb1 =*/ nb1,
+                    /*.nb2 =*/ nb2,
+                    /*.nb3 =*/ nb3,
+                    /*.sf0 =*/ sf0,
+                    /*.sf1 =*/ sf1,
+                    /*.sf2 =*/ sf2,
+                    /*.sf3 =*/ sf3
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
-                [encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
-                [encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
-                [encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
-                [encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
-                [encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
-                [encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
-                [encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
-                [encoder setBytes:&nb03 length:sizeof(nb03) atIndex:9];
-                [encoder setBytes:&ne0  length:sizeof(ne0)  atIndex:10];
-                [encoder setBytes:&ne1  length:sizeof(ne1)  atIndex:11];
-                [encoder setBytes:&ne2  length:sizeof(ne2)  atIndex:12];
-                [encoder setBytes:&ne3  length:sizeof(ne3)  atIndex:13];
-                [encoder setBytes:&nb0  length:sizeof(nb0)  atIndex:14];
-                [encoder setBytes:&nb1  length:sizeof(nb1)  atIndex:15];
-                [encoder setBytes:&nb2  length:sizeof(nb2)  atIndex:16];
-                [encoder setBytes:&nb3  length:sizeof(nb3)  atIndex:17];
-                [encoder setBytes:&sf0  length:sizeof(sf0)  atIndex:18];
-                [encoder setBytes:&sf1  length:sizeof(sf1)  atIndex:19];
-                [encoder setBytes:&sf2  length:sizeof(sf2)  atIndex:20];
-                [encoder setBytes:&sf3  length:sizeof(sf3)  atIndex:21];
+                [encoder setBytes:&args length:sizeof(args) atIndex:2];
 
                 const int nth = MIN((int) pipeline.maxTotalThreadsPerThreadgroup, ne0);
 
@@ -3420,26 +3465,29 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_PAD_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_pad args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.ne02 =*/ ne02,
+                    /*.ne03 =*/ ne03,
+                    /*.nb00 =*/ nb00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.nb03 =*/ nb03,
+                    /*.ne0 =*/ ne0,
+                    /*.ne1 =*/ ne1,
+                    /*.ne2 =*/ ne2,
+                    /*.ne3 =*/ ne3,
+                    /*.nb0 =*/ nb0,
+                    /*.nb1 =*/ nb1,
+                    /*.nb2 =*/ nb2,
+                    /*.nb3 =*/ nb3
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
-                [encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
-                [encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
-                [encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
-                [encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
-                [encoder setBytes:&nb00 length:sizeof(nb00) atIndex:6];
-                [encoder setBytes:&nb01 length:sizeof(nb01) atIndex:7];
-                [encoder setBytes:&nb02 length:sizeof(nb02) atIndex:8];
-                [encoder setBytes:&nb03 length:sizeof(nb03) atIndex:9];
-                [encoder setBytes:&ne0  length:sizeof(ne0)  atIndex:10];
-                [encoder setBytes:&ne1  length:sizeof(ne1)  atIndex:11];
-                [encoder setBytes:&ne2  length:sizeof(ne2)  atIndex:12];
-                [encoder setBytes:&ne3  length:sizeof(ne3)  atIndex:13];
-                [encoder setBytes:&nb0  length:sizeof(nb0)  atIndex:14];
-                [encoder setBytes:&nb1  length:sizeof(nb1)  atIndex:15];
-                [encoder setBytes:&nb2  length:sizeof(nb2)  atIndex:16];
-                [encoder setBytes:&nb3  length:sizeof(nb3)  atIndex:17];
+                [encoder setBytes:&args length:sizeof(args) atIndex:2];
 
                 const int nth = MIN(1024, ne0);
 
@@ -3454,24 +3502,31 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_PAD_REFLECT_1D_F32].pipeline;
 
+                ggml_metal_kargs_pad_reflect_1d args = {
+                    /*.ne00 =*/ ne00,
+                    /*.ne01 =*/ ne01,
+                    /*.ne02 =*/ ne02,
+                    /*.ne03 =*/ ne03,
+                    /*.nb00 =*/ nb00,
+                    /*.nb01 =*/ nb01,
+                    /*.nb02 =*/ nb02,
+                    /*.nb03 =*/ nb03,
+                    /*.ne0 =*/ ne0,
+                    /*.ne1 =*/ ne1,
+                    /*.ne2 =*/ ne2,
+                    /*.ne3 =*/ ne3,
+                    /*.nb0 =*/ nb0,
+                    /*.nb1 =*/ nb1,
+                    /*.nb2 =*/ nb2,
+                    /*.nb3 =*/ nb3,
+                    /*.p0 =*/ p0,
+                    /*.p1 =*/ p1
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
-                [encoder setBytes:&ne00 length:sizeof(ne00) atIndex:2];
-                [encoder setBytes:&ne01 length:sizeof(ne01) atIndex:3];
-                [encoder setBytes:&ne02 length:sizeof(ne02) atIndex:4];
-                [encoder setBytes:&ne03 length:sizeof(ne03) atIndex:5];
-                [encoder setBytes:&ne0  length:sizeof(ne0)  atIndex:6];
-                [encoder setBytes:&nb00 length:sizeof(nb00) atIndex:7];
-                [encoder setBytes:&nb01 length:sizeof(nb01) atIndex:8];
-                [encoder setBytes:&nb02 length:sizeof(nb02) atIndex:9];
-                [encoder setBytes:&nb03 length:sizeof(nb03) atIndex:10];
-                [encoder setBytes:&nb0  length:sizeof(nb0)  atIndex:11];
-                [encoder setBytes:&nb1  length:sizeof(nb1)  atIndex:12];
-                [encoder setBytes:&nb2  length:sizeof(nb2)  atIndex:13];
-                [encoder setBytes:&nb3  length:sizeof(nb3)  atIndex:14];
-                [encoder setBytes:&p0   length:sizeof(p0)   atIndex:15];
-                [encoder setBytes:&p1   length:sizeof(p1)   atIndex:16];
+                [encoder setBytes:&args length:sizeof(args) atIndex:2];
 
                 const int nth = MIN(1024, ne0);
 
@@ -3489,12 +3544,15 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_ARANGE_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_arange args = {
+                    /*.ne0 =*/ ne0,
+                    /*.start =*/ start,
+                    /*.step =*/ step
+                };
+
                 [encoder setComputePipelineState:pipeline];
-                [encoder setBuffer:id_dst  offset:offs_dst    atIndex:0];
-                [encoder setBytes:&ne0   length:sizeof(ne0)   atIndex:1];
-                [encoder setBytes:&start length:sizeof(start) atIndex:2];
-                [encoder setBytes:&step  length:sizeof(step)  atIndex:3];
+                [encoder setBuffer:id_dst  offset:offs_dst  atIndex:0];
+                [encoder setBytes:&args length:sizeof(args) atIndex:1];
 
                 const int nth = MIN(1024, ne0);
 
@@ -3511,13 +3569,16 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_TIMESTEP_EMBEDDING_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_timestep_embedding args = {
+                    /*.nb1 =*/ nb1,
+                    /*.dim =*/ dim,
+                    /*.max_period =*/ max_period
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
-                [encoder setBytes:&nb1   length:sizeof(nb1) atIndex:2];
-                [encoder setBytes:&dim   length:sizeof(dim) atIndex:3];
-                [encoder setBytes:&max_period length:sizeof(max_period) atIndex:4];
+                [encoder setBytes:&args length:sizeof(args) atIndex:2];
 
                 const int nth = MIN(1024, half);
 
@@ -3550,12 +3611,15 @@ static void ggml_metal_encode_node(
                     default: GGML_ABORT("fatal error");
                 };
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_argsort args = {
+                    /*.ncols =*/ ne00,
+                    /*.ncols_pad =*/ ne00_padded
+                };
+
                 [encoder setComputePipelineState:pipeline];
-                [encoder setBuffer:id_src0     offset:offs_src0        atIndex:0];
-                [encoder setBuffer:id_dst      offset:offs_dst         atIndex:1];
-                [encoder setBytes:&ne00        length:sizeof( int64_t) atIndex:2];
-                [encoder setBytes:&ne00_padded length:sizeof( int64_t) atIndex:3];
+                [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+                [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
+                [encoder setBytes:&args length:sizeof(args) atIndex:2];
                 [encoder setThreadgroupMemoryLength:mem_size atIndex:0];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(1, nrows, 1) threadsPerThreadgroup:MTLSizeMake(ne00_padded, 1, 1)];
@@ -3569,11 +3633,14 @@ static void ggml_metal_encode_node(
 
                 id<MTLComputePipelineState> pipeline = ctx->kernels[GGML_METAL_KERNEL_TYPE_LEAKY_RELU_F32].pipeline;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_leaky_relu args = {
+                    /*.slope =*/ slope
+                };
+
                 [encoder setComputePipelineState:pipeline];
                 [encoder setBuffer:id_src0 offset:offs_src0   atIndex:0];
                 [encoder setBuffer:id_dst  offset:offs_dst    atIndex:1];
-                [encoder setBytes:&slope length:sizeof(slope) atIndex:2];
+                [encoder setBytes:&args length:sizeof(args)   atIndex:2];
 
                 const int64_t n = ggml_nelements(dst);
 
@@ -4149,21 +4216,24 @@ static void ggml_metal_encode_node(
                 const int64_t n_threads = MIN((int64_t)[pipeline maxTotalThreadsPerThreadgroup], parallel_elements);
                 const int64_t n_tg = (parallel_elements + n_threads - 1) / n_threads;
 
-                // TODO: add ggml_metal_kargs struct
+                ggml_metal_kargs_pool_2d args_pool_2d = {
+                    /* .k0 = */ k0,
+                    /* .k1 = */ k1,
+                    /* .s0 = */ s0,
+                    /* .s1 = */ s1,
+                    /* .p0 = */ p0,
+                    /* .p1 = */ p1,
+                    /* .IH = */ IH,
+                    /* .IW = */ IW,
+                    /* .OH = */ OH,
+                    /* .OW = */ OW,
+                    /* .parallel_elements = */ parallel_elements
+                };
+
                 [encoder setComputePipelineState:pipeline];
-                [encoder setBuffer:id_src0 offset:offs_src0       atIndex:0];
-                [encoder setBuffer:id_dst  offset:offs_dst        atIndex:1];
-                [encoder setBytes:&k0      length:sizeof(int32_t) atIndex:2];
-                [encoder setBytes:&k1      length:sizeof(int32_t) atIndex:3];
-                [encoder setBytes:&s0      length:sizeof(int32_t) atIndex:4];
-                [encoder setBytes:&s1      length:sizeof(int32_t) atIndex:5];
-                [encoder setBytes:&p0      length:sizeof(int32_t) atIndex:6];
-                [encoder setBytes:&p1      length:sizeof(int32_t) atIndex:7];
-                [encoder setBytes:&IH      length:sizeof(int64_t) atIndex:8];
-                [encoder setBytes:&IW      length:sizeof(int64_t) atIndex:9];
-                [encoder setBytes:&OH      length:sizeof(int64_t) atIndex:10];
-                [encoder setBytes:&OW      length:sizeof(int64_t) atIndex:11];
-                [encoder setBytes:&parallel_elements length:sizeof(int64_t) atIndex:12];
+                [encoder setBuffer:id_src0 offset:offs_src0 atIndex:0];
+                [encoder setBuffer:id_dst  offset:offs_dst  atIndex:1];
+                [encoder setBytes:&args_pool_2d length:sizeof(args_pool_2d) atIndex:2];
 
                 [encoder dispatchThreadgroups:MTLSizeMake(n_tg, 1, 1) threadsPerThreadgroup:MTLSizeMake(n_threads, 1, 1)];
             } break;

ggml/src/ggml-musa/CMakeLists.txt

@@ -21,7 +21,7 @@ if (MUSAToolkit_FOUND)
     message(STATUS "MUSA Toolkit found")
 
     if (NOT DEFINED MUSA_ARCHITECTURES)
-        set(MUSA_ARCHITECTURES "21;22")
+        set(MUSA_ARCHITECTURES "21;22;31")
     endif()
     message(STATUS "Using MUSA architectures: ${MUSA_ARCHITECTURES}")
 

ggml/src/ggml-opencl/CMakeLists.txt

@@ -15,6 +15,7 @@ if (GGML_OPENCL_PROFILING)
 endif ()
 
 add_compile_definitions(GGML_OPENCL_SOA_Q)
+add_compile_definitions(GGML_OPENCL_TARGET_VERSION=${GGML_OPENCL_TARGET_VERSION})
 
 if (GGML_OPENCL_USE_ADRENO_KERNELS)
     message(STATUS "OpenCL will use matmul kernels optimized for Adreno")

ggml/src/ggml-opencl/ggml-opencl.cpp

@@ -1,4 +1,4 @@
-#define CL_TARGET_OPENCL_VERSION 220
+#define CL_TARGET_OPENCL_VERSION GGML_OPENCL_TARGET_VERSION
 #define CL_USE_DEPRECATED_OPENCL_1_2_APIS
 
 // suppress warnings in CL headers for GCC and Clang
@@ -25,6 +25,8 @@
 #include <vector>
 #include <string>
 #include <cmath>
+#include <memory>
+#include <charconv>
 
 #undef MIN
 #undef MAX
@@ -62,6 +64,97 @@ enum ADRENO_GPU_GEN {
     X1E,
 };
 
+struct ggml_cl_version {
+    cl_uint major = 0;
+    cl_uint minor = 0;
+};
+
+// Parses a version string of form "XX.YY ". On an error returns ggml_cl_version with all zeroes.
+static ggml_cl_version parse_cl_version(std::string_view str) {
+    size_t major_str_begin = 0;
+    size_t major_str_end   = str.find(".", major_str_begin);
+    if (major_str_end == std::string::npos) {
+        return {};
+    }
+
+    size_t minor_str_begin = major_str_end + 1;
+    size_t minor_str_end   = str.find(" ", minor_str_begin);
+    if (minor_str_end == std::string::npos) {
+        return {};
+    }
+
+    cl_uint version_major;
+    if (std::from_chars(str.data() + major_str_begin, str.data() + major_str_end, version_major).ec != std::errc{}) {
+        return {};
+    }
+
+    cl_uint version_minor;
+    if (std::from_chars(str.data() + minor_str_begin, str.data() + minor_str_end, version_minor).ec != std::errc{}) {
+        return {};
+    }
+    return { version_major, version_minor };
+}
+
+// Returns OpenCL platform's version. On an error returns ggml_cl_version with all zeroes.
+static ggml_cl_version get_opencl_platform_version(cl_platform_id platform) {
+    size_t param_size;
+    CL_CHECK(clGetPlatformInfo(platform, CL_PLATFORM_VERSION, 0, nullptr, &param_size));
+    std::unique_ptr<char[]> param_storage(new char[param_size]);
+    CL_CHECK(clGetPlatformInfo(platform, CL_PLATFORM_VERSION, param_size, param_storage.get(), nullptr));
+
+    auto              param_value    = std::string_view(param_storage.get(), param_size);
+    const std::string version_prefix = "OpenCL ";  // Suffix: "XX.YY <platform-specific-info>"
+    if (param_value.find(version_prefix) != 0) {
+        return {};
+    }
+    param_value.remove_prefix(version_prefix.length());
+    return parse_cl_version(param_value);
+}
+
+// Return a version to use in OpenCL C compilation. On an error returns ggml_cl_version with all zeroes.
+static ggml_cl_version get_opencl_c_version(ggml_cl_version platform_version, cl_device_id device) {
+    size_t param_size;
+
+#if CL_TARGET_OPENCL_VERSION >= 300
+    if (platform_version.major >= 3) {
+        CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_OPENCL_C_ALL_VERSIONS, 0, nullptr, &param_size));
+        if (!param_size) {
+            return {};
+        }
+
+        std::unique_ptr<cl_name_version[]> versions(new cl_name_version[param_size]);
+        CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_OPENCL_C_ALL_VERSIONS, param_size, versions.get(), nullptr));
+        unsigned versions_count = param_size / sizeof(cl_name_version);
+
+        cl_version version_max = 0;
+        for (unsigned i = 0; i < versions_count; i++) {
+            version_max = std::max<cl_version>(versions[i].version, version_max);
+        }
+
+        return { CL_VERSION_MAJOR(version_max), CL_VERSION_MINOR(version_max) };
+    }
+#else
+    GGML_UNUSED(platform_version);
+#endif  // CL_TARGET_OPENCL_VERSION >= 300
+
+    CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_OPENCL_C_VERSION, 0, nullptr, &param_size));
+    if (!param_size) {
+        return {};
+    }
+
+    std::unique_ptr<char[]> param_storage(new char[param_size]);
+    CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_OPENCL_C_VERSION, param_size, param_storage.get(), nullptr));
+    auto param_value = std::string_view(param_storage.get(), param_size);
+
+    const std::string version_prefix = "OpenCL C ";  // Suffix: "XX.YY <platform-specific-info>"
+    if (param_value.find(version_prefix) != 0) {
+        return {};
+    }
+    param_value.remove_prefix(version_prefix.length());
+
+    return parse_cl_version(param_value);
+}
+
 static ADRENO_GPU_GEN get_adreno_gpu_gen(const char *device_name) {
     if (strstr(device_name, "730") ||
         strstr(device_name, "740") ||
@@ -278,7 +371,7 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
 
     cl_int err;
 
-#ifdef GGML_PROFILE_OPENCL
+#ifdef GGML_OPENCL_PROFILING
     GGML_LOG_INFO("ggml_opencl: OpenCL profiling enabled\n");
 #endif
 
@@ -470,16 +563,11 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     // A local ref of cl_device_id for convenience
     cl_device_id device = backend_ctx->device;
 
-    // Check device OpenCL version, OpenCL 2.0 or above is required
-    size_t device_ver_str_size;
-    clGetDeviceInfo(device, CL_DEVICE_VERSION, 0, NULL, &device_ver_str_size);
-    char *device_ver_buffer = (char *)alloca(device_ver_str_size + 1);
-    clGetDeviceInfo(device, CL_DEVICE_VERSION, device_ver_str_size, device_ver_buffer, NULL);
-    device_ver_buffer[device_ver_str_size] = '\0';
-    GGML_LOG_INFO("ggml_opencl: device OpenCL version: %s\n", device_ver_buffer);
+    ggml_cl_version platform_version = get_opencl_platform_version(default_device->platform->id);
 
-    if (strstr(device_ver_buffer, "OpenCL 2") == NULL &&
-        strstr(device_ver_buffer, "OpenCL 3") == NULL) {
+    // Check device OpenCL version, OpenCL 2.0 or above is required
+    ggml_cl_version opencl_c_version = get_opencl_c_version(platform_version, device);
+    if (opencl_c_version.major < 2) {
         GGML_LOG_ERROR("ggml_opencl: OpenCL 2.0 or above is required\n");
         return backend_ctx;
     }
@@ -516,15 +604,17 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
 
     // If OpenCL 3.0 is supported, then check for cl_khr_subgroups, which becomes
     // optional in OpenCL 3.0 (cl_khr_subgroup is mandatory in OpenCL 2.x)
-    if (strstr(device_ver_buffer, "OpenCL 3") &&
-        strstr(ext_buffer, "cl_khr_subgroups") == NULL &&
+    if (opencl_c_version.major == 3 && strstr(ext_buffer, "cl_khr_subgroups") == NULL &&
         strstr(ext_buffer, "cl_intel_subgroups") == NULL) {
         GGML_LOG_ERROR("ggml_opencl: device does not support subgroups (cl_khr_subgroups or cl_intel_subgroups) "
             "(note that subgroups is an optional feature in OpenCL 3.0)\n");
         return backend_ctx;
     }
 
-    CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_MEM_BASE_ADDR_ALIGN, sizeof(cl_uint), &backend_ctx->alignment, NULL));
+    cl_uint base_align_in_bits;
+    CL_CHECK(clGetDeviceInfo(device, CL_DEVICE_MEM_BASE_ADDR_ALIGN, sizeof(cl_uint), &base_align_in_bits, NULL));
+    GGML_ASSERT(base_align_in_bits % 8u == 0);
+    backend_ctx->alignment = base_align_in_bits / 8u;
     GGML_LOG_INFO("ggml_opencl: mem base addr align: %u\n", backend_ctx->alignment);
 
     clGetDeviceInfo(device, CL_DEVICE_MAX_MEM_ALLOC_SIZE, sizeof(size_t), &backend_ctx->max_alloc_size, NULL);
@@ -578,9 +668,12 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     const std::string kernel_src = read_file("ggml-opencl.cl");
 #endif
 
-    std::string compile_opts =
-        "-cl-std=CL2.0 -cl-mad-enable -cl-unsafe-math-optimizations "
-        "-cl-finite-math-only -cl-fast-relaxed-math ";
+    auto opencl_c_std =
+        std::string("CL") + std::to_string(opencl_c_version.major) + "." + std::to_string(opencl_c_version.minor);
+
+    std::string compile_opts = std::string("-cl-std=") + opencl_c_std +
+                               " -cl-mad-enable -cl-unsafe-math-optimizations"
+                               " -cl-finite-math-only -cl-fast-relaxed-math";
     backend_ctx->program = build_program_from_source(context, device, kernel_src.c_str(), compile_opts);
 
     // Non matmul kernels.
@@ -690,10 +783,10 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     CL_CHECK((backend_ctx->kernel_transpose_16 = clCreateKernel(backend_ctx->program_transpose_16, "kernel_transpose_16", &err), err));
 
     // Gemv general
-    std::string CL_gemv_compile_opts =
-        " -cl-std=CL2.0 "
-        " -cl-mad-enable "
-        " -DSIMDGROUP_WIDTH=" + std::to_string(backend_ctx->adreno_wave_size);
+    std::string CL_gemv_compile_opts = std::string("-cl-std=") + opencl_c_std +
+                                       " -cl-mad-enable "
+                                       " -DSIMDGROUP_WIDTH=" +
+                                       std::to_string(backend_ctx->adreno_wave_size);
     if (has_vector_subgroup_broadcast) {
         CL_gemv_compile_opts += " -DVECTOR_SUB_GROUP_BROADCAT ";
     }
@@ -710,12 +803,12 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     CL_CHECK((backend_ctx->CL_mul_mat_vec_q4_0_f32_1d_4x_flat_general = clCreateKernel(backend_ctx->program_CL_gemv_general, "kernel_gemv_noshuffle", &err), err));
 
     // Gemv 2048, 16384
-    CL_gemv_compile_opts =
-        " -cl-std=CL2.0 "
-        " -cl-mad-enable "
-        " -DLINE_STRIDE_A=2048 "
-        " -DBLOCK_STRIDE_A=16384 "
-        " -DSIMDGROUP_WIDTH=" + std::to_string(backend_ctx->adreno_wave_size);
+    CL_gemv_compile_opts = std::string("-cl-std=") + opencl_c_std +
+                           " -cl-mad-enable "
+                           " -DLINE_STRIDE_A=2048 "
+                           " -DBLOCK_STRIDE_A=16384 "
+                           " -DSIMDGROUP_WIDTH=" +
+                           std::to_string(backend_ctx->adreno_wave_size);
     if (has_vector_subgroup_broadcast) {
         CL_gemv_compile_opts += " -DVECTOR_SUB_GROUP_BROADCAT ";
     }
@@ -732,12 +825,12 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     CL_CHECK((backend_ctx->CL_mul_mat_vec_q4_0_f32_1d_4x_flat_4096_1_4096 = clCreateKernel(backend_ctx->program_CL_gemv_4096_1_4096, "kernel_gemv_noshuffle", &err), err));
 
     // Gemv 2048, 16384
-    CL_gemv_compile_opts =
-        " -cl-std=CL2.0 "
-        " -cl-mad-enable "
-        " -DLINE_STRIDE_A=2048 "
-        " -DBLOCK_STRIDE_A=16384 "
-        " -DSIMDGROUP_WIDTH=" + std::to_string(backend_ctx->adreno_wave_size);
+    CL_gemv_compile_opts = std::string("-cl-std=") + opencl_c_std +
+                           " -cl-mad-enable "
+                           " -DLINE_STRIDE_A=2048 "
+                           " -DBLOCK_STRIDE_A=16384 "
+                           " -DSIMDGROUP_WIDTH=" +
+                           std::to_string(backend_ctx->adreno_wave_size);
     if (has_vector_subgroup_broadcast) {
         CL_gemv_compile_opts += " -DVECTOR_SUB_GROUP_BROADCAT ";
     }
@@ -747,12 +840,12 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     CL_CHECK((backend_ctx->CL_mul_mat_vec_q4_0_f32_1d_4x_flat_4096_1_11008 = clCreateKernel(backend_ctx->program_CL_gemv_4096_1_11008, "kernel_gemv_noshuffle", &err), err));
 
     // Gemv 5504, 44032
-    CL_gemv_compile_opts =
-        " -cl-std=CL2.0 "
-        " -cl-mad-enable "
-        " -DLINE_STRIDE_A=5504 "
-        " -DBLOCK_STRIDE_A=44032 "
-        " -DSIMDGROUP_WIDTH=" + std::to_string(backend_ctx->adreno_wave_size);
+    CL_gemv_compile_opts = std::string("-cl-std=") + opencl_c_std +
+                           " -cl-mad-enable "
+                           " -DLINE_STRIDE_A=5504 "
+                           " -DBLOCK_STRIDE_A=44032 "
+                           " -DSIMDGROUP_WIDTH=" +
+                           std::to_string(backend_ctx->adreno_wave_size);
     if (has_vector_subgroup_broadcast) {
         CL_gemv_compile_opts += " -DVECTOR_SUB_GROUP_BROADCAT ";
     }
@@ -762,12 +855,12 @@ static ggml_backend_opencl_context * ggml_cl2_init(ggml_backend_dev_t dev) {
     CL_CHECK((backend_ctx->CL_mul_mat_vec_q4_0_f32_1d_4x_flat_11008_1_4096 = clCreateKernel(backend_ctx->program_CL_gemv_11008_1_4096, "kernel_gemv_noshuffle", &err), err));
 
     // Gemv 16000, 128000
-    CL_gemv_compile_opts =
-        " -cl-std=CL2.0 "
-        " -cl-mad-enable "
-        " -DLINE_STRIDE_A=16000 "
-        " -DBLOCK_STRIDE_A=128000 "
-        " -DSIMDGROUP_WIDTH=" + std::to_string(backend_ctx->adreno_wave_size);
+    CL_gemv_compile_opts = std::string("-cl-std=") + opencl_c_std +
+                           " -cl-mad-enable "
+                           " -DLINE_STRIDE_A=16000 "
+                           " -DBLOCK_STRIDE_A=128000 "
+                           " -DSIMDGROUP_WIDTH=" +
+                           std::to_string(backend_ctx->adreno_wave_size);
     if (has_vector_subgroup_broadcast) {
         CL_gemv_compile_opts += " -DVECTOR_SUB_GROUP_BROADCAT ";
     }
@@ -1004,17 +1097,18 @@ static bool ggml_opencl_supports_op(ggml_backend_dev_t dev, const struct ggml_te
         case GGML_OP_ADD:
         case GGML_OP_SCALE:
         case GGML_OP_MUL:
-            return true;
+            return op->src[0]->type == GGML_TYPE_F32;
         case GGML_OP_UNARY:
             switch (ggml_get_unary_op(op)) {
                 case GGML_UNARY_OP_GELU:
                 case GGML_UNARY_OP_SILU:
                 case GGML_UNARY_OP_RELU:
-                   return ggml_is_contiguous(op->src[0]);
+                   return ggml_is_contiguous(op->src[0]) && op->src[0]->type == GGML_TYPE_F32;
                 default:
                     return false;
             }
         case GGML_OP_CLAMP:
+            return op->src[0]->type == GGML_TYPE_F32;
         case GGML_OP_SOFT_MAX:
         case GGML_OP_NORM:
         case GGML_OP_RMS_NORM:
@@ -1198,17 +1292,14 @@ struct ggml_backend_opencl_buffer_context {
     std::string name;
 };
 
-static void * const cl_ptr_base = (void *)(uintptr_t) 0x1000;
-
 static void ggml_backend_opencl_buffer_free_buffer(ggml_backend_buffer_t buffer) {
     ggml_backend_opencl_buffer_context * ctx = (ggml_backend_opencl_buffer_context *) buffer->context;
     delete ctx;
 }
 
 static void * ggml_backend_opencl_buffer_get_base(ggml_backend_buffer_t buffer) {
-    return cl_ptr_base;
-
-    GGML_UNUSED(buffer);
+    ggml_backend_opencl_context * backend_ctx = ggml_cl2_init(buffer->buft->device);
+    return (void *) (uintptr_t) backend_ctx->alignment;
 }
 
 static enum ggml_status ggml_backend_opencl_buffer_init_tensor(ggml_backend_buffer_t buffer, ggml_tensor * tensor) {
@@ -1241,7 +1332,7 @@ static enum ggml_status ggml_backend_opencl_buffer_init_tensor(ggml_backend_buff
         tensor->extra = view_extra;
     } else {
         {
-            size_t offset = (char *)tensor->data - (char *)cl_ptr_base;
+            size_t offset = (char *) tensor->data - (char *) ggml_backend_opencl_buffer_get_base(buffer);
 
             ggml_tensor_extra_cl * extra = ctx->ggml_opencl_alloc_temp_tensor_extra();
             extra->offset = offset;
@@ -2573,26 +2664,33 @@ static void ggml_cl_norm(ggml_backend_t backend, const ggml_tensor * src0, const
     memcpy(&eps, dst->op_params, sizeof(float));
 
     const int ne00 = src0 ? src0->ne[0] : 0;
-    const cl_ulong nb01 = src0 ? src0->nb[1] : 0;
+    const int ne01 = src0 ? src0->ne[1] : 0;
+    const int ne02 = src0 ? src0->ne[2] : 0;
+    const int ne03 = src0 ? src0->ne[3] : 0;
 
-    GGML_ASSERT(ggml_is_contiguous_1(src0));
+    const cl_ulong nb01 = src0 ? src0->nb[1] : 0;
+    const cl_ulong nb02 = src0 ? src0->nb[2] : 0;
+    const cl_ulong nb03 = src0 ? src0->nb[3] : 0;
 
     const int nth = MIN(64, ne00);
 
     cl_kernel kernel = backend_ctx->kernel_norm;
 
-    CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem),    &extra0->data_device));
-    CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_ulong),  &offset0));
-    CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem),    &extrad->data_device));
-    CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_ulong),  &offsetd));
-    CL_CHECK(clSetKernelArg(kernel, 4, sizeof(int),       &ne00));
-    CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_ulong),  &nb01));
-    CL_CHECK(clSetKernelArg(kernel, 6, sizeof(float),     &eps));
-    CL_CHECK(clSetKernelArg(kernel, 7, sizeof(float)*nth, NULL));
+    CL_CHECK(clSetKernelArg(kernel,  0, sizeof(cl_mem),    &extra0->data_device));
+    CL_CHECK(clSetKernelArg(kernel,  1, sizeof(cl_ulong),  &offset0));
+    CL_CHECK(clSetKernelArg(kernel,  2, sizeof(cl_mem),    &extrad->data_device));
+    CL_CHECK(clSetKernelArg(kernel,  3, sizeof(cl_ulong),  &offsetd));
+    CL_CHECK(clSetKernelArg(kernel,  4, sizeof(int),       &ne00));
+    CL_CHECK(clSetKernelArg(kernel,  5, sizeof(int),       &ne01));
+    CL_CHECK(clSetKernelArg(kernel,  6, sizeof(int),       &ne02));
+    CL_CHECK(clSetKernelArg(kernel,  7, sizeof(int),       &ne03));
+    CL_CHECK(clSetKernelArg(kernel,  8, sizeof(cl_ulong),  &nb01));
+    CL_CHECK(clSetKernelArg(kernel,  9, sizeof(cl_ulong),  &nb02));
+    CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_ulong),  &nb03));
+    CL_CHECK(clSetKernelArg(kernel, 11, sizeof(float),     &eps));
+    CL_CHECK(clSetKernelArg(kernel, 12, sizeof(float)*nth, NULL));
 
-    const int64_t nrows = ggml_nrows(src0);
-
-    size_t global_work_size[] = {(size_t)nrows*nth, 1, 1};
+    size_t global_work_size[] = {(size_t)ne01*nth, (size_t)ne02, (size_t)ne03};
     size_t local_work_size[] = {(size_t)nth, 1, 1};
 
 #ifdef GGML_OPENCL_PROFILING
@@ -2630,16 +2728,19 @@ static void ggml_cl_rms_norm(ggml_backend_t backend, const ggml_tensor * src0, c
     memcpy(&eps, dst->op_params, sizeof(float));
 
     const int ne00 = src0 ? src0->ne[0] : 0;
+    const int ne01 = src0 ? src0->ne[1] : 0;
+    const int ne02 = src0 ? src0->ne[2] : 0;
+    const int ne03 = src0 ? src0->ne[3] : 0;
+
     const cl_ulong nb01 = src0 ? src0->nb[1] : 0;
+    const cl_ulong nb02 = src0 ? src0->nb[2] : 0;
+    const cl_ulong nb03 = src0 ? src0->nb[3] : 0;
 
     GGML_ASSERT(ne00 % 4 == 0);
-    GGML_ASSERT(ggml_is_contiguous_1(src0));
 
     const int nth = MIN(64, ne00);
 
-    const int64_t nrows = ggml_nrows(src0);
-
-    size_t global_work_size[] = {(size_t)nrows*nth, 1, 1};
+    size_t global_work_size[] = {(size_t)ne01*nth, (size_t)ne02, (size_t)ne03};
     size_t local_work_size[] = {(size_t)nth, 1, 1};
 
     cl_kernel kernel = backend_ctx->kernel_rms_norm;
@@ -2654,15 +2755,20 @@ static void ggml_cl_rms_norm(ggml_backend_t backend, const ggml_tensor * src0, c
         sizeof(local_work_size), local_work_size,
         sizeof(size_t), &sgs, NULL));
 
-    CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem),    &extra0->data_device));
-    CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_ulong),  &offset0));
-    CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem),    &extrad->data_device));
-    CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_ulong),  &offsetd));
-    CL_CHECK(clSetKernelArg(kernel, 4, sizeof(int),       &ne00));
-    CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_ulong),  &nb01));
-    CL_CHECK(clSetKernelArg(kernel, 6, sizeof(float),     &eps));
+    CL_CHECK(clSetKernelArg(kernel,  0, sizeof(cl_mem),    &extra0->data_device));
+    CL_CHECK(clSetKernelArg(kernel,  1, sizeof(cl_ulong),  &offset0));
+    CL_CHECK(clSetKernelArg(kernel,  2, sizeof(cl_mem),    &extrad->data_device));
+    CL_CHECK(clSetKernelArg(kernel,  3, sizeof(cl_ulong),  &offsetd));
+    CL_CHECK(clSetKernelArg(kernel,  4, sizeof(int),       &ne00));
+    CL_CHECK(clSetKernelArg(kernel,  5, sizeof(int),       &ne01));
+    CL_CHECK(clSetKernelArg(kernel,  6, sizeof(int),       &ne02));
+    CL_CHECK(clSetKernelArg(kernel,  7, sizeof(int),       &ne03));
+    CL_CHECK(clSetKernelArg(kernel,  8, sizeof(cl_ulong),  &nb01));
+    CL_CHECK(clSetKernelArg(kernel,  9, sizeof(cl_ulong),  &nb02));
+    CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_ulong),  &nb03));
+    CL_CHECK(clSetKernelArg(kernel, 11, sizeof(float),     &eps));
     // This is local memory - the size depends on subgroup size.
-    CL_CHECK(clSetKernelArg(kernel, 7, sizeof(float)*nth/sgs,  NULL));
+    CL_CHECK(clSetKernelArg(kernel, 12, sizeof(float)*nth/sgs,  NULL));
 
 #ifdef GGML_OPENCL_PROFILING
     cl_event evt;
@@ -3023,6 +3129,7 @@ static void ggml_cl_mul_mat(ggml_backend_t backend, const ggml_tensor * src0, co
         // enqueue kernel with profiling
         // <--------------------------------------------> //
     #ifdef GGML_OPENCL_PROFILING
+        cl_event evt;
         CL_CHECK(clEnqueueNDRangeKernel(queue, kernel, 3, NULL, global_work_size, local_work_size, 0, NULL, &evt));
 
         g_profiling_info.emplace_back();
@@ -3764,10 +3871,10 @@ static void ggml_cl_rope(ggml_backend_t backend, const ggml_tensor * src0, const
     const int  ne02 = src0 ? src0->ne[2] : 0;
     const int  ne03 = src0 ? src0->ne[3] : 0;
 
-    const int  nb00 = src0 ? src0->nb[0] : 0;
-    const int  nb01 = src0 ? src0->nb[1] : 0;
-    const int  nb02 = src0 ? src0->nb[2] : 0;
-    const int  nb03 = src0 ? src0->nb[3] : 0;
+    const cl_ulong  nb00 = src0 ? src0->nb[0] : 0;
+    const cl_ulong  nb01 = src0 ? src0->nb[1] : 0;
+    const cl_ulong  nb02 = src0 ? src0->nb[2] : 0;
+    const cl_ulong  nb03 = src0 ? src0->nb[3] : 0;
 
     const int ne10 = src1 ? src1->ne[0] : 0;
     const int ne11 = src1 ? src1->ne[1] : 0; UNUSED(ne11);
@@ -3779,10 +3886,10 @@ static void ggml_cl_rope(ggml_backend_t backend, const ggml_tensor * src0, const
     const int  ne2 = dst ? dst->ne[2] : 0;
     const int  ne3 = dst ? dst->ne[3] : 0;
 
-    const int  nb0 = dst ? dst->nb[0] : 0;
-    const int  nb1 = dst ? dst->nb[1] : 0;
-    const int  nb2 = dst ? dst->nb[2] : 0;
-    const int  nb3 = dst ? dst->nb[3] : 0;
+    const cl_ulong  nb0 = dst ? dst->nb[0] : 0;
+    const cl_ulong  nb1 = dst ? dst->nb[1] : 0;
+    const cl_ulong  nb2 = dst ? dst->nb[2] : 0;
+    const cl_ulong  nb3 = dst ? dst->nb[3] : 0;
 
     GGML_ASSERT(ne10 % ne02 == 0);
     GGML_ASSERT(ne10 >= ne02);

ggml/src/ggml-opencl/kernels/ggml-opencl.cl

@@ -506,14 +506,23 @@ kernel void kernel_norm(
         global float * dst,
         ulong offsetd,
         int ne00,
+        int ne01,
+        int ne02,
+        int ne03,
         ulong nb01,
+        ulong nb02,
+        ulong nb03,
         float eps,
         local float * sum
 ) {
     src0 = (global void*)((global char*)src0 + offset0);
     dst = (global void*)((global char*)dst + offsetd);
 
-    global float * x = (global float *) ((global char *) src0 + get_group_id(0)*nb01);
+    int i03 = get_group_id(2);
+    int i02 = get_group_id(1);
+    int i01 = get_group_id(0);
+
+    global float * x = (global float *) ((global char *) src0 + i03*nb03 + i02*nb02 + i01*nb01);
 
     // MEAN
     // parallel sum
@@ -533,7 +542,7 @@ kernel void kernel_norm(
 
     // recenter and VARIANCE
     barrier(CLK_LOCAL_MEM_FENCE);
-    global float * y = dst + get_group_id(0)*ne00;
+    global float * y = dst + i03*ne02*ne01*ne00 + i02*ne01*ne00 + i01*ne00;
     sum[get_local_id(0)] = 0.0f;
     for (int i00 = get_local_id(0); i00 < ne00; i00 += get_local_size(0)) {
         y[i00] = x[i00] - mean;
@@ -566,14 +575,23 @@ kernel void kernel_rms_norm(
         global float * dst,
         ulong offsetd,
         int ne00,
+        int ne01,
+        int ne02,
+        int ne03,
         ulong nb01,
+        ulong nb02,
+        ulong nb03,
         float eps,
         local float * sum // Note, the size depends on number of subgroups
 ) {
     src0 = (global void*)((global char*)src0 + offset0);
     dst = (global float*)((global char*)dst + offsetd);
 
-    global float4 * x = (global float4 *) ((global char *) src0 + get_group_id(0)*nb01);
+    int i03 = get_group_id(2);
+    int i02 = get_group_id(1);
+    int i01 = get_group_id(0);
+
+    global float4 * x = (global float4 *) ((global char *) src0 + i03*nb03 + i02*nb02 + i01*nb01);
     global float * x_scalar = (global float *) x;
     float4 sumf = 0;
     float all_sum = 0;
@@ -607,7 +625,7 @@ kernel void kernel_rms_norm(
     const float mean  = sum[0];
     const float scale = 1.0f/sqrt(mean + eps);
 
-    global float4 * y = (global float4 *) (dst + get_group_id(0)*ne00);
+    global float4 * y = (global float4 *) (dst + i03*ne02*ne01*ne00 + i02*ne01*ne00 + i01*ne00);
     global float * y_scalar = (global float *) y;
     for (int i00 = get_local_id(0); i00 < ne00/4; i00 += get_local_size(0)) {
         y[i00] = x[i00] * scale;