server : chunked prefill support

ggml-ci

.clang-tidy

@@ -17,8 +17,10 @@ Checks: >
     -clang-analyzer-security.insecureAPI.DeprecatedOrUnsafeBufferHandling,
     performance-*,
     portability-*,
+    -portability-simd-intrinsics,
     misc-*,
     -misc-const-correctness,
     -misc-non-private-member-variables-in-classes,
     -misc-no-recursion,
+    -misc-use-anonymous-namespace,
 FormatStyle: none

.devops/full.Dockerfile

@@ -3,23 +3,36 @@ ARG UBUNTU_VERSION=22.04
 FROM ubuntu:$UBUNTU_VERSION AS build
 
 RUN apt-get update && \
-    apt-get install -y build-essential python3 python3-pip git libcurl4-openssl-dev libgomp1
-
-COPY requirements.txt   requirements.txt
-COPY requirements       requirements
-
-RUN pip install --upgrade pip setuptools wheel \
-    && pip install -r requirements.txt
+    apt-get install -y build-essential git cmake libcurl4-openssl-dev
 
 WORKDIR /app
 
 COPY . .
 
-ENV LLAMA_CURL=1
+RUN cmake -S . -B build -DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ALL_VARIANTS=ON -DLLAMA_CURL=ON -DCMAKE_BUILD_TYPE=Release && \
+    cmake --build build -j $(nproc) && \
+    mkdir -p /app/lib && \
+    find build -name "*.so" -exec cp {} /app/lib/ \;
 
+FROM ubuntu:$UBUNTU_VERSION as runtime
 
-RUN make -j$(nproc)
+WORKDIR /app
+
+RUN apt-get update && \
+    apt-get install -y build-essential python3 python3-pip git libcurl4-openssl-dev libgomp1
+
+COPY requirements.txt   /app/requirements.txt
+COPY requirements       /app/requirements
+COPY .devops/tools.sh   /app/tools.sh
+
+RUN pip install --upgrade pip setuptools wheel && \
+    pip install -r /app/requirements.txt
+
+COPY --from=build /app/build/bin/ /app/
+COPY --from=build /app/lib/ /app/
+COPY --from=build /app/convert_hf_to_gguf.py /app/
+COPY --from=build /app/gguf-py /app/gguf-py
 
 ENV LC_ALL=C.utf8
 
-ENTRYPOINT ["/app/.devops/tools.sh"]
+ENTRYPOINT ["/app/tools.sh"]

.devops/llama-cli.Dockerfile

@@ -3,21 +3,27 @@ ARG UBUNTU_VERSION=22.04
 FROM ubuntu:$UBUNTU_VERSION AS build
 
 RUN apt-get update && \
-    apt-get install -y build-essential git
+    apt-get install -y build-essential git cmake libcurl4-openssl-dev
 
 WORKDIR /app
 
 COPY . .
 
-RUN make -j$(nproc) llama-cli
+RUN cmake -S . -B build -DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ALL_VARIANTS=ON -DLLAMA_CURL=ON -DCMAKE_BUILD_TYPE=Release && \
+    cmake --build build -j $(nproc) && \
+    mkdir -p /app/lib && \
+    find build -name "*.so" -exec cp {} /app/lib/ \;
 
 FROM ubuntu:$UBUNTU_VERSION AS runtime
 
-RUN apt-get update && \
-    apt-get install -y libgomp1
+WORKDIR /app
 
-COPY --from=build /app/llama-cli /llama-cli
+RUN apt-get update && \
+    apt-get install -y libcurl4-openssl-dev libgomp1 curl
+
+COPY --from=build /app/build/bin/llama-cli /app/
+COPY --from=build /app/lib/ /app/
 
 ENV LC_ALL=C.utf8
 
-ENTRYPOINT [ "/llama-cli" ]
+ENTRYPOINT [ "/app/llama-cli" ]

.devops/llama-server.Dockerfile

@@ -3,22 +3,26 @@ ARG UBUNTU_VERSION=22.04
 FROM ubuntu:$UBUNTU_VERSION AS build
 
 RUN apt-get update && \
-    apt-get install -y build-essential git libcurl4-openssl-dev
+    apt-get install -y build-essential git cmake libcurl4-openssl-dev
 
 WORKDIR /app
 
 COPY . .
 
-ENV LLAMA_CURL=1
-
-RUN make -j$(nproc) llama-server
+RUN cmake -S . -B build -DGGML_BACKEND_DL=ON -DGGML_NATIVE=OFF -DGGML_CPU_ALL_VARIANTS=ON -DLLAMA_CURL=ON -DCMAKE_BUILD_TYPE=Release && \
+    cmake --build build -j $(nproc) && \
+    mkdir -p /app/lib && \
+    find build -name "*.so" -exec cp {} /app/lib/ \;
 
 FROM ubuntu:$UBUNTU_VERSION AS runtime
 
+WORKDIR /app
+
 RUN apt-get update && \
     apt-get install -y libcurl4-openssl-dev libgomp1 curl
 
-COPY --from=build /app/llama-server /llama-server
+COPY --from=build /app/build/bin/llama-server /app/
+COPY --from=build /app/lib/ /app/
 
 ENV LC_ALL=C.utf8
 # Must be set to 0.0.0.0 so it can listen to requests from host machine
@@ -26,4 +30,4 @@ ENV LLAMA_ARG_HOST=0.0.0.0
 
 HEALTHCHECK CMD [ "curl", "-f", "http://localhost:8080/health" ]
 
-ENTRYPOINT [ "/llama-server" ]
+ENTRYPOINT [ "/app/llama-server" ]

.github/pull_request_template.md

@@ -1,7 +1 @@
-
-
-- [x] I have read the [contributing guidelines](https://github.com/ggerganov/llama.cpp/blob/master/CONTRIBUTING.md)
-- Self-reported review complexity:
-  - [ ] Low
-  - [ ] Medium
-  - [ ] High
+*Make sure to read the [contributing guidelines](https://github.com/ggerganov/llama.cpp/blob/master/CONTRIBUTING.md) before submitting a PR*

.github/workflows/build.yml

@@ -160,66 +160,6 @@ jobs:
           path: llama-${{ steps.tag.outputs.name }}-bin-macos-x64.zip
           name: llama-bin-macos-x64.zip
 
-  ubuntu-focal-make:
-    runs-on: ubuntu-20.04
-    env:
-      LLAMA_NODE_AVAILABLE: true
-      LLAMA_PYTHON_AVAILABLE: true
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install build-essential gcc-8
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-
-      - uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-
-      - name: Build
-        id: make_build
-        env:
-            LLAMA_FATAL_WARNINGS: 1
-        run: |
-          CC=gcc-8 make -j $(nproc)
-
-      - name: Test
-        id: make_test
-        run: |
-          CC=gcc-8 make tests -j $(nproc)
-          make test -j $(nproc)
-
-  ubuntu-focal-make-curl:
-    runs-on: ubuntu-20.04
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Dependencies
-        id: depends
-        run: |
-          sudo apt-get update
-          sudo apt-get install build-essential gcc-8 libcurl4-openssl-dev
-
-      - name: Build
-        id: make_build
-        env:
-          LLAMA_FATAL_WARNINGS: 1
-          LLAMA_CURL: 1
-        run: |
-          CC=gcc-8 make -j $(nproc)
-
   ubuntu-latest-cmake:
     runs-on: ubuntu-latest
 
@@ -517,36 +457,6 @@ jobs:
           cmake -DGGML_SYCL=ON -DCMAKE_C_COMPILER=icx -DCMAKE_CXX_COMPILER=icpx -DGGML_SYCL_F16=ON ..
           cmake --build . --config Release -j $(nproc)
 
-  # TODO: build with GGML_NO_METAL because test-backend-ops fail on "Apple Paravirtual device" and I don't know
-  #       how to debug it.
-  #       ref: https://github.com/ggerganov/llama.cpp/actions/runs/7131777249/job/19420981052#step:5:1124
-  macOS-latest-make:
-    runs-on: macos-latest
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Dependencies
-        id: depends
-        continue-on-error: true
-        run: |
-          brew update
-
-      - name: Build
-        id: make_build
-        env:
-            LLAMA_FATAL_WARNINGS: 1
-        run: |
-          GGML_NO_METAL=1 make -j $(sysctl -n hw.logicalcpu)
-
-      - name: Test
-        id: make_test
-        run: |
-          GGML_NO_METAL=1 make tests -j $(sysctl -n hw.logicalcpu)
-          GGML_NO_METAL=1 make test  -j $(sysctl -n hw.logicalcpu)
-
   # TODO: build with GGML_METAL=OFF because test-backend-ops fail on "Apple Paravirtual device" and I don't know
   #       how to debug it.
   #       ref: https://github.com/ggerganov/llama.cpp/actions/runs/7132125951/job/19422043567?pr=4359#step:5:6584
@@ -642,33 +552,35 @@ jobs:
             -DCMAKE_XCODE_ATTRIBUTE_DEVELOPMENT_TEAM=ggml
           cmake --build . --config Release -j $(sysctl -n hw.logicalcpu) -- CODE_SIGNING_ALLOWED=NO
 
-  macOS-latest-swift:
-    runs-on: macos-latest
-
-    strategy:
-      matrix:
-        destination: ['generic/platform=macOS', 'generic/platform=iOS', 'generic/platform=tvOS']
-
-    steps:
-      - name: Clone
-        id: checkout
-        uses: actions/checkout@v4
-
-      - name: Dependencies
-        id: depends
-        continue-on-error: true
-        run: |
-          brew update
-
-      - name: xcodebuild for swift package
-        id: xcodebuild
-        run: |
-          xcodebuild -scheme llama -destination "${{ matrix.destination }}"
-
-      - name: Build Swift Example
-        id: make_build_swift_example
-        run: |
-            make swift
+# TODO: tmp disabled. see for possible re-enable:
+#       https://github.com/ggerganov/llama.cpp/pull/10525
+#  macOS-latest-swift:
+#    runs-on: macos-latest
+#
+#    strategy:
+#      matrix:
+#        destination: ['generic/platform=macOS', 'generic/platform=iOS', 'generic/platform=tvOS']
+#
+#    steps:
+#      - name: Clone
+#        id: checkout
+#        uses: actions/checkout@v4
+#
+#      - name: Dependencies
+#        id: depends
+#        continue-on-error: true
+#        run: |
+#          brew update
+#
+#      - name: xcodebuild for swift package
+#        id: xcodebuild
+#        run: |
+#          xcodebuild -scheme llama -destination "${{ matrix.destination }}"
+#
+#      - name: Build Swift Example
+#        id: make_build_swift_example
+#        run: |
+#            make swift
 
   windows-msys2:
     runs-on: windows-latest
@@ -695,21 +607,6 @@ jobs:
             mingw-w64-${{matrix.env}}-cmake
             mingw-w64-${{matrix.env}}-openblas
 
-      - name: Build using make
-        shell: msys2 {0}
-        run: |
-            make -j $(nproc)
-
-      - name: Clean after building using make
-        shell: msys2 {0}
-        run: |
-            make clean
-
-      - name: Build using make w/ OpenBLAS
-        shell: msys2 {0}
-        run: |
-            make GGML_OPENBLAS=1 -j $(nproc)
-
       - name: Build using CMake
         shell: msys2 {0}
         run: |
@@ -1121,6 +1018,11 @@ jobs:
         run: |
           & 'C:\Program Files\AMD\ROCm\*\bin\clang.exe' --version
 
+      - name: Install ccache
+        uses: hendrikmuhs/ccache-action@v1.2
+        with:
+          key: ${{ github.job }}
+
       - name: Build
         id: cmake_build
         run: |
@@ -1252,9 +1154,7 @@ jobs:
     runs-on: ubuntu-latest
 
     needs:
-      - ubuntu-focal-make
       - ubuntu-latest-cmake
-      - macOS-latest-make
       - macOS-latest-cmake
       - windows-latest-cmake
       - windows-2019-cmake-cuda

.github/workflows/server.yml

@@ -76,20 +76,26 @@ jobs:
         run: |
           pip install -r examples/server/tests/requirements.txt
 
-      - name: Verify server deps
-        id: verify_server_deps
+      # Setup nodejs (to be used for verifying bundled index.html)
+      - uses: actions/setup-node@v4
+        with:
+          node-version: 22
+
+      - name: Verify bundled index.html
+        id: verify_server_index_html
         run: |
           git config --global --add safe.directory $(realpath .)
-          cd examples/server
-          git ls-files --others --modified
+          cd examples/server/webui
           git status
-          ./deps.sh
+          npm ci
+          npm run build
           git status
-          not_ignored_files="$(git ls-files --others --modified)"
-          echo "Modified files: ${not_ignored_files}"
-          if [ -n "${not_ignored_files}" ]; then
-            echo "Repository is dirty or server deps are not built as expected"
-            echo "${not_ignored_files}"
+          modified_files="$(git status -s)"
+          echo "Modified files: ${modified_files}"
+          if [ -n "${modified_files}" ]; then
+            echo "Repository is dirty or server/webui is not built as expected"
+            echo "Hint: You may need to follow Web UI build guide in server/README.md"
+            echo "${modified_files}"
             exit 1
           fi
 

.gitignore

@@ -104,6 +104,10 @@ examples/server/*.mjs.hpp
 !examples/sycl/*.bat
 !examples/sycl/*.sh
 
+# Server Web UI temporary files
+node_modules
+examples/server/webui/dist
+
 # Python
 
 /.venv

CMakeLists.txt

@@ -96,10 +96,6 @@ if (NOT DEFINED GGML_LLAMAFILE)
     set(GGML_LLAMAFILE_DEFAULT ON)
 endif()
 
-if (NOT DEFINED GGML_AMX)
-    set(GGML_AMX ON)
-endif()
-
 if (NOT DEFINED GGML_CUDA_GRAPHS)
     set(GGML_CUDA_GRAPHS_DEFAULT ON)
 endif()

CODEOWNERS

@@ -0,0 +1,3 @@
+# collaborators can optionally add themselves here to indicate their availability for reviewing related PRs
+
+ci/ @ggerganov

CONTRIBUTING.md

@@ -1,9 +1,10 @@
 # Pull requests (for contributors)
 
 - Test your changes:
-  - Using the commands in the [`tests`](tests) folder. For instance, running the `./tests/test-backend-ops` command tests different backend implementations of the `ggml` library
   - Execute [the full CI locally on your machine](ci/README.md) before publishing
-- Optionally rate the complexity of your PR (i.e. `Review Complexity : Low`, `Review Complexity : Medium`, `Review Complexity : High`). This makes it easier for maintainers to triage the PRs
+  - Verify that the perplexity and the performance are not affected negatively by your changes (use `llama-perplexity` and `llama-bench`)
+  - If you modified the `ggml` source, run the `test-backend-ops` tool to check whether different backend implementations of the `ggml` operators produce consistent results (this requires access to at least two different `ggml` backends)
+  - If you modified a `ggml` operator or added a new one, add the corresponding test cases to `test-backend-ops`
 - Consider allowing write access to your branch for faster reviews, as reviewers can push commits directly
 - If your PR becomes stale, don't hesitate to ping the maintainers in the comments
 
@@ -12,6 +13,7 @@
 - Squash-merge PRs
 - Use the following format for the squashed commit title: `<module> : <commit title> (#<issue_number>)`. For example: `utils : fix typo in utils.py (#1234)`
 - Optionally pick a `<module>` from here: https://github.com/ggerganov/llama.cpp/wiki/Modules
+- Consider adding yourself to [CODEOWNERS](CODEOWNERS)
 
 # Coding guidelines
 

Makefile

@@ -1,3 +1,7 @@
+ifndef LLAMA_MAKEFILE
+$(error The Makefile build is deprecated. Use the CMake build instead. For more details, see https://github.com/ggerganov/llama.cpp/blob/master/docs/build.md)
+endif
+
 # Define the default target now so that it is always the first target
 BUILD_TARGETS = \
 	libllava.a \
@@ -251,11 +255,11 @@ endif
 # Compile flags
 #
 
-# keep standard at C11 and C++11
+# keep standard at C11 and C++17
 MK_CPPFLAGS  = -Iggml/include -Iggml/src -Iinclude -Isrc -Icommon -DGGML_USE_CPU
 MK_CFLAGS    = -std=c11   -fPIC
-MK_CXXFLAGS  = -std=c++11 -fPIC
-MK_NVCCFLAGS = -std=c++11
+MK_CXXFLAGS  = -std=c++17 -fPIC
+MK_NVCCFLAGS = -std=c++17
 
 ifdef LLAMA_NO_CCACHE
 GGML_NO_CCACHE := 1
@@ -441,6 +445,10 @@ ifeq ($(UNAME_M),$(filter $(UNAME_M),x86_64 i686 amd64))
 	MK_CFLAGS     += -march=native -mtune=native
 	HOST_CXXFLAGS += -march=native -mtune=native
 
+	# Usage AMX build test
+	#MK_CFLAGS     += -march=graniterapids -mtune=graniterapids
+	#HOST_CXXFLAGS += -march=graniterapids -mtune=graniterapids
+
 	# Usage AVX-only
 	#MK_CFLAGS   += -mfma -mf16c -mavx
 	#MK_CXXFLAGS += -mfma -mf16c -mavx
@@ -575,9 +583,12 @@ endif
 
 ifndef GGML_NO_AMX
 	MK_CPPFLAGS += -DGGML_USE_AMX
-	OBJ_GGML_EXT += ggml/src/ggml-amx/ggml-amx.o ggml/src/ggml-amx/mmq.o
+	OBJ_GGML_EXT += ggml/src/ggml-cpu/amx/amx.o ggml/src/ggml-cpu/amx/mmq.o
 endif
 
+# only necessary for the CPU backend files
+MK_CPPFLAGS += -Iggml/src/ggml-cpu
+
 ifdef GGML_RPC
 	MK_CPPFLAGS  += -DGGML_USE_RPC
 	OBJ_GGML_EXT += ggml/src/ggml-rpc.o
@@ -941,7 +952,6 @@ DIR_COMMON = common
 
 OBJ_GGML = \
 	$(DIR_GGML)/src/ggml.o \
-	$(DIR_GGML)/src/ggml-aarch64.o \
 	$(DIR_GGML)/src/ggml-alloc.o \
 	$(DIR_GGML)/src/ggml-backend.o \
 	$(DIR_GGML)/src/ggml-backend-reg.o \
@@ -949,9 +959,11 @@ OBJ_GGML = \
 	$(DIR_GGML)/src/ggml-quants.o \
 	$(DIR_GGML)/src/ggml-threading.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu.o \
-	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-cpp.o \
+	$(DIR_GGML)/src/ggml-cpu/ggml-cpu_cpp.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-aarch64.o \
+	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-hbm.o \
 	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-quants.o \
+	$(DIR_GGML)/src/ggml-cpu/ggml-cpu-traits.o \
 	$(OBJ_GGML_EXT)
 
 OBJ_LLAMA = \
@@ -1091,17 +1103,10 @@ DEP_FILES = $(OBJ_GGML:.o=.d) $(OBJ_LLAMA:.o=.d) $(OBJ_COMMON:.o=.d)
 # Default target
 all: $(BUILD_TARGETS)
 
+# force c++ build for source file that have same name as c file
 # Note: need this exception because `ggml-cpu.c` and `ggml-cpu.cpp` both produce the same obj/dep files
-#       g++ -M -I ./ggml/include/ -I ./ggml/src ggml/src/ggml-cpu/ggml-cpu.cpp | grep ggml
-$(DIR_GGML)/src/ggml-cpu/ggml-cpu-cpp.o: \
-	ggml/src/ggml-cpu/ggml-cpu.cpp \
-	ggml/include/ggml-backend.h \
-	ggml/include/ggml.h \
-	ggml/include/ggml-alloc.h \
-	ggml/src/ggml-backend-impl.h \
-	ggml/include/ggml-cpu.h \
-	ggml/src/ggml-impl.h
-	$(CXX) $(CXXFLAGS)   -c $< -o $@
+$(DIR_GGML)/%_cpp.o: $(DIR_GGML)/%.cpp
+	$(CXX) $(CXXFLAGS) -MMD -c $< -o $@
 
 # Rules for building object files
 $(DIR_GGML)/%.o: $(DIR_GGML)/%.c
@@ -1138,8 +1143,15 @@ $(LIB_COMMON_S): $(OBJ_COMMON)
 # Include dependency files
 -include $(DEP_FILES)
 
+# Clean generated server assets
+clean-server-assets:
+	find examples/server -type f -name "*.js.hpp"   -delete
+	find examples/server -type f -name "*.mjs.hpp"  -delete
+	find examples/server -type f -name "*.css.hpp"  -delete
+	find examples/server -type f -name "*.html.hpp" -delete
+
 # Clean rule
-clean:
+clean: clean-server-assets
 	rm -vrf $(BUILD_TARGETS) $(TEST_TARGETS)
 	rm -rvf *.a *.dll *.so *.dot
 	find ggml src common tests examples pocs -type f -name "*.o" -delete
@@ -1347,20 +1359,14 @@ llama-server: \
 	examples/server/utils.hpp \
 	examples/server/httplib.h \
 	examples/server/index.html.hpp \
-	examples/server/completion.js.hpp \
 	examples/server/loading.html.hpp \
-	examples/server/deps_daisyui.min.css.hpp \
-	examples/server/deps_markdown-it.js.hpp \
-	examples/server/deps_tailwindcss.js.hpp \
-	examples/server/deps_vue.esm-browser.js.hpp \
 	common/json.hpp \
-	common/stb_image.h \
 	$(OBJ_ALL)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h %.hpp $<,$^) -Iexamples/server $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS) $(LWINSOCK2)
 
 # Portable equivalent of `cd examples/server/public && xxd -i $(notdir $<) ../$(notdir $<).hpp`:
-examples/server/%.hpp: examples/server/public/% Makefile
+examples/server/%.hpp: examples/server/public/% FORCE Makefile
 	@( export NAME=$(subst .,_,$(subst -,_,$(notdir $<))) && \
 		echo "unsigned char $${NAME}[] = {" && \
 		cat $< | od -v -t x1 -An | sed -E 's/([0-9a-fA-F]+)/0x\1, /g' && \
@@ -1535,7 +1541,7 @@ llama-q8dot: pocs/vdot/q8dot.cpp ggml/src/ggml.o \
 # Deprecated binaries that we want to keep around long enough for people to migrate to the new filenames, then these can be removed.
 #
 # Mark legacy binary targets as .PHONY so that they are always checked.
-.PHONY: main quantize perplexity embedding server
+.PHONY: FORCE main quantize perplexity embedding server
 
 # Define the object file target
 examples/deprecation-warning/deprecation-warning.o: examples/deprecation-warning/deprecation-warning.cpp

Package.swift

@@ -10,14 +10,15 @@ var sources = [
     "src/unicode.cpp",
     "src/unicode-data.cpp",
     "ggml/src/ggml.c",
-    "ggml/src/ggml-aarch64.c",
     "ggml/src/ggml-alloc.c",
     "ggml/src/ggml-backend.cpp",
     "ggml/src/ggml-backend-reg.cpp",
     "ggml/src/ggml-cpu/ggml-cpu.c",
     "ggml/src/ggml-cpu/ggml-cpu.cpp",
-    "ggml/src/ggml-cpu/ggml-cpu-aarch64.c",
+    "ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp",
+    "ggml/src/ggml-cpu/ggml-cpu-hbm.cpp",
     "ggml/src/ggml-cpu/ggml-cpu-quants.c",
+    "ggml/src/ggml-cpu/ggml-cpu-traits.cpp",
     "ggml/src/ggml-threading.cpp",
     "ggml/src/ggml-quants.c",
 ]
@@ -28,13 +29,16 @@ var cSettings: [CSetting] =  [
     .unsafeFlags(["-Wno-shorten-64-to-32", "-O3", "-DNDEBUG"]),
     .unsafeFlags(["-fno-objc-arc"]),
     .headerSearchPath("ggml/src"),
+    .headerSearchPath("ggml/src/ggml-cpu"),
     // NOTE: NEW_LAPACK will required iOS version 16.4+
     // We should consider add this in the future when we drop support for iOS 14
     // (ref: ref: https://developer.apple.com/documentation/accelerate/1513264-cblas_sgemm?language=objc)
     // .define("ACCELERATE_NEW_LAPACK"),
     // .define("ACCELERATE_LAPACK_ILP64")
+    .define("GGML_USE_CPU"),
 ]
 
+
 #if canImport(Darwin)
 sources.append("ggml/src/ggml-common.h")
 sources.append("ggml/src/ggml-metal/ggml-metal.m")
@@ -44,7 +48,6 @@ cSettings.append(
     contentsOf: [
         .define("GGML_USE_ACCELERATE"),
         .define("GGML_USE_METAL"),
-        .define("GGML_USE_CPU")
     ]
 )
 #endif
@@ -86,5 +89,5 @@ let package = Package(
             linkerSettings: linkerSettings
         )
     ],
-    cxxLanguageStandard: .cxx11
+    cxxLanguageStandard: .cxx17
 )

README.md

@@ -4,7 +4,6 @@
 
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](https://opensource.org/licenses/MIT)
 [![Server](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml/badge.svg)](https://github.com/ggerganov/llama.cpp/actions/workflows/server.yml)
-[![Conan Center](https://shields.io/conan/v/llama-cpp)](https://conan.io/center/llama-cpp)
 
 [Roadmap](https://github.com/users/ggerganov/projects/7) / [Project status](https://github.com/ggerganov/llama.cpp/discussions/3471) / [Manifesto](https://github.com/ggerganov/llama.cpp/discussions/205) / [ggml](https://github.com/ggerganov/ggml)
 
@@ -26,7 +25,7 @@ Inference of Meta's [LLaMA](https://arxiv.org/abs/2302.13971) model (and others)
 ## Description
 
 The main goal of `llama.cpp` is to enable LLM inference with minimal setup and state-of-the-art performance on a wide
-variety of hardware - locally and in the cloud.
+range of hardware - locally and in the cloud.
 
 - Plain C/C++ implementation without any dependencies
 - Apple silicon is a first-class citizen - optimized via ARM NEON, Accelerate and Metal frameworks
@@ -36,14 +35,17 @@ variety of hardware - locally and in the cloud.
 - Vulkan and SYCL backend support
 - CPU+GPU hybrid inference to partially accelerate models larger than the total VRAM capacity
 
-Since its [inception](https://github.com/ggerganov/llama.cpp/issues/33#issuecomment-1465108022), the project has
-improved significantly thanks to many contributions. It is the main playground for developing new features for the
-[ggml](https://github.com/ggerganov/ggml) library.
+The `llama.cpp` project is the main playground for developing new features for the [ggml](https://github.com/ggerganov/ggml) library.
 
-**Supported models:**
+<details>
+<summary>Models</summary>
 
 Typically finetunes of the base models below are supported as well.
 
+Instructions for adding support for new models: [HOWTO-add-model.md](docs/development/HOWTO-add-model.md)
+
+#### Text-only
+
 - [X] LLaMA 🦙
 - [x] LLaMA 2 🦙🦙
 - [x] LLaMA 3 🦙🦙🦙
@@ -97,9 +99,7 @@ Typically finetunes of the base models below are supported as well.
 - [x] [Bielik-11B-v2.3](https://huggingface.co/collections/speakleash/bielik-11b-v23-66ee813238d9b526a072408a)
 - [x] [RWKV-6](https://github.com/BlinkDL/RWKV-LM)
 
-(instructions for supporting more models: [HOWTO-add-model.md](./docs/development/HOWTO-add-model.md))
-
-**Multimodal models:**
+#### Multimodal
 
 - [x] [LLaVA 1.5 models](https://huggingface.co/collections/liuhaotian/llava-15-653aac15d994e992e2677a7e), [LLaVA 1.6 models](https://huggingface.co/collections/liuhaotian/llava-16-65b9e40155f60fd046a5ccf2)
 - [x] [BakLLaVA](https://huggingface.co/models?search=SkunkworksAI/Bakllava)
@@ -111,7 +111,10 @@ Typically finetunes of the base models below are supported as well.
 - [x] [Moondream](https://huggingface.co/vikhyatk/moondream2)
 - [x] [Bunny](https://github.com/BAAI-DCAI/Bunny)
 
-**Bindings:**
+</details>
+
+<details>
+<summary>Bindings</summary>
 
 - Python: [abetlen/llama-cpp-python](https://github.com/abetlen/llama-cpp-python)
 - Go: [go-skynet/go-llama.cpp](https://github.com/go-skynet/go-llama.cpp)
@@ -138,316 +141,314 @@ Typically finetunes of the base models below are supported as well.
 - Swift [srgtuszy/llama-cpp-swift](https://github.com/srgtuszy/llama-cpp-swift)
 - Swift [ShenghaiWang/SwiftLlama](https://github.com/ShenghaiWang/SwiftLlama)
 
-**UI:**
+</details>
 
-Unless otherwise noted these projects are open-source with permissive licensing:
-
-- [MindWorkAI/AI-Studio](https://github.com/MindWorkAI/AI-Studio) (FSL-1.1-MIT)
-- [iohub/collama](https://github.com/iohub/coLLaMA)
-- [janhq/jan](https://github.com/janhq/jan) (AGPL)
-- [nat/openplayground](https://github.com/nat/openplayground)
-- [Faraday](https://faraday.dev/) (proprietary)
-- [LMStudio](https://lmstudio.ai/) (proprietary)
-- [Layla](https://play.google.com/store/apps/details?id=com.laylalite) (proprietary)
-- [ramalama](https://github.com/containers/ramalama) (MIT)
-- [LocalAI](https://github.com/mudler/LocalAI) (MIT)
-- [LostRuins/koboldcpp](https://github.com/LostRuins/koboldcpp) (AGPL)
-- [Mozilla-Ocho/llamafile](https://github.com/Mozilla-Ocho/llamafile)
-- [nomic-ai/gpt4all](https://github.com/nomic-ai/gpt4all)
-- [ollama/ollama](https://github.com/ollama/ollama)
-- [oobabooga/text-generation-webui](https://github.com/oobabooga/text-generation-webui) (AGPL)
-- [psugihara/FreeChat](https://github.com/psugihara/FreeChat)
-- [cztomsik/ava](https://github.com/cztomsik/ava) (MIT)
-- [ptsochantaris/emeltal](https://github.com/ptsochantaris/emeltal)
-- [pythops/tenere](https://github.com/pythops/tenere) (AGPL)
-- [RAGNA Desktop](https://ragna.app/) (proprietary)
-- [RecurseChat](https://recurse.chat/) (proprietary)
-- [semperai/amica](https://github.com/semperai/amica)
-- [withcatai/catai](https://github.com/withcatai/catai)
-- [Mobile-Artificial-Intelligence/maid](https://github.com/Mobile-Artificial-Intelligence/maid) (MIT)
-- [Msty](https://msty.app) (proprietary)
-- [LLMFarm](https://github.com/guinmoon/LLMFarm?tab=readme-ov-file) (MIT)
-- [KanTV](https://github.com/zhouwg/kantv?tab=readme-ov-file)(Apachev2.0 or later)
-- [Dot](https://github.com/alexpinel/Dot) (GPL)
-- [MindMac](https://mindmac.app) (proprietary)
-- [KodiBot](https://github.com/firatkiral/kodibot) (GPL)
-- [eva](https://github.com/ylsdamxssjxxdd/eva) (MIT)
-- [AI Sublime Text plugin](https://github.com/yaroslavyaroslav/OpenAI-sublime-text) (MIT)
-- [AIKit](https://github.com/sozercan/aikit) (MIT)
-- [LARS - The LLM & Advanced Referencing Solution](https://github.com/abgulati/LARS) (AGPL)
-- [LLMUnity](https://github.com/undreamai/LLMUnity) (MIT)
-- [Llama Assistant](https://github.com/vietanhdev/llama-assistant) (GPL)
-- [PocketPal AI - An iOS and Android App](https://github.com/a-ghorbani/pocketpal-ai) (MIT)
+<details>
+<summary>UIs</summary>
 
 *(to have a project listed here, it should clearly state that it depends on `llama.cpp`)*
 
-**Tools:**
+- [AI Sublime Text plugin](https://github.com/yaroslavyaroslav/OpenAI-sublime-text) (MIT)
+- [cztomsik/ava](https://github.com/cztomsik/ava) (MIT)
+- [Dot](https://github.com/alexpinel/Dot) (GPL)
+- [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)
+- [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)
+- [LARS](https://github.com/abgulati/LARS) (AGPL)
+- [Llama Assistant](https://github.com/vietanhdev/llama-assistant) (GPL)
+- [LLMFarm](https://github.com/guinmoon/LLMFarm?tab=readme-ov-file) (MIT)
+- [LLMUnity](https://github.com/undreamai/LLMUnity) (MIT)
+- [LMStudio](https://lmstudio.ai/) (proprietary)
+- [LocalAI](https://github.com/mudler/LocalAI) (MIT)
+- [LostRuins/koboldcpp](https://github.com/LostRuins/koboldcpp) (AGPL)
+- [MindMac](https://mindmac.app) (proprietary)
+- [MindWorkAI/AI-Studio](https://github.com/MindWorkAI/AI-Studio) (FSL-1.1-MIT)
+- [Mobile-Artificial-Intelligence/maid](https://github.com/Mobile-Artificial-Intelligence/maid) (MIT)
+- [Mozilla-Ocho/llamafile](https://github.com/Mozilla-Ocho/llamafile) (Apache-2.0)
+- [nat/openplayground](https://github.com/nat/openplayground) (MIT)
+- [nomic-ai/gpt4all](https://github.com/nomic-ai/gpt4all) (MIT)
+- [ollama/ollama](https://github.com/ollama/ollama) (MIT)
+- [oobabooga/text-generation-webui](https://github.com/oobabooga/text-generation-webui) (AGPL)
+- [PocketPal AI](https://github.com/a-ghorbani/pocketpal-ai) (MIT)
+- [psugihara/FreeChat](https://github.com/psugihara/FreeChat) (MIT)
+- [ptsochantaris/emeltal](https://github.com/ptsochantaris/emeltal) (MIT)
+- [pythops/tenere](https://github.com/pythops/tenere) (AGPL)
+- [ramalama](https://github.com/containers/ramalama) (MIT)
+- [semperai/amica](https://github.com/semperai/amica) (MIT)
+- [withcatai/catai](https://github.com/withcatai/catai) (MIT)
+
+</details>
+
+<details>
+<summary>Tools</summary>
 
 - [akx/ggify](https://github.com/akx/ggify) – download PyTorch models from HuggingFace Hub and convert them to GGML
 - [akx/ollama-dl](https://github.com/akx/ollama-dl) – download models from the Ollama library to be used directly with llama.cpp
 - [crashr/gppm](https://github.com/crashr/gppm) – launch llama.cpp instances utilizing NVIDIA Tesla P40 or P100 GPUs with reduced idle power consumption
 - [gpustack/gguf-parser](https://github.com/gpustack/gguf-parser-go/tree/main/cmd/gguf-parser) - review/check the GGUF file and estimate the memory usage
-- [Styled Lines](https://marketplace.unity.com/packages/tools/generative-ai/styled-lines-llama-cpp-model-292902) (proprietary licensed, async wrapper of inference part for game development in Unity3d with prebuild Mobile and Web platform wrappers and a model example)
+- [Styled Lines](https://marketplace.unity.com/packages/tools/generative-ai/styled-lines-llama-cpp-model-292902) (proprietary licensed, async wrapper of inference part for game development in Unity3d with pre-built Mobile and Web platform wrappers and a model example)
 
-**Infrastructure:**
+</details>
+
+<details>
+<summary>Infrastructure</summary>
 
 - [Paddler](https://github.com/distantmagic/paddler) - Stateful load balancer custom-tailored for llama.cpp
 - [GPUStack](https://github.com/gpustack/gpustack) - Manage GPU clusters for running LLMs
 - [llama_cpp_canister](https://github.com/onicai/llama_cpp_canister) - llama.cpp as a smart contract on the Internet Computer, using WebAssembly
 
-**Games:**
+</details>
+
+<details>
+<summary>Games</summary>
+
 - [Lucy's Labyrinth](https://github.com/MorganRO8/Lucys_Labyrinth) - A simple maze game where agents controlled by an AI model will try to trick you.
 
-## Demo
-
-<details>
-<summary>Typical run using LLaMA v2 13B on M2 Ultra</summary>
-
-```
-$ make -j && ./llama-cli -m models/llama-13b-v2/ggml-model-q4_0.gguf -p "Building a website can be done in 10 simple steps:\nStep 1:" -n 400 -e
-I llama.cpp build info:
-I UNAME_S:  Darwin
-I UNAME_P:  arm
-I UNAME_M:  arm64
-I CFLAGS:   -I.            -O3 -std=c11   -fPIC -DNDEBUG -Wall -Wextra -Wpedantic -Wcast-qual -Wdouble-promotion -Wshadow -Wstrict-prototypes -Wpointer-arith -Wmissing-prototypes -pthread -DGGML_USE_K_QUANTS -DGGML_USE_ACCELERATE
-I CXXFLAGS: -I. -I./common -O3 -std=c++11 -fPIC -DNDEBUG -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wno-multichar -pthread -DGGML_USE_K_QUANTS
-I LDFLAGS:   -framework Accelerate
-I CC:       Apple clang version 14.0.3 (clang-1403.0.22.14.1)
-I CXX:      Apple clang version 14.0.3 (clang-1403.0.22.14.1)
-
-make: Nothing to be done for `default'.
-main: build = 1041 (cf658ad)
-main: seed  = 1692823051
-llama_model_loader: loaded meta data with 16 key-value pairs and 363 tensors from models/llama-13b-v2/ggml-model-q4_0.gguf (version GGUF V1 (latest))
-llama_model_loader: - type  f32:   81 tensors
-llama_model_loader: - type q4_0:  281 tensors
-llama_model_loader: - type q6_K:    1 tensors
-llm_load_print_meta: format         = GGUF V1 (latest)
-llm_load_print_meta: arch           = llama
-llm_load_print_meta: vocab type     = SPM
-llm_load_print_meta: n_vocab        = 32000
-llm_load_print_meta: n_merges       = 0
-llm_load_print_meta: n_ctx_train    = 4096
-llm_load_print_meta: n_ctx          = 512
-llm_load_print_meta: n_embd         = 5120
-llm_load_print_meta: n_head         = 40
-llm_load_print_meta: n_head_kv      = 40
-llm_load_print_meta: n_layer        = 40
-llm_load_print_meta: n_rot          = 128
-llm_load_print_meta: n_gqa          = 1
-llm_load_print_meta: f_norm_eps     = 1.0e-05
-llm_load_print_meta: f_norm_rms_eps = 1.0e-05
-llm_load_print_meta: n_ff           = 13824
-llm_load_print_meta: freq_base      = 10000.0
-llm_load_print_meta: freq_scale     = 1
-llm_load_print_meta: model type     = 13B
-llm_load_print_meta: model ftype    = mostly Q4_0
-llm_load_print_meta: model size     = 13.02 B
-llm_load_print_meta: general.name   = LLaMA v2
-llm_load_print_meta: BOS token = 1 '<s>'
-llm_load_print_meta: EOS token = 2 '</s>'
-llm_load_print_meta: UNK token = 0 '<unk>'
-llm_load_print_meta: LF token  = 13 '<0x0A>'
-llm_load_tensors: ggml ctx size =    0.11 MB
-llm_load_tensors: mem required  = 7024.01 MB (+  400.00 MB per state)
-...................................................................................................
-llama_new_context_with_model: kv self size  =  400.00 MB
-llama_new_context_with_model: compute buffer total size =   75.41 MB
-
-system_info: n_threads = 16 / 24 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 |
-sampling: repeat_last_n = 64, repeat_penalty = 1.100000, presence_penalty = 0.000000, frequency_penalty = 0.000000, top_k = 40, tfs_z = 1.000000, top_p = 0.950000, typical_p = 1.000000, temp = 0.800000, mirostat = 0, mirostat_lr = 0.100000, mirostat_ent = 5.000000
-generate: n_ctx = 512, n_batch = 512, n_predict = 400, n_keep = 0
-
-
- Building a website can be done in 10 simple steps:
-Step 1: Find the right website platform.
-Step 2: Choose your domain name and hosting plan.
-Step 3: Design your website layout.
-Step 4: Write your website content and add images.
-Step 5: Install security features to protect your site from hackers or spammers
-Step 6: Test your website on multiple browsers, mobile devices, operating systems etc…
-Step 7: Test it again with people who are not related to you personally – friends or family members will work just fine!
-Step 8: Start marketing and promoting the website via social media channels or paid ads
-Step 9: Analyze how many visitors have come to your site so far, what type of people visit more often than others (e.g., men vs women) etc…
-Step 10: Continue to improve upon all aspects mentioned above by following trends in web design and staying up-to-date on new technologies that can enhance user experience even further!
-How does a Website Work?
-A website works by having pages, which are made of HTML code. This code tells your computer how to display the content on each page you visit – whether it’s an image or text file (like PDFs). In order for someone else’s browser not only be able but also want those same results when accessing any given URL; some additional steps need taken by way of programming scripts that will add functionality such as making links clickable!
-The most common type is called static HTML pages because they remain unchanged over time unless modified manually (either through editing files directly or using an interface such as WordPress). They are usually served up via HTTP protocols – this means anyone can access them without having any special privileges like being part of a group who is allowed into restricted areas online; however, there may still exist some limitations depending upon where one lives geographically speaking.
-How to
-llama_print_timings:        load time =   576.45 ms
-llama_print_timings:      sample time =   283.10 ms /   400 runs   (    0.71 ms per token,  1412.91 tokens per second)
-llama_print_timings: prompt eval time =   599.83 ms /    19 tokens (   31.57 ms per token,    31.68 tokens per second)
-llama_print_timings:        eval time = 24513.59 ms /   399 runs   (   61.44 ms per token,    16.28 tokens per second)
-llama_print_timings:       total time = 25431.49 ms
-```
-
 </details>
 
-<details>
-<summary>Demo of running both LLaMA-7B and whisper.cpp on a single M1 Pro MacBook</summary>
-
-And here is another demo of running both LLaMA-7B and [whisper.cpp](https://github.com/ggerganov/whisper.cpp) on a single M1 Pro MacBook:
-
-https://user-images.githubusercontent.com/1991296/224442907-7693d4be-acaa-4e01-8b4f-add84093ffff.mp4
-
-</details>
-
-## Usage
-
-Here are the end-to-end binary build and model conversion steps for most supported models.
-
-### Basic usage
-
-Firstly, you need to get the binary. There are different methods that you can follow:
-- Method 1: Clone this repository and build locally, see [how to build](./docs/build.md)
-- Method 2: If you are using MacOS or Linux, you can install llama.cpp via [brew, flox or nix](./docs/install.md)
-- Method 3: Use a Docker image, see [documentation for Docker](./docs/docker.md)
-- Method 4: Download pre-built binary from [releases](https://github.com/ggerganov/llama.cpp/releases)
-
-You can run a basic completion using this command:
-
-```bash
-llama-cli -m your_model.gguf -p "I believe the meaning of life is" -n 128
-
-# Output:
-# I believe the meaning of life is to find your own truth and to live in accordance with it. For me, this means being true to myself and following my passions, even if they don't align with societal expectations. I think that's what I love about yoga – it's not just a physical practice, but a spiritual one too. It's about connecting with yourself, listening to your inner voice, and honoring your own unique journey.
-```
-
-See [this page](./examples/main/README.md) for a full list of parameters.
-
-### Conversation mode
-
-If you want a more ChatGPT-like experience, you can run in conversation mode by passing `-cnv` as a parameter:
-
-```bash
-llama-cli -m your_model.gguf -p "You are a helpful assistant" -cnv
-
-# Output:
-# > hi, who are you?
-# Hi there! I'm your helpful assistant! I'm an AI-powered chatbot designed to assist and provide information to users like you. I'm here to help answer your questions, provide guidance, and offer support on a wide range of topics. I'm a friendly and knowledgeable AI, and I'm always happy to help with anything you need. What's on your mind, and how can I assist you today?
-#
-# > what is 1+1?
-# Easy peasy! The answer to 1+1 is... 2!
-```
-
-By default, the chat template will be taken from the input model. If you want to use another chat template, pass `--chat-template NAME` as a parameter. See the list of [supported templates](https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template)
-
-```bash
-./llama-cli -m your_model.gguf -p "You are a helpful assistant" -cnv --chat-template chatml
-```
-
-You can also use your own template via in-prefix, in-suffix and reverse-prompt parameters:
-
-```bash
-./llama-cli -m your_model.gguf -p "You are a helpful assistant" -cnv --in-prefix 'User: ' --reverse-prompt 'User:'
-```
-
-### Web server
-
-[llama.cpp web server](./examples/server/README.md) is a lightweight [OpenAI API](https://github.com/openai/openai-openapi) compatible HTTP server that can be used to serve local models and easily connect them to existing clients.
-
-Example usage:
-
-```bash
-./llama-server -m your_model.gguf --port 8080
-
-# Basic web UI can be accessed via browser: http://localhost:8080
-# Chat completion endpoint: http://localhost:8080/v1/chat/completions
-```
-
-### Interactive mode
-
-> [!NOTE]
-> If you prefer basic usage, please consider using conversation mode instead of interactive mode
-
-In this mode, you can always interrupt generation by pressing Ctrl+C and entering one or more lines of text, which will be converted into tokens and appended to the current context. You can also specify a *reverse prompt* with the parameter `-r "reverse prompt string"`. This will result in user input being prompted whenever the exact tokens of the reverse prompt string are encountered in the generation. A typical use is to use a prompt that makes LLaMA emulate a chat between multiple users, say Alice and Bob, and pass `-r "Alice:"`.
-
-Here is an example of a few-shot interaction, invoked with the command
-
-```bash
-# default arguments using a 7B model
-./examples/chat.sh
-
-# advanced chat with a 13B model
-./examples/chat-13B.sh
-
-# custom arguments using a 13B model
-./llama-cli -m ./models/13B/ggml-model-q4_0.gguf -n 256 --repeat_penalty 1.0 --color -i -r "User:" -f prompts/chat-with-bob.txt
-```
-
-Note the use of `--color` to distinguish between user input and generated text. Other parameters are explained in more detail in the [README](examples/main/README.md) for the `llama-cli` example program.
-
-![image](https://user-images.githubusercontent.com/1991296/224575029-2af3c7dc-5a65-4f64-a6bb-517a532aea38.png)
-
-### Persistent Interaction
-
-The prompt, user inputs, and model generations can be saved and resumed across calls to `./llama-cli` by leveraging `--prompt-cache` and `--prompt-cache-all`. The `./examples/chat-persistent.sh` script demonstrates this with support for long-running, resumable chat sessions. To use this example, you must provide a file to cache the initial chat prompt and a directory to save the chat session, and may optionally provide the same variables as `chat-13B.sh`. The same prompt cache can be reused for new chat sessions. Note that both prompt cache and chat directory are tied to the initial prompt (`PROMPT_TEMPLATE`) and the model file.
-
-```bash
-# Start a new chat
-PROMPT_CACHE_FILE=chat.prompt.bin CHAT_SAVE_DIR=./chat/default ./examples/chat-persistent.sh
-
-# Resume that chat
-PROMPT_CACHE_FILE=chat.prompt.bin CHAT_SAVE_DIR=./chat/default ./examples/chat-persistent.sh
-
-# Start a different chat with the same prompt/model
-PROMPT_CACHE_FILE=chat.prompt.bin CHAT_SAVE_DIR=./chat/another ./examples/chat-persistent.sh
-
-# Different prompt cache for different prompt/model
-PROMPT_TEMPLATE=./prompts/chat-with-bob.txt PROMPT_CACHE_FILE=bob.prompt.bin \
-    CHAT_SAVE_DIR=./chat/bob ./examples/chat-persistent.sh
-```
-
-### Constrained output with grammars
-
-`llama.cpp` supports grammars to constrain model output. For example, you can force the model to output JSON only:
-
-```bash
-./llama-cli -m ./models/13B/ggml-model-q4_0.gguf -n 256 --grammar-file grammars/json.gbnf -p 'Request: schedule a call at 8pm; Command:'
-```
-
-The `grammars/` folder contains a handful of sample grammars. To write your own, check out the [GBNF Guide](./grammars/README.md).
-
-For authoring more complex JSON grammars, you can also check out https://grammar.intrinsiclabs.ai/, a browser app that lets you write TypeScript interfaces which it compiles to GBNF grammars that you can save for local use. Note that the app is built and maintained by members of the community, please file any issues or FRs on [its repo](http://github.com/intrinsiclabsai/gbnfgen) and not this one.
-
-## Build
-
-Please refer to [Build llama.cpp locally](./docs/build.md)
-
 ## Supported backends
 
 | Backend | Target devices |
 | --- | --- |
-| [Metal](./docs/build.md#metal-build) | Apple Silicon |
-| [BLAS](./docs/build.md#blas-build) | All |
-| [BLIS](./docs/backend/BLIS.md) | All |
-| [SYCL](./docs/backend/SYCL.md) | Intel and Nvidia GPU |
-| [MUSA](./docs/build.md#musa) | Moore Threads MTT GPU |
-| [CUDA](./docs/build.md#cuda) | Nvidia GPU |
-| [hipBLAS](./docs/build.md#hipblas) | AMD GPU |
-| [Vulkan](./docs/build.md#vulkan) | GPU |
-| [CANN](./docs/build.md#cann) | Ascend NPU |
+| [Metal](docs/build.md#metal-build) | Apple Silicon |
+| [BLAS](docs/build.md#blas-build) | All |
+| [BLIS](docs/backend/BLIS.md) | All |
+| [SYCL](docs/backend/SYCL.md) | Intel and Nvidia GPU |
+| [MUSA](docs/build.md#musa) | Moore Threads MTT GPU |
+| [CUDA](docs/build.md#cuda) | Nvidia GPU |
+| [hipBLAS](docs/build.md#hipblas) | AMD GPU |
+| [Vulkan](docs/build.md#vulkan) | GPU |
+| [CANN](docs/build.md#cann) | Ascend NPU |
 
-## Tools
+## Building the project
 
-### Prepare and Quantize
+The main product of this project is the `llama` library. Its C-style interface can be found in [include/llama.h](include/llama.h).
+The project also includes many example programs and tools using the `llama` library. The examples range from simple, minimal code snippets to sophisticated sub-projects such as an OpenAI-compatible HTTP server. Possible methods for obtaining the binaries:
 
-> [!NOTE]
-> You can use the [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space on Hugging Face to quantise your model weights without any setup too. It is synced from `llama.cpp` main every 6 hours.
+- Clone this repository and build locally, see [how to build](docs/build.md)
+- On MacOS or Linux, install `llama.cpp` via [brew, flox or nix](docs/install.md)
+- Use a Docker image, see [documentation for Docker](docs/docker.md)
+- Download pre-built binaries from [releases](https://github.com/ggerganov/llama.cpp/releases)
 
-To obtain the official LLaMA 2 weights please see the <a href="#obtaining-and-using-the-facebook-llama-2-model">Obtaining and using the Facebook LLaMA 2 model</a> section. There is also a large selection of pre-quantized `gguf` models available on Hugging Face.
+## Obtaining and quantizing models
 
-Note: `convert.py` has been moved to `examples/convert_legacy_llama.py` and shouldn't be used for anything other than `Llama/Llama2/Mistral` models and their derivatives.
-It does not support LLaMA 3, you can use `convert_hf_to_gguf.py` with LLaMA 3 downloaded from Hugging Face.
+The [Hugging Face](https://huggingface.co) platform hosts a [number of LLMs](https://huggingface.co/models?library=gguf&sort=trending) compatible with `llama.cpp`:
 
-To learn more about quantizing model, [read this documentation](./examples/quantize/README.md)
+- [Trending](https://huggingface.co/models?library=gguf&sort=trending)
+- [LLaMA](https://huggingface.co/models?sort=trending&search=llama+gguf)
 
-### Perplexity (measuring model quality)
+After downloading a model, use the CLI tools to run it locally - see below.
 
-You can use the `perplexity` example to measure perplexity over a given prompt (lower perplexity is better).
-For more information, see [https://huggingface.co/docs/transformers/perplexity](https://huggingface.co/docs/transformers/perplexity).
+`llama.cpp` requires the model to be stored in the [GGUF](https://github.com/ggerganov/ggml/blob/master/docs/gguf.md) file format. Models in other data formats can be converted to GGUF using the `convert_*.py` Python scripts in this repo.
+
+The Hugging Face platform provides a variety of online tools for converting, quantizing and hosting models with `llama.cpp`:
+
+- Use the [GGUF-my-repo space](https://huggingface.co/spaces/ggml-org/gguf-my-repo) to convert to GGUF format and quantize model weights to smaller sizes
+- Use the [GGUF-my-LoRA space](https://huggingface.co/spaces/ggml-org/gguf-my-lora) to convert LoRA adapters to GGUF format (more info: https://github.com/ggerganov/llama.cpp/discussions/10123)
+- Use the [GGUF-editor space](https://huggingface.co/spaces/CISCai/gguf-editor) to edit GGUF meta data in the browser (more info: https://github.com/ggerganov/llama.cpp/discussions/9268)
+- Use the [Inference Endpoints](https://ui.endpoints.huggingface.co/) to directly host `llama.cpp` in the cloud (more info: https://github.com/ggerganov/llama.cpp/discussions/9669)
+
+To learn more about model quantization, [read this documentation](examples/quantize/README.md)
+
+## [`llama-cli`](examples/main)
+
+#### A CLI tool for accessing and experimenting with most of `llama.cpp`'s functionality.
+
+- <details open>
+    <summary>Run simple text completion</summary>
+
+    ```bash
+    llama-cli -m model.gguf -p "I believe the meaning of life is" -n 128
+
+    # I believe the meaning of life is to find your own truth and to live in accordance with it. For me, this means being true to myself and following my passions, even if they don't align with societal expectations. I think that's what I love about yoga – it's not just a physical practice, but a spiritual one too. It's about connecting with yourself, listening to your inner voice, and honoring your own unique journey.
+    ```
+
+    </details>
+
+- <details>
+    <summary>Run in conversation mode</summary>
+
+    ```bash
+    llama-cli -m model.gguf -p "You are a helpful assistant" -cnv
+
+    # > hi, who are you?
+    # Hi there! I'm your helpful assistant! I'm an AI-powered chatbot designed to assist and provide information to users like you. I'm here to help answer your questions, provide guidance, and offer support on a wide range of topics. I'm a friendly and knowledgeable AI, and I'm always happy to help with anything you need. What's on your mind, and how can I assist you today?
+    #
+    # > what is 1+1?
+    # Easy peasy! The answer to 1+1 is... 2!
+    ```
+
+    </details>
+
+- <details>
+    <summary>Run with custom chat template</summary>
+
+    ```bash
+    # use the "chatml" template
+    llama-cli -m model.gguf -p "You are a helpful assistant" -cnv --chat-template chatml
+
+    # use a custom template
+    llama-cli -m model.gguf -p "You are a helpful assistant" -cnv --in-prefix 'User: ' --reverse-prompt 'User:'
+    ```
+
+    [Supported templates](https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template)
+
+    </details>
+
+- <details>
+    <summary>Constrain the output with a custom grammar</summary>
+
+    ```bash
+    llama-cli -m model.gguf -n 256 --grammar-file grammars/json.gbnf -p 'Request: schedule a call at 8pm; Command:'
+
+    # {"appointmentTime": "8pm", "appointmentDetails": "schedule a a call"}
+    ```
+
+    The [grammars/](grammars/) folder contains a handful of sample grammars. To write your own, check out the [GBNF Guide](grammars/README.md).
+
+    For authoring more complex JSON grammars, check out https://grammar.intrinsiclabs.ai/
+
+    </details>
+
+
+## [`llama-server`](examples/server)
+
+#### A lightweight, [OpenAI API](https://github.com/openai/openai-openapi) compatible, HTTP server for serving LLMs.
+
+- <details open>
+    <summary>Start a local HTTP server with default configuration on port 8080</summary>
+
+    ```bash
+    llama-server -m model.gguf --port 8080
+
+    # Basic web UI can be accessed via browser: http://localhost:8080
+    # Chat completion endpoint: http://localhost:8080/v1/chat/completions
+    ```
+
+    </details>
+
+- <details>
+    <summary>Support multiple-users and parallel decoding</summary>
+
+    ```bash
+    # up to 4 concurrent requests, each with 4096 max context
+    llama-server -m model.gguf -c 16384 -np 4
+    ```
+
+    </details>
+
+- <details>
+    <summary>Enable speculative decoding</summary>
+
+    ```bash
+    # the draft.gguf model should be a small variant of the target model.gguf
+    llama-server -m model.gguf -md draft.gguf
+    ```
+
+    </details>
+
+- <details>
+    <summary>Serve an embedding model</summary>
+
+    ```bash
+    # use the /embedding endpoint
+    llama-server -m model.gguf --embedding --pooling cls -ub 8192
+    ```
+
+    </details>
+
+- <details>
+    <summary>Serve a reranking model</summary>
+
+    ```bash
+    # use the /reranking endpoint
+    llama-server -m model.gguf --reranking
+    ```
+
+    </details>
+
+- <details>
+    <summary>Constrain all outputs with a grammar</summary>
+
+    ```bash
+    # custom grammar
+    llama-server -m model.gguf --grammar-file grammar.gbnf
+
+    # JSON
+    llama-server -m model.gguf --grammar-file grammars/json.gbnf
+    ```
+
+    </details>
+
+
+## [`llama-perplexity`](examples/perplexity)
+
+#### A tool for measuring the perplexity [^1][^2] (and other quality metrics) of a model over a given text.
+
+- <details open>
+    <summary>Measure the perplexity over a text file</summary>
+
+    ```bash
+    llama-perplexity -m model.gguf -f file.txt
+
+    # [1]15.2701,[2]5.4007,[3]5.3073,[4]6.2965,[5]5.8940,[6]5.6096,[7]5.7942,[8]4.9297, ...
+    # Final estimate: PPL = 5.4007 +/- 0.67339
+    ```
+
+    </details>
+
+- <details>
+    <summary>Measure KL divergence</summary>
+
+    ```bash
+    # TODO
+    ```
+
+    </details>
+
+[^1]: [examples/perplexity/README.md](examples/perplexity/README.md)
+[^2]: [https://huggingface.co/docs/transformers/perplexity](https://huggingface.co/docs/transformers/perplexity)
+
+## [`llama-bench`](example/bench)
+
+#### Benchmark the performance of the inference for various parameters.
+
+- <details open>
+    <summary>Run default benchmark</summary>
+
+    ```bash
+    llama-bench -m model.gguf
+
+    # Output:
+    # | model               |       size |     params | backend    | threads |          test |                  t/s |
+    # | ------------------- | ---------: | ---------: | ---------- | ------: | ------------: | -------------------: |
+    # | qwen2 1.5B Q4_0     | 885.97 MiB |     1.54 B | Metal,BLAS |      16 |         pp512 |      5765.41 ± 20.55 |
+    # | qwen2 1.5B Q4_0     | 885.97 MiB |     1.54 B | Metal,BLAS |      16 |         tg128 |        197.71 ± 0.81 |
+    #
+    # build: 3e0ba0e60 (4229)
+    ```
+
+    </details>
+
+
+## [`llama-simple`](examples/simple)
+
+#### A minimal example for implementing apps with `llama.cpp`. Useful for developers.
+
+- <details>
+    <summary>Basic text completion</summary>
+
+    ```bash
+    llama-simple -m model.gguf
+
+    # Hello my name is Kaitlyn and I am a 16 year old girl. I am a junior in high school and I am currently taking a class called "The Art of
+    ```
+
+    </details>
 
-To learn more how to measure perplexity using llama.cpp, [read this documentation](./examples/perplexity/README.md)
 
 ## Contributing
 
@@ -462,20 +463,19 @@ To learn more how to measure perplexity using llama.cpp, [read this documentatio
 
 ## Other documentation
 
-- [main (cli)](./examples/main/README.md)
-- [server](./examples/server/README.md)
-- [jeopardy](./examples/jeopardy/README.md)
-- [GBNF grammars](./grammars/README.md)
+- [main (cli)](examples/main/README.md)
+- [server](examples/server/README.md)
+- [GBNF grammars](grammars/README.md)
 
-**Development documentation**
+#### Development documentation
 
-- [How to build](./docs/build.md)
-- [Running on Docker](./docs/docker.md)
-- [Build on Android](./docs/android.md)
-- [Performance troubleshooting](./docs/development/token_generation_performance_tips.md)
+- [How to build](docs/build.md)
+- [Running on Docker](docs/docker.md)
+- [Build on Android](docs/android.md)
+- [Performance troubleshooting](docs/development/token_generation_performance_tips.md)
 - [GGML tips & tricks](https://github.com/ggerganov/llama.cpp/wiki/GGML-Tips-&-Tricks)
 
-**Seminal papers and background on the models**
+#### Seminal papers and background on the models
 
 If your issue is with model generation quality, then please at least scan the following links and papers to understand the limitations of LLaMA models. This is especially important when choosing an appropriate model size and appreciating both the significant and subtle differences between LLaMA models and ChatGPT:
 - LLaMA:
@@ -486,3 +486,6 @@ If your issue is with model generation quality, then please at least scan the fo
 - GPT-3.5 / InstructGPT / ChatGPT:
     - [Aligning language models to follow instructions](https://openai.com/research/instruction-following)
     - [Training language models to follow instructions with human feedback](https://arxiv.org/abs/2203.02155)
+
+#### References
+

ci/run.sh

@@ -815,7 +815,10 @@ if [ -z ${GG_BUILD_LOW_PERF} ]; then
     ln -sfn ${mnt_models} ${SRC}/models-mnt
 
     # Create a fresh python3 venv and enter it
-    python3 -m venv "$MNT/venv"
+    if ! python3 -m venv "$MNT/venv"; then
+        echo "Error: Failed to create Python virtual environment at $MNT/venv."
+        exit 1
+    fi
     source "$MNT/venv/bin/activate"
 
     pip install -r ${SRC}/requirements.txt --disable-pip-version-check

common/CMakeLists.txt

@@ -88,5 +88,5 @@ if (LLAMA_CURL)
 endif ()
 
 target_include_directories(${TARGET} PUBLIC .)
-target_compile_features   (${TARGET} PUBLIC cxx_std_11)
+target_compile_features   (${TARGET} PUBLIC cxx_std_17)
 target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)

common/arg.cpp

@@ -348,6 +348,18 @@ bool common_params_parse(int argc, char ** argv, common_params & params, llama_e
     return true;
 }
 
+static std::string list_builtin_chat_templates() {
+    std::vector<const char *> supported_tmpl;
+    int32_t res = llama_chat_builtin_templates(nullptr, 0);
+    supported_tmpl.resize(res);
+    res = llama_chat_builtin_templates(supported_tmpl.data(), supported_tmpl.size());
+    std::ostringstream msg;
+    for (auto & tmpl : supported_tmpl) {
+        msg << tmpl << (&tmpl == &supported_tmpl.back() ? "" : ", ");
+    }
+    return msg.str();
+}
+
 common_params_context common_params_parser_init(common_params & params, llama_example ex, void(*print_usage)(int, char **)) {
     // load dynamic backends
     ggml_backend_load_all();
@@ -774,7 +786,7 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
         [](common_params & params) {
             params.warmup = false;
         }
-    ).set_examples({LLAMA_EXAMPLE_MAIN}));
+    ).set_examples({LLAMA_EXAMPLE_MAIN, LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--spm-infill"},
         string_format(
@@ -1814,9 +1826,11 @@ common_params_context common_params_parser_init(common_params & params, llama_ex
     ).set_examples({LLAMA_EXAMPLE_SERVER}));
     add_opt(common_arg(
         {"--chat-template"}, "JINJA_TEMPLATE",
-        "set custom jinja chat template (default: template taken from model's metadata)\n"
-        "if suffix/prefix are specified, template will be disabled\n"
-        "only commonly used templates are accepted:\nhttps://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template",
+        string_format(
+            "set custom jinja chat template (default: template taken from model's metadata)\n"
+            "if suffix/prefix are specified, template will be disabled\n"
+            "list of built-in templates:\n%s", list_builtin_chat_templates().c_str()
+        ),
         [](common_params & params, const std::string & value) {
             if (!common_chat_verify_template(value)) {
                 throw std::runtime_error(string_format(

common/common.cpp

@@ -652,7 +652,17 @@ bool fs_validate_filename(const std::string & filename) {
 
     std::u32string filename_utf32;
     try {
+#if defined(__clang__)
+        // disable C++17 deprecation warning for std::codecvt_utf8
+#    pragma clang diagnostic push
+#    pragma clang diagnostic ignored "-Wdeprecated-declarations"
+#endif
         std::wstring_convert<std::codecvt_utf8<char32_t>, char32_t> converter;
+
+#if defined(__clang__)
+#    pragma clang diagnostic pop
+#endif
+
         filename_utf32 = converter.from_bytes(filename);
 
         // If the reverse conversion mismatches, it means overlong UTF-8 sequences were used,

common/common.h

@@ -133,6 +133,7 @@ struct common_params_sampling {
     bool    penalize_nl        = false; // consider newlines as a repeatable token
     bool    ignore_eos         = false;
     bool    no_perf            = false; // disable performance metrics
+    bool    timing_per_token   = false;
 
     std::vector<std::string> dry_sequence_breakers = {"\n", ":", "\"", "*"};     // default sequence breakers for DRY
 
@@ -214,7 +215,7 @@ struct common_params {
     struct common_params_speculative speculative;
 
     std::string model                = ""; // model path                                                    // NOLINT
-    std::string model_alias          = "unknown"; // model alias                                            // NOLINT
+    std::string model_alias          = ""; // model alias                                                   // NOLINT
     std::string model_url            = ""; // model url to download                                         // NOLINT
     std::string hf_token             = ""; // HF token                                                      // NOLINT
     std::string hf_repo              = ""; // HF repo                                                       // NOLINT

common/speculative.cpp

@@ -62,6 +62,10 @@ struct common_speculative * common_speculative_init(
 }
 
 void common_speculative_free(struct common_speculative * spec) {
+    if (spec == nullptr) {
+        return;
+    }
+
     common_sampler_free(spec->smpl);
 
     llama_batch_free(spec->batch);

convert_hf_to_gguf.py

@@ -658,6 +658,12 @@ class Model:
         if chkhsh == "60824e3c0d9401f89943cbb2fff727f0e2d4c545ba4df2d6e4f09a6db0f5b450":
             # ref: https://huggingface.co/facebook/chameleon-7b
             res = "chameleon"
+        if chkhsh == "1431a23e583c97432bc230bff598d103ddb5a1f89960c8f1d1051aaa944d0b35":
+            # ref: https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0
+            res = "minerva-7b"
+        if chkhsh == "8b5a93ed704057481f240da0be7e7dca721d7f8f4755263b6807227a2cbeae65":
+            # ref: https://huggingface.co/sentence-transformers/stsb-roberta-base
+            res = "roberta-bpe"
 
         if res is None:
             logger.warning("\n")
@@ -1831,29 +1837,40 @@ class MiniCPMModel(Model):
     model_arch = gguf.MODEL_ARCH.MINICPM
 
     def set_gguf_parameters(self):
-        block_count = self.hparams["num_hidden_layers"]
-        self.gguf_writer.add_context_length(self.hparams["max_position_embeddings"])
-        self.gguf_writer.add_embedding_length(self.hparams["hidden_size"])
-        self.gguf_writer.add_block_count(block_count)
-        self.gguf_writer.add_feed_forward_length(self.hparams["intermediate_size"])
-        self.gguf_writer.add_rope_dimension_count(self.hparams["hidden_size"] // self.hparams["num_attention_heads"])
-        self.gguf_writer.add_head_count(self.hparams["num_attention_heads"])
-        self.gguf_writer.add_head_count_kv(self.hparams["num_key_value_heads"])
-        self.gguf_writer.add_layer_norm_rms_eps(self.hparams["rms_norm_eps"])
-        self.gguf_writer.add_file_type(self.ftype)
+        super().set_gguf_parameters()
+        embedding_scale = float(self.hparams["scale_emb"])
+        self.gguf_writer.add_embedding_scale(embedding_scale)
+        logger.info(f"gguf: (minicpm) embedding_scale = {embedding_scale}")
+        residual_scale = self.hparams["scale_depth"] / self.hparams["num_hidden_layers"] ** 0.5
+        self.gguf_writer.add_residual_scale(residual_scale)
+        logger.info(f"gguf: (minicpm) residual_scale = {residual_scale}")
+        logit_scale = self.hparams["hidden_size"] / self.hparams["dim_model_base"]
+        self.gguf_writer.add_logit_scale(logit_scale)
+        logger.info(f"gguf: (minicpm) logit_scale = {logit_scale}")
+        if self.hparams.get("rope_scaling") is not None:
+            if self.hparams["rope_scaling"].get("type") == "longrope":
+                self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.LONGROPE)
+                logger.info(f"gguf: (minicpm) rope_scaling_type = {gguf.RopeScalingType.LONGROPE}")
+
+    def generate_extra_tensors(self) -> Iterable[tuple[str, Tensor]]:
+        rope_dims = self.hparams["hidden_size"] // self.hparams["num_attention_heads"]
+
+        rope_scaling = self.find_hparam(['rope_scaling'], True)
+        if rope_scaling is not None:
+            long_factors = rope_scaling.get('long_factor', None)
+            short_factors = rope_scaling.get('short_factor', None)
+
+            if long_factors is None or short_factors is None:
+                raise KeyError('Missing the required key rope_scaling.long_factor or rope_scaling_short_factor')
+
+            if len(long_factors) != len(short_factors) or len(long_factors) != rope_dims / 2:
+                raise ValueError(f'The length of rope long and short factors must be {rope_dims / 2}')
+
+            yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_LONG), torch.tensor(long_factors, dtype=torch.float32))
+            yield (self.format_tensor_name(gguf.MODEL_TENSOR.ROPE_FACTORS_SHORT), torch.tensor(short_factors, dtype=torch.float32))
 
     def set_vocab(self):
-        self._set_vocab_llama_hf()
-
-    def _reverse_hf_permute(self, weights: Tensor, n_head: int, n_kv_head: int | None = None) -> Tensor:
-        if n_kv_head is not None and n_head != n_kv_head:
-            n_head //= n_kv_head
-
-        return (
-            weights.reshape(n_head, 2, weights.shape[0] // n_head // 2, *weights.shape[1:])
-            .swapaxes(1, 2)
-            .reshape(weights.shape)
-        )
+        self._set_vocab_sentencepiece()
 
     def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
         del bid  # unused
@@ -1863,9 +1880,9 @@ class MiniCPMModel(Model):
 
         # HF models permute some of the tensors, so we need to undo that
         if name.endswith(("q_proj.weight")):
-            data_torch = self._reverse_hf_permute(data_torch, n_head, n_head)
+            data_torch = LlamaModel.permute(data_torch, n_head, n_head)
         if name.endswith(("k_proj.weight")):
-            data_torch = self._reverse_hf_permute(data_torch, n_head, n_kv_head)
+            data_torch = LlamaModel.permute(data_torch, n_head, n_kv_head)
 
         return [(self.map_tensor_name(name), data_torch)]
 
@@ -1975,6 +1992,14 @@ class Qwen2Model(Model):
         except FileNotFoundError:
             self._set_vocab_gpt2()
 
+    def set_gguf_parameters(self):
+        super().set_gguf_parameters()
+        if self.hparams.get("rope_scaling") is not None and "factor" in self.hparams["rope_scaling"]:
+            if self.hparams["rope_scaling"].get("type") == "yarn":
+                self.gguf_writer.add_rope_scaling_type(gguf.RopeScalingType.YARN)
+                self.gguf_writer.add_rope_scaling_factor(self.hparams["rope_scaling"]["factor"])
+                self.gguf_writer.add_rope_scaling_orig_ctx_len(self.hparams["rope_scaling"]["original_max_position_embeddings"])
+
 
 @Model.register("Qwen2MoeForCausalLM")
 class Qwen2MoeModel(Model):
@@ -2519,7 +2544,7 @@ class InternLM2Model(Model):
             return [(self.map_tensor_name(name), data_torch)]
 
 
-@Model.register("BertModel", "CamembertModel")
+@Model.register("BertModel", "CamembertModel", "RobertaModel")
 class BertModel(Model):
     model_arch = gguf.MODEL_ARCH.BERT
 
@@ -2560,7 +2585,8 @@ class BertModel(Model):
 
         # we need this to validate the size of the token_type embeddings
         # though currently we are passing all zeros to the token_type embeddings
-        self.gguf_writer.add_token_type_count(2)  # "Sequence A" or "Sequence B"
+        # "Sequence A" or "Sequence B"
+        self.gguf_writer.add_token_type_count(self.hparams.get("type_vocab_size", 1))
 
         # convert to phantom space vocab
         def phantom(tok):

convert_hf_to_gguf_update.py

@@ -17,7 +17,7 @@
 #
 #   python3 convert_hf_to_gguf_update.py <huggingface_token>
 #
-# - Copy-paste the generated get_vocab_base_pre() function into convert_hf_to_gguf.py
+# - The convert_hf_to_gguf.py script will have had its get_vocab_base_pre() function updated
 # - Update llama.cpp with the new pre-tokenizer if necessary
 #
 # TODO: generate tokenizer tests for llama.cpp
@@ -102,6 +102,8 @@ models = [
     {"name": "exaone",         "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/LGAI-EXAONE/EXAONE-3.0-7.8B-Instruct", },
     {"name": "phi-2",          "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/microsoft/phi-2", },
     {"name": "chameleon",      "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/facebook/chameleon-7b", },
+    {"name": "minerva-7b",     "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sapienzanlp/Minerva-7B-base-v1.0", },
+    {"name": "roberta-bpe",    "tokt": TOKENIZER_TYPE.BPE, "repo": "https://huggingface.co/sentence-transformers/stsb-roberta-base"},
 ]
 
 

docs/backend/BLIS.md

@@ -27,13 +27,6 @@ We recommend using openmp since it's easier to modify the cores being used.
 
 ### llama.cpp compilation
 
-Makefile:
-
-```bash
-make GGML_BLIS=1 -j
-# make GGML_BLIS=1 llama-benchmark-matmult
-```
-
 CMake:
 
 ```bash

docs/build.md

@@ -7,124 +7,68 @@ git clone https://github.com/ggerganov/llama.cpp
 cd llama.cpp
 ```
 
-In order to build llama.cpp you have four different options.
+The following sections describe how to build with different backends and options.
 
-- Using `make`:
-  - On Linux or MacOS:
+## CPU Build
 
-      ```bash
-      make
-      ```
+Build llama.cpp using `CMake`:
 
-  - On Windows (x86/x64 only, arm64 requires cmake):
+```bash
+cmake -B build
+cmake --build build --config Release
+```
 
-    1. Download the latest fortran version of [w64devkit](https://github.com/skeeto/w64devkit/releases).
-    2. Extract `w64devkit` on your pc.
-    3. Run `w64devkit.exe`.
-    4. Use the `cd` command to reach the `llama.cpp` folder.
-    5. From here you can run:
-        ```bash
-        make
-        ```
+**Notes**:
 
-  - Notes:
-    - For `Q4_0_4_4` quantization type build, add the `GGML_NO_LLAMAFILE=1` flag. For example, use `make GGML_NO_LLAMAFILE=1`.
-    - For faster compilation, add the `-j` argument to run multiple jobs in parallel. For example, `make -j 8` will run 8 jobs in parallel.
-    - For faster repeated compilation, install [ccache](https://ccache.dev/).
-    - For debug builds, run `make LLAMA_DEBUG=1`
+- For faster compilation, add the `-j` argument to run multiple jobs in parallel, or use a generator that does this automatically such as Ninja. For example, `cmake --build build --config Release -j 8` will run 8 jobs in parallel.
+- For faster repeated compilation, install [ccache](https://ccache.dev/)
+- For debug builds, there are two cases:
 
-- Using `CMake`:
+    1. Single-config generators (e.g. default = `Unix Makefiles`; note that they just ignore the `--config` flag):
 
-  ```bash
-  cmake -B build
+       ```bash
+       cmake -B build -DCMAKE_BUILD_TYPE=Debug
+       cmake --build build
+       ```
+
+    2. Multi-config generators (`-G` param set to Visual Studio, XCode...):
+
+       ```bash
+       cmake -B build -G "Xcode"
+       cmake --build build --config Debug
+       ```
+
+    For more details and a list of supported generators, see the [CMake documentation](https://cmake.org/cmake/help/latest/manual/cmake-generators.7.html).
+- For static builds, add `-DBUILD_SHARED_LIBS=OFF`:
+  ```
+  cmake -B build -DBUILD_SHARED_LIBS=OFF
   cmake --build build --config Release
   ```
 
-  **Notes**:
-
-    - For `Q4_0_4_4` quantization type build, add the `-DGGML_LLAMAFILE=OFF` cmake option. For example, use `cmake -B build -DGGML_LLAMAFILE=OFF`.
-    - For faster compilation, add the `-j` argument to run multiple jobs in parallel. For example, `cmake --build build --config Release -j 8` will run 8 jobs in parallel.
-    - For faster repeated compilation, install [ccache](https://ccache.dev/).
-    - For debug builds, there are two cases:
-
-      1. Single-config generators (e.g. default = `Unix Makefiles`; note that they just ignore the `--config` flag):
-
-      ```bash
-      cmake -B build -DCMAKE_BUILD_TYPE=Debug
-      cmake --build build
-      ```
-
-      2. Multi-config generators (`-G` param set to Visual Studio, XCode...):
-
-      ```bash
-      cmake -B build -G "Xcode"
-      cmake --build build --config Debug
-      ```
-    - Building for Windows (x86, x64 and arm64) with MSVC or clang as compilers:
-      - Install Visual Studio 2022, e.g. via the [Community Edition](https://visualstudio.microsoft.com/de/vs/community/). In the installer, select at least the following options (this also automatically installs the required additional tools like CMake,...):
-        - Tab Workload: Desktop-development with C++
-        - Tab Components (select quickly via search): C++-_CMake_ Tools for Windows, _Git_ for Windows, C++-_Clang_ Compiler for Windows, MS-Build Support for LLVM-Toolset (clang)
-      - Please remember to always use a Developer Command Prompt / PowerShell for VS2022 for git, build, test
-      - For Windows on ARM (arm64, WoA) build with:
-        ```bash
-        cmake --preset arm64-windows-llvm-release -D GGML_OPENMP=OFF
-        cmake --build build-arm64-windows-llvm-release
-        ```
-        Note: Building for arm64 could also be done just with MSVC (with the build-arm64-windows-MSVC preset, or the standard CMake build instructions). But MSVC does not support inline ARM assembly-code, used e.g. for the accelerated Q4_0_4_8 CPU kernels.
-
--   Using `gmake` (FreeBSD):
-
-    1. Install and activate [DRM in FreeBSD](https://wiki.freebsd.org/Graphics)
-    2. Add your user to **video** group
-    3. Install compilation dependencies.
-
-        ```bash
-        sudo pkg install gmake automake autoconf pkgconf llvm15 openblas
-
-        gmake CC=/usr/local/bin/clang15 CXX=/usr/local/bin/clang++15 -j4
-        ```
-
-## Metal Build
-
-On MacOS, Metal is enabled by default. Using Metal makes the computation run on the GPU.
-To disable the Metal build at compile time use the `GGML_NO_METAL=1` flag or the `GGML_METAL=OFF` cmake option.
-
-When built with Metal support, you can explicitly disable GPU inference with the `--n-gpu-layers|-ngl 0` command-line
-argument.
+- Building for Windows (x86, x64 and arm64) with MSVC or clang as compilers:
+    - Install Visual Studio 2022, e.g. via the [Community Edition](https://visualstudio.microsoft.com/de/vs/community/). In the installer, select at least the following options (this also automatically installs the required additional tools like CMake,...):
+    - Tab Workload: Desktop-development with C++
+    - Tab Components (select quickly via search): C++-_CMake_ Tools for Windows, _Git_ for Windows, C++-_Clang_ Compiler for Windows, MS-Build Support for LLVM-Toolset (clang)
+    - Please remember to always use a Developer Command Prompt / PowerShell for VS2022 for git, build, test
+    - For Windows on ARM (arm64, WoA) build with:
+    ```bash
+    cmake --preset arm64-windows-llvm-release -D GGML_OPENMP=OFF
+    cmake --build build-arm64-windows-llvm-release
+    ```
+    Building for arm64 can also be done with the MSVC compiler with the build-arm64-windows-MSVC preset, or the standard CMake build instructions. However, note that the MSVC compiler does not support inline ARM assembly code, used e.g. for the accelerated Q4_0_N_M CPU kernels.
 
 ## BLAS Build
 
-Building the program with BLAS support may lead to some performance improvements in prompt processing using batch sizes higher than 32 (the default is 512). Support with CPU-only BLAS implementations doesn't affect the normal generation performance. We may see generation performance improvements with GPU-involved BLAS implementations, e.g. cuBLAS, hipBLAS. There are currently several different BLAS implementations available for build and use:
+Building the program with BLAS support may lead to some performance improvements in prompt processing using batch sizes higher than 32 (the default is 512). Using BLAS doesn't affect the generation performance. There are currently several different BLAS implementations available for build and use:
 
-### Accelerate Framework:
+### Accelerate Framework
 
 This is only available on Mac PCs and it's enabled by default. You can just build using the normal instructions.
 
-### OpenBLAS:
+### OpenBLAS
 
 This provides BLAS acceleration using only the CPU. Make sure to have OpenBLAS installed on your machine.
 
-- Using `make`:
-  - On Linux:
-    ```bash
-    make GGML_OPENBLAS=1
-    ```
-
-  - On Windows:
-
-    1. Download the latest fortran version of [w64devkit](https://github.com/skeeto/w64devkit/releases).
-    2. Download the latest version of [OpenBLAS for Windows](https://github.com/xianyi/OpenBLAS/releases).
-    3. Extract `w64devkit` on your pc.
-    4. From the OpenBLAS zip that you just downloaded copy `libopenblas.a`, located inside the `lib` folder, inside `w64devkit\x86_64-w64-mingw32\lib`.
-    5. From the same OpenBLAS zip copy the content of the `include` folder inside `w64devkit\x86_64-w64-mingw32\include`.
-    6. Run `w64devkit.exe`.
-    7. Use the `cd` command to reach the `llama.cpp` folder.
-    8. From here you can run:
-
-        ```bash
-        make GGML_OPENBLAS=1
-        ```
-
 - Using `CMake` on Linux:
 
     ```bash
@@ -136,14 +80,6 @@ This provides BLAS acceleration using only the CPU. Make sure to have OpenBLAS i
 
 Check [BLIS.md](./backend/BLIS.md) for more information.
 
-### SYCL
-
-SYCL is a higher-level programming model to improve programming productivity on various hardware accelerators.
-
-llama.cpp based on SYCL is used to **support Intel GPU** (Data Center Max series, Flex series, Arc series, Built-in GPU and iGPU).
-
-For detailed info, please refer to [llama.cpp for SYCL](./backend/SYCL.md).
-
 ### Intel oneMKL
 
 Building through oneAPI compilers will make avx_vnni instruction set available for intel processors that do not support avx512 and avx512_vnni. Please note that this build config **does not support Intel GPU**. For Intel GPU support, please refer to [llama.cpp for SYCL](./backend/SYCL.md).
@@ -161,16 +97,29 @@ Building through oneAPI compilers will make avx_vnni instruction set available f
 
 Check [Optimizing and Running LLaMA2 on Intel® CPU](https://www.intel.com/content/www/us/en/content-details/791610/optimizing-and-running-llama2-on-intel-cpu.html) for more information.
 
-### CUDA
+### Other BLAS libraries
 
-This provides GPU acceleration using the CUDA cores of your Nvidia GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager (e.g. `apt install nvidia-cuda-toolkit`) or from here: [CUDA Toolkit](https://developer.nvidia.com/cuda-downloads).
+Any other BLAS library can be used by setting the `GGML_BLAS_VENDOR` option. See the [CMake documentation](https://cmake.org/cmake/help/latest/module/FindBLAS.html#blas-lapack-vendors) for a list of supported vendors.
 
-For Jetson user, if you have Jetson Orin, you can try this: [Offical Support](https://www.jetson-ai-lab.com/tutorial_text-generation.html). If you are using an old model(nano/TX2), need some additional operations before compiling.
+## Metal Build
+
+On MacOS, Metal is enabled by default. Using Metal makes the computation run on the GPU.
+To disable the Metal build at compile time use the `-DGGML_METAL=OFF` cmake option.
+
+When built with Metal support, you can explicitly disable GPU inference with the `--n-gpu-layers 0` command-line argument.
+
+## SYCL
+
+SYCL is a higher-level programming model to improve programming productivity on various hardware accelerators.
+
+llama.cpp based on SYCL is used to **support Intel GPU** (Data Center Max series, Flex series, Arc series, Built-in GPU and iGPU).
+
+For detailed info, please refer to [llama.cpp for SYCL](./backend/SYCL.md).
+
+## CUDA
+
+This provides GPU acceleration using an NVIDIA GPU. Make sure to have the CUDA toolkit installed. You can download it from your Linux distro's package manager (e.g. `apt install nvidia-cuda-toolkit`) or from the [NVIDIA developer site](https://developer.nvidia.com/cuda-downloads).
 
-- Using `make`:
-  ```bash
-  make GGML_CUDA=1
-  ```
 - Using `CMake`:
 
   ```bash
@@ -192,14 +141,10 @@ The following compilation options are also available to tweak performance:
 | GGML_CUDA_PEER_MAX_BATCH_SIZE | Positive integer       | 128     | Maximum batch size for which to enable peer access between multiple GPUs. Peer access requires either Linux or NVLink. When using NVLink enabling peer access for larger batch sizes is potentially beneficial.                                                                         |
 | GGML_CUDA_FA_ALL_QUANTS       | Boolean                | false   | Compile support for all KV cache quantization type (combinations) for the FlashAttention CUDA kernels. More fine-grained control over KV cache size but compilation takes much longer.                                                                                                  |
 
-### MUSA
+## 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).
 
-- Using `make`:
-  ```bash
-  make GGML_MUSA=1
-  ```
 - Using `CMake`:
 
   ```bash
@@ -213,16 +158,12 @@ The environment variable `GGML_CUDA_ENABLE_UNIFIED_MEMORY=1` can be used to enab
 
 Most of the compilation options available for CUDA should also be available for MUSA, though they haven't been thoroughly tested yet.
 
-### hipBLAS
+## HIP
 
-This provides BLAS acceleration on HIP-supported AMD GPUs.
+This provides GPU acceleration on HIP-supported AMD GPUs.
 Make sure to have ROCm installed.
 You can download it from your Linux distro's package manager or from here: [ROCm Quick Start (Linux)](https://rocm.docs.amd.com/projects/install-on-linux/en/latest/tutorial/quick-start.html#rocm-install-quick).
 
-- Using `make`:
-  ```bash
-  make GGML_HIP=1
-  ```
 - Using `CMake` for Linux (assuming a gfx1030-compatible AMD GPU):
   ```bash
   HIPCXX="$(hipconfig -l)/clang" HIP_PATH="$(hipconfig -R)" \
@@ -247,11 +188,6 @@ You can download it from your Linux distro's package manager or from here: [ROCm
       && cmake --build build -- -j 16
   ```
 
-- Using `make` (example for target gfx1030, build with 16 CPU threads):
-  ```bash
-  make -j16 GGML_HIP=1 GGML_HIP_UMA=1 AMDGPU_TARGETS=gfx1030
-  ```
-
 - Using `CMake` for Windows (using x64 Native Tools Command Prompt for VS, and assuming a gfx1100-compatible AMD GPU):
   ```bash
   set PATH=%HIP_PATH%\bin;%PATH%
@@ -265,11 +201,11 @@ You can download it from your Linux distro's package manager or from here: [ROCm
 The environment variable [`HIP_VISIBLE_DEVICES`](https://rocm.docs.amd.com/en/latest/understand/gpu_isolation.html#hip-visible-devices) can be used to specify which GPU(s) will be used.
 If your GPU is not officially supported you can use the environment variable [`HSA_OVERRIDE_GFX_VERSION`] set to a similar GPU, for example 10.3.0 on RDNA2 (e.g. gfx1030, gfx1031, or gfx1035) or 11.0.0 on RDNA3.
 
-### Vulkan
+## Vulkan
 
 **Windows**
 
-#### w64devkit
+### w64devkit
 
 Download and extract [`w64devkit`](https://github.com/skeeto/w64devkit/releases).
 
@@ -289,9 +225,14 @@ Libs: -lvulkan-1
 EOF
 
 ```
-Switch into the `llama.cpp` directory and run `make GGML_VULKAN=1`.
 
-#### Git Bash MINGW64
+Switch into the `llama.cpp` directory and build using CMake.
+```sh
+cmake -B build -DGGML_VULKAN=ON
+cmake --build build --config Release
+```
+
+### Git Bash MINGW64
 
 Download and install [`Git-SCM`](https://git-scm.com/downloads/win) with the default settings
 
@@ -310,20 +251,21 @@ cmake --build build --config Release
 
 Now you can load the model in conversation mode using `Vulkan`
 
-```
-build/bin/release/llama-cli -m "[PATH TO MODEL]" -ngl 100 -c 16384 -t 10 -n -2 -cnv
+```sh
+build/bin/Release/llama-cli -m "[PATH TO MODEL]" -ngl 100 -c 16384 -t 10 -n -2 -cnv
 ```
 
-#### MSYS2
+### MSYS2
 Install [MSYS2](https://www.msys2.org/) and then run the following commands in a UCRT terminal to install dependencies.
-  ```sh
-  pacman -S git \
-      mingw-w64-ucrt-x86_64-gcc \
-      mingw-w64-ucrt-x86_64-cmake \
-      mingw-w64-ucrt-x86_64-vulkan-devel \
-      mingw-w64-ucrt-x86_64-shaderc
-  ```
-Switch into `llama.cpp` directory and build using CMake.
+```sh
+pacman -S git \
+    mingw-w64-ucrt-x86_64-gcc \
+    mingw-w64-ucrt-x86_64-cmake \
+    mingw-w64-ucrt-x86_64-vulkan-devel \
+    mingw-w64-ucrt-x86_64-shaderc
+```
+
+Switch into the `llama.cpp` directory and build using CMake.
 ```sh
 cmake -B build -DGGML_VULKAN=ON
 cmake --build build --config Release
@@ -372,7 +314,7 @@ cmake --build build --config Release
 # ggml_vulkan: Using Intel(R) Graphics (ADL GT2) | uma: 1 | fp16: 1 | warp size: 32
 ```
 
-### CANN
+## CANN
 This provides NPU acceleration using the AI cores of your Ascend NPU. And [CANN](https://www.hiascend.com/en/software/cann) is a hierarchical APIs to help you to quickly build AI applications and service based on Ascend NPU.
 
 For more information about Ascend NPU in [Ascend Community](https://www.hiascend.com/en/).
@@ -387,22 +329,26 @@ cmake --build build --config release
 
 You can test with:
 
-`./build/bin/llama-cli -m PATH_TO_MODEL -p "Building a website can be done in 10 steps:" -ngl 32`
-
-If the fllowing info is output on screen, you are using `llama.cpp by CANN backend`:
 ```bash
-llm_load_tensors:       CANN buffer size = 13313.00 MiB
+./build/bin/llama-cli -m PATH_TO_MODEL -p "Building a website can be done in 10 steps:" -ngl 32
+```
+
+If the following info is output on screen, you are using `llama.cpp` with the CANN backend:
+```bash
+llm_load_tensors:       CANN model buffer size = 13313.00 MiB
 llama_new_context_with_model:       CANN compute buffer size =  1260.81 MiB
 ```
 
 For detailed info, such as model/device supports, CANN install, please refer to [llama.cpp for CANN](./backend/CANN.md).
 
-### Android
+## Android
 
 To read documentation for how to build on Android, [click here](./android.md)
 
-### Arm CPU optimized mulmat kernels
+## Notes about GPU-accelerated backends
 
-Llama.cpp includes a set of optimized mulmat kernels for the Arm architecture, leveraging Arm® Neon™, int8mm and SVE instructions. These kernels are enabled at build time through the appropriate compiler cpu-type flags, such as `-DCMAKE_C_FLAGS=-march=armv8.2a+i8mm+sve`. Note that these optimized kernels require the model to be quantized into one of the formats: `Q4_0_4_4` (Arm Neon), `Q4_0_4_8` (int8mm) or `Q4_0_8_8` (SVE). The SVE mulmat kernel specifically requires a vector width of 256 bits. When running on devices with a different vector width, it is recommended to use the `Q4_0_4_8` (int8mm) or `Q4_0_4_4` (Arm Neon) formats for better performance. Refer to [examples/quantize/README.md](../examples/quantize/README.md) for more information on the quantization formats.
+The GPU may still be used to accelerate some parts of the computation even when using the `-ngl 0` option. You can fully disable GPU acceleration by using `--device none`.
 
-To support `Q4_0_4_4`, you must build with `GGML_NO_LLAMAFILE=1` (`make`) or `-DGGML_LLAMAFILE=OFF` (`cmake`).
+In most cases, it is possible to build and use multiple backends at the same time. For example, you can build llama.cpp with both CUDA and Vulkan support by using the `-DGGML_CUDA=ON -DGGML_VULKAN=ON` options with CMake. At runtime, you can specify which backend devices to use with the `--device` option. To see a list of available devices, use the `--list-devices` option.
+
+Backends can be built as dynamic libraries that can be loaded dynamically at runtime. This allows you to use the same llama.cpp binary on different machines with different GPUs. To enable this feature, use the `GGML_BACKEND_DL` option when building.

examples/base-translate.sh

@@ -1,61 +0,0 @@
-#!/bin/bash
-#
-# Few-shot translation example.
-# Requires a base model (i.e. no fine-tuned or instruct models).
-#
-# Usage:
-#
-#   cd llama.cpp
-#   make -j
-#
-#   ./examples/base-translate.sh <model-base> "<text>" [extra-main-args]
-#
-
-if [ $# -lt 2 ]; then
-  echo "Usage: ./base-translate.sh <model-base> \"<text>\" [extra-main-args]"
-  exit 1
-fi
-
-eargs=""
-if [ $# -gt 2 ]; then
-  eargs="${@:3}"
-fi
-
-ftmp="__llama.cpp_example_tmp__.txt"
-trap "rm -f $ftmp" EXIT
-
-echo "Translate from English to French:
-
-===
-
-sea otter, peppermint, plush girafe:
-
-sea otter => loutre de mer
-peppermint => menthe poivrée
-plush girafe => girafe peluche
-
-===
-
-violin
-
-violin => violon
-
-===
-
-phone, computer, mouse, keyboard:
-
-phone => téléphone
-computer => ordinateur
-mouse => souris
-keyboard => clavier
-
-===
-" > $ftmp
-
-echo "$2
-" >> $ftmp
-
-model=$1
-
-# generate the most likely continuation until the string "===" is found
-./llama-cli -m $model -f $ftmp -n 64 --temp 0 --repeat-penalty 1.0 --no-penalize-nl -r "===" $eargs

examples/batched-bench/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-batched-bench)
 add_executable(${TARGET} batched-bench.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/batched/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-batched)
 add_executable(${TARGET} batched.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/convert-llama2c-to-ggml/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-convert-llama2c-to-ggml)
 add_executable(${TARGET} convert-llama2c-to-ggml.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/convert-llama2c-to-ggml/README.md

@@ -2,11 +2,8 @@
 
 This example reads weights from project [llama2.c](https://github.com/karpathy/llama2.c) and saves them in ggml compatible format. The vocab that is available in `models/ggml-vocab.bin` is used by default.
 
-To convert the model first download the models from the [llama2.c](https://github.com/karpathy/llama2.c) repository:
+To convert the model first download the models from the [llama2.c](https://github.com/karpathy/llama2.c) repository.
 
-`$ make -j`
-
-After successful compilation, following usage options are available:
 ```
 usage: ./llama-convert-llama2c-to-ggml [options]
 

examples/cvector-generator/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-cvector-generator)
 add_executable(${TARGET} cvector-generator.cpp pca.hpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/deprecation-warning/deprecation-warning.cpp

@@ -12,7 +12,7 @@ int main(int argc, char** argv) {
     }
 
     // Get only the program name from the full path
-    auto pos = filename.find_last_of('/');
+    auto pos = filename.find_last_of("/\\");
     if (pos != std::string::npos) {
         filename = filename.substr(pos+1);
     }

examples/embedding/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-embedding)
 add_executable(${TARGET} embedding.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/eval-callback/CMakeLists.txt

@@ -2,7 +2,7 @@ set(TARGET llama-eval-callback)
 add_executable(${TARGET} eval-callback.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TEST_TARGET test-eval-callback)
 add_test(NAME ${TEST_TARGET}

examples/export-lora/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-export-lora)
 add_executable(${TARGET} export-lora.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/gbnf-validator/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-gbnf-validator)
 add_executable(${TARGET} gbnf-validator.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/gen-docs/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-gen-docs)
 add_executable(${TARGET} gen-docs.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/gguf-hash/CMakeLists.txt

@@ -19,4 +19,4 @@ add_library(sha256 OBJECT deps/sha256/sha256.c deps/sha256/sha256.h)
 target_link_libraries(${TARGET} PRIVATE sha256)
 
 target_link_libraries(${TARGET} PRIVATE ggml ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/gguf-split/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-gguf-split)
 add_executable(${TARGET} gguf-split.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/gguf/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-gguf)
 add_executable(${TARGET} gguf.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE ggml ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/gritlm/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-gritlm)
 add_executable(${TARGET} gritlm.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/imatrix/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-imatrix)
 add_executable(${TARGET} imatrix.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/imatrix/README.md

@@ -25,8 +25,6 @@ For faster computation, make sure to use GPU offloading via the `-ngl` argument
 ## Example
 
 ```bash
-GGML_CUDA=1 make -j
-
 # generate importance matrix (imatrix.dat)
 ./llama-imatrix -m ggml-model-f16.gguf -f train-data.txt -ngl 99
 

examples/imatrix/imatrix.cpp

@@ -637,10 +637,19 @@ int main(int argc, char ** argv) {
         LOG_INF("%s\n", common_params_get_system_info(params).c_str());
     }
 
-    if (!compute_imatrix(ctx, params)) {
-        return 1;
+    if (params.prompt.empty()) {
+        if (params.in_files.empty()) {
+            LOG_ERR("Error: No prompt provided and no precomputed matrices (--in-file) to combine.\n");
+            return 1;
+        }
+        LOG_INF("No prompt provided; combining precomputed matrices only.\n");
+    } else {
+        if (!compute_imatrix(ctx, params)) {
+            return 1;
+        }
     }
 
+
     g_collector.save_imatrix();
 
     LOG("\n");

examples/infill/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-infill)
 add_executable(${TARGET} infill.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/infill/README.md

@@ -14,7 +14,7 @@ In this section, we cover the most commonly used options for running the `infill
 -   `-m FNAME, --model FNAME`: Specify the path to the LLaMA model file (e.g., `models/7B/ggml-model.bin`).
 -   `-i, --interactive`: Run the program in interactive mode, allowing you to provide input directly and receive real-time responses.
 -   `-n N, --n-predict N`: Set the number of tokens to predict when generating text. Adjusting this value can influence the length of the generated text.
--   `-c N, --ctx-size N`: Set the size of the prompt context. The default is 512, but LLaMA models were built with a context of 2048, which will provide better results for longer input/inference.
+-   `-c N, --ctx-size N`: Set the size of the prompt context. The default is 4096, but if a LLaMA model was built with a longer context, increasing this value will provide better results for longer input/inference.
 -   `--spm-infill`: Use Suffix/Prefix/Middle pattern for infill (instead of Prefix/Suffix/Middle) as some models prefer this.
 
 ## Input Prompts

examples/llama-bench/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-bench)
 add_executable(${TARGET} llama-bench.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/llava/CMakeLists.txt

@@ -11,7 +11,7 @@ target_include_directories(llava PUBLIC .)
 target_include_directories(llava PUBLIC ../..)
 target_include_directories(llava PUBLIC ../../common)
 
-target_compile_features(llava PRIVATE cxx_std_11)
+target_compile_features(llava PRIVATE cxx_std_17)
 
 add_library(llava_static STATIC $<TARGET_OBJECTS:llava>)
 if (BUILD_SHARED_LIBS)
@@ -35,11 +35,11 @@ add_executable(${TARGET} llava-cli.cpp)
 set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama-llava-cli)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llava ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-minicpmv-cli)
 add_executable(${TARGET} minicpmv-cli.cpp)
 set_target_properties(${TARGET} PROPERTIES OUTPUT_NAME llama-minicpmv-cli)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llava ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/llava/clip.cpp

@@ -12,6 +12,10 @@
 #include "ggml-cuda.h"
 #endif
 
+#ifdef GGML_USE_SYCL
+#include "ggml-sycl.h"
+#endif
+
 #ifdef GGML_USE_METAL
 #include "ggml-metal.h"
 #endif
@@ -40,10 +44,17 @@
 #include <cinttypes>
 #include <limits>
 
-#define LOG_INF(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
-#define LOG_WRN(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
-#define LOG_ERR(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
-#define LOG_DBG(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
+#if defined(LLAVA_LOG_OFF)
+#   define LOG_INF(...)
+#   define LOG_WRN(...)
+#   define LOG_ERR(...)
+#   define LOG_DBG(...)
+#else // defined(LLAVA_LOG_OFF)
+#   define LOG_INF(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
+#   define LOG_WRN(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
+#   define LOG_ERR(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
+#   define LOG_DBG(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
+#endif // defined(LLAVA_LOG_OFF)
 
 //#define CLIP_DEBUG_FUNCTIONS
 
@@ -1162,6 +1173,11 @@ struct clip_ctx * clip_model_load(const char * fname, const int verbosity = 1) {
     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__);

examples/llava/llava.cpp

@@ -11,13 +11,17 @@
 #include <limits>
 #include <vector>
 
-#define die(msg)          do { fputs("error: " msg "\n", stderr);                exit(1); } while (0)
-#define die_fmt(fmt, ...) do { fprintf(stderr, "error: " fmt "\n", __VA_ARGS__); exit(1); } while (0)
-
-#define LOG_INF(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
-#define LOG_WRN(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
-#define LOG_ERR(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
-#define LOG_DBG(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
+#if defined(LLAVA_LOG_OFF)
+#   define LOG_INF(...)
+#   define LOG_WRN(...)
+#   define LOG_ERR(...)
+#   define LOG_DBG(...)
+#else // defined(LLAVA_LOG_OFF)
+#   define LOG_INF(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
+#   define LOG_WRN(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
+#   define LOG_ERR(...) do { fprintf(stderr, __VA_ARGS__); } while (0)
+#   define LOG_DBG(...) do { fprintf(stdout, __VA_ARGS__); } while (0)
+#endif // defined(LLAVA_LOG_OFF)
 
 // RGB uint8 image
 struct clip_image_u8 {
@@ -498,10 +502,16 @@ static bool load_file_to_bytes(const char* path, unsigned char** bytesOut, long
     errno = 0;
     size_t ret = fread(buffer, 1, fileSize, file); // Read the file into the buffer
     if (ferror(file)) {
-        die_fmt("read error: %s", strerror(errno));
+        LOG_ERR("read error: %s", strerror(errno));
+        free(buffer);
+        fclose(file);
+        return false;
     }
     if (ret != (size_t) fileSize) {
-        die("unexpectedly reached end of file");
+        LOG_ERR("unexpectedly reached end of file");
+        free(buffer);
+        fclose(file);
+        return false;
     }
     fclose(file); // Close the file
 

examples/lookahead/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-lookahead)
 add_executable(${TARGET} lookahead.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/lookup/CMakeLists.txt

@@ -2,22 +2,22 @@ set(TARGET llama-lookup)
 add_executable(${TARGET} lookup.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-lookup-create)
 add_executable(${TARGET} lookup-create.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-lookup-merge)
 add_executable(${TARGET} lookup-merge.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-lookup-stats)
 add_executable(${TARGET} lookup-stats.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/main-cmake-pkg/CMakeLists.txt

@@ -29,4 +29,4 @@ add_executable(${TARGET} ${CMAKE_CURRENT_LIST_DIR}/../main/main.cpp)
 target_include_directories(${TARGET} PRIVATE ${_common_path})
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/main/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-cli)
 add_executable(${TARGET} main.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/main/README.md

@@ -66,7 +66,7 @@ In this section, we cover the most commonly used options for running the `llama-
 -   `-mu MODEL_URL --model-url MODEL_URL`: Specify a remote http url to download the file (e.g [https://huggingface.co/ggml-org/gemma-1.1-7b-it-Q4_K_M-GGUF/resolve/main/gemma-1.1-7b-it.Q4_K_M.gguf?download=true](https://huggingface.co/ggml-org/gemma-1.1-7b-it-Q4_K_M-GGUF/resolve/main/gemma-1.1-7b-it.Q4_K_M.gguf?download=true)).
 -   `-i, --interactive`: Run the program in interactive mode, allowing you to provide input directly and receive real-time responses.
 -   `-n N, --n-predict N`: Set the number of tokens to predict when generating text. Adjusting this value can influence the length of the generated text.
--   `-c N, --ctx-size N`: Set the size of the prompt context. The default is 512, but LLaMA models were built with a context of 2048, which will provide better results for longer input/inference.
+-   `-c N, --ctx-size N`: Set the size of the prompt context. The default is 4096, but if a LLaMA model was built with a longer context, increasing this value will provide better results for longer input/inference.
 -   `-mli, --multiline-input`: Allows you to write or paste multiple lines without ending each in '\'
 -   `-t N, --threads N`: Set the number of threads to use during generation. For optimal performance, it is recommended to set this value to the number of physical CPU cores your system has.
 -   `-ngl N, --n-gpu-layers N`: When compiled with GPU support, this option allows offloading some layers to the GPU for computation. Generally results in increased performance.
@@ -131,7 +131,7 @@ During text generation, LLaMA models have a limited context size, which means th
 
 ### Context Size
 
-- `-c N, --ctx-size N`: Set the size of the prompt context (default: 0, 0 = loaded from model). The LLaMA models were built with a context of 2048-8192, which will yield the best results on longer input/inference.
+- `-c N, --ctx-size N`: Set the size of the prompt context (default: 4096, 0 = loaded from model). If a LLaMA model was built with a longer context, increasing this value will yield the best results on longer input/inference.
 
 ### Extended Context Size
 
@@ -348,6 +348,7 @@ These options provide extra functionality and customization when running the LLa
 
 -   `-h, --help`: Display a help message showing all available options and their default values. This is particularly useful for checking the latest options and default values, as they can change frequently, and the information in this document may become outdated.
 -   `--verbose-prompt`: Print the prompt before generating text.
+-   `--no-display-prompt`: Don't print prompt at generation.
 -   `-mg i, --main-gpu i`: When using multiple GPUs this option controls which GPU is used for small tensors for which the overhead of splitting the computation across all GPUs is not worthwhile. The GPU in question will use slightly more VRAM to store a scratch buffer for temporary results. By default GPU 0 is used.
 -   `-ts SPLIT, --tensor-split SPLIT`: When using multiple GPUs this option controls how large tensors should be split across all GPUs. `SPLIT` is a comma-separated list of non-negative values that assigns the proportion of data that each GPU should get in order. For example, "3,2" will assign 60% of the data to GPU 0 and 40% to GPU 1. By default the data is split in proportion to VRAM but this may not be optimal for performance.
 -   `-hfr URL --hf-repo URL`: The url to the Hugging Face model repository. Used in conjunction with `--hf-file` or `-hff`. The model is downloaded and stored in the file provided by `-m` or `--model`. If `-m` is not provided, the model is auto-stored in the path specified by the `LLAMA_CACHE` environment variable  or in an OS-specific local cache.

examples/parallel/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-parallel)
 add_executable(${TARGET} parallel.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/passkey/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-passkey)
 add_executable(${TARGET} passkey.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/perplexity/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-perplexity)
 add_executable(${TARGET} perplexity.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/quantize-stats/CMakeLists.txt

@@ -3,4 +3,4 @@ add_executable(${TARGET} quantize-stats.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE llama build_info ${CMAKE_THREAD_LIBS_INIT})
 target_include_directories(${TARGET} PRIVATE ../../common)
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/quantize/CMakeLists.txt

@@ -3,4 +3,4 @@ add_executable(${TARGET} quantize.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
 target_include_directories(${TARGET} PRIVATE ../../common)
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/quantize/README.md

@@ -54,8 +54,6 @@ As the models are currently fully loaded into memory, you will need adequate dis
 
 Several quantization methods are supported. They differ in the resulting model disk size and inference speed.
 
-The quantization formats `Q4_0_4_4`, `Q4_0_4_8` and `Q4_0_8_8` are block interleaved variants of the `Q4_0` format, providing a data layout that is better suited for specific implementations of optimized mulmat kernels. Since these formats differ only in data layout, they have the same quantized size as the `Q4_0` format.
-
 *(outdated)*
 
 | Model | Measure      |    F16 |   Q4_0 |   Q4_1 |   Q5_0 |   Q5_1 |   Q8_0 |

examples/quantize/quantize.cpp

@@ -48,9 +48,6 @@ static const std::vector<struct quant_option> QUANT_OPTIONS = {
     { "Q5_K_M",   LLAMA_FTYPE_MOSTLY_Q5_K_M,   " 5.33G, +0.0569 ppl @ Llama-3-8B",  },
     { "Q6_K",     LLAMA_FTYPE_MOSTLY_Q6_K,     " 6.14G, +0.0217 ppl @ Llama-3-8B",  },
     { "Q8_0",     LLAMA_FTYPE_MOSTLY_Q8_0,     " 7.96G, +0.0026 ppl @ Llama-3-8B",  },
-    { "Q4_0_4_4", LLAMA_FTYPE_MOSTLY_Q4_0_4_4, " 4.34G, +0.4685 ppl @ Llama-3-8B",  },
-    { "Q4_0_4_8", LLAMA_FTYPE_MOSTLY_Q4_0_4_8, " 4.34G, +0.4685 ppl @ Llama-3-8B",  },
-    { "Q4_0_8_8", LLAMA_FTYPE_MOSTLY_Q4_0_8_8, " 4.34G, +0.4685 ppl @ Llama-3-8B",  },
     { "F16",      LLAMA_FTYPE_MOSTLY_F16,      "14.00G, +0.0020 ppl @ Mistral-7B",  },
     { "BF16",     LLAMA_FTYPE_MOSTLY_BF16,     "14.00G, -0.0050 ppl @ Mistral-7B",  },
     { "F32",      LLAMA_FTYPE_ALL_F32,         "26.00G              @ 7B",          },

examples/retrieval/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-retrieval)
 add_executable(${TARGET} retrieval.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/run/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-run)
 add_executable(${TARGET} run.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/save-load-state/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-save-load-state)
 add_executable(${TARGET} save-load-state.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/server/CMakeLists.txt

@@ -16,12 +16,7 @@ set(TARGET_SRCS
 )
 set(PUBLIC_ASSETS
     index.html
-    completion.js
     loading.html
-    deps_daisyui.min.css
-    deps_markdown-it.js
-    deps_tailwindcss.js
-    deps_vue.esm-browser.js
 )
 
 foreach(asset ${PUBLIC_ASSETS})
@@ -33,6 +28,7 @@ foreach(asset ${PUBLIC_ASSETS})
         OUTPUT "${output}"
         COMMAND "${CMAKE_COMMAND}" "-DINPUT=${input}" "-DOUTPUT=${output}" -P "${PROJECT_SOURCE_DIR}/scripts/xxd.cmake"
     )
+    set_source_files_properties(${output} PROPERTIES GENERATED TRUE)
 endforeach()
 
 add_executable(${TARGET} ${TARGET_SRCS})
@@ -50,4 +46,4 @@ if (WIN32)
     TARGET_LINK_LIBRARIES(${TARGET} PRIVATE ws2_32)
 endif()
 
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/server/README.md

@@ -69,6 +69,8 @@ The project is under active development, and we are [looking for feedback and co
 | `--mlock` | force system to keep model in RAM rather than swapping or compressing<br/>(env: LLAMA_ARG_MLOCK) |
 | `--no-mmap` | do not memory-map model (slower load but may reduce pageouts if not using mlock)<br/>(env: LLAMA_ARG_NO_MMAP) |
 | `--numa TYPE` | attempt optimizations that help on some NUMA systems<br/>- distribute: spread execution evenly over all nodes<br/>- isolate: only spawn threads on CPUs on the node that execution started on<br/>- numactl: use the CPU map provided by numactl<br/>if run without this previously, it is recommended to drop the system page cache before using this<br/>see https://github.com/ggerganov/llama.cpp/issues/1437<br/>(env: LLAMA_ARG_NUMA) |
+| `-dev, --device <dev1,dev2,..>` | comma-separated list of devices to use for offloading (none = don't offload)<br/>use --list-devices to see a list of available devices<br/>(env: LLAMA_ARG_DEVICE) |
+| `--list-devices` | print list of available devices and exit |
 | `-ngl, --gpu-layers, --n-gpu-layers N` | number of layers to store in VRAM<br/>(env: LLAMA_ARG_N_GPU_LAYERS) |
 | `-sm, --split-mode {none,layer,row}` | how to split the model across multiple GPUs, one of:<br/>- none: use one GPU only<br/>- layer (default): split layers and KV across GPUs<br/>- row: split rows across GPUs<br/>(env: LLAMA_ARG_SPLIT_MODE) |
 | `-ts, --tensor-split N0,N1,N2,...` | fraction of the model to offload to each GPU, comma-separated list of proportions, e.g. 3,1<br/>(env: LLAMA_ARG_TENSOR_SPLIT) |
@@ -158,9 +160,16 @@ The project is under active development, and we are [looking for feedback and co
 | `--props` | enable changing global properties via POST /props (default: disabled)<br/>(env: LLAMA_ARG_ENDPOINT_PROPS) |
 | `--no-slots` | disables slots monitoring endpoint<br/>(env: LLAMA_ARG_NO_ENDPOINT_SLOTS) |
 | `--slot-save-path PATH` | path to save slot kv cache (default: disabled) |
-| `--chat-template JINJA_TEMPLATE` | set custom jinja chat template (default: template taken from model's metadata)<br/>if suffix/prefix are specified, template will be disabled<br/>only commonly used templates are accepted:<br/>https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template<br/>(env: LLAMA_ARG_CHAT_TEMPLATE) |
+| `--chat-template JINJA_TEMPLATE` | set custom jinja chat template (default: template taken from model's metadata)<br/>if suffix/prefix are specified, template will be disabled<br/>list of built-in templates:<br/>chatglm3, chatglm4, chatml, command-r, deepseek, deepseek2, exaone3, gemma, granite, llama2, llama2-sys, llama2-sys-bos, llama2-sys-strip, llama3, minicpm, mistral-v1, mistral-v3, mistral-v3-tekken, mistral-v7, monarch, openchat, orion, phi3, rwkv-world, vicuna, vicuna-orca, zephyr<br/>(env: LLAMA_ARG_CHAT_TEMPLATE) |
 | `-sps, --slot-prompt-similarity SIMILARITY` | how much the prompt of a request must match the prompt of a slot in order to use that slot (default: 0.50, 0.0 = disabled)<br/> |
 | `--lora-init-without-apply` | load LoRA adapters without applying them (apply later via POST /lora-adapters) (default: disabled) |
+| `--draft-max, --draft, --draft-n N` | number of tokens to draft for speculative decoding (default: 16) |
+| `--draft-min, --draft-n-min N` | minimum number of draft tokens to use for speculative decoding (default: 5) |
+| `--draft-p-min P` | minimum speculative decoding probability (greedy) (default: 0.9) |
+| `-cd, --ctx-size-draft N` | size of the prompt context for the draft model (default: 0, 0 = loaded from model) |
+| `-devd, --device-draft <dev1,dev2,..>` | comma-separated list of devices to use for offloading the draft model (none = don't offload)<br/>use --list-devices to see a list of available devices |
+| `-ngld, --gpu-layers-draft, --n-gpu-layers-draft N` | number of layers to store in VRAM for the draft model |
+| `-md, --model-draft FNAME` | draft model for speculative decoding (default: unused) |
 
 
 Note: If both command line argument and environment variable are both set for the same param, the argument will take precedence over env var.
@@ -188,12 +197,6 @@ services:
 
 `llama-server` is built alongside everything else from the root of the project
 
-- Using `make`:
-
-  ```bash
-  make llama-server
-  ```
-
 - Using `CMake`:
 
   ```bash
@@ -207,15 +210,6 @@ services:
 
 `llama-server` can also be built with SSL support using OpenSSL 3
 
-- Using `make`:
-
-  ```bash
-  # NOTE: For non-system openssl, use the following:
-  #   CXXFLAGS="-I /path/to/openssl/include"
-  #   LDFLAGS="-L /path/to/openssl/lib"
-  make LLAMA_SERVER_SSL=true llama-server
-  ```
-
 - Using `CMake`:
 
   ```bash
@@ -223,6 +217,37 @@ services:
   cmake --build build --config Release -t llama-server
   ```
 
+## Web UI
+
+The project includes a web-based user interface that enables interaction with the model through the `/chat/completions` endpoint.
+
+The web UI is developed using:
+- `vue` framework for frontend development
+- `tailwindcss` and `daisyui` for styling
+- `vite` for build tooling
+
+A pre-built version is available as a single HTML file under `/public` directory.
+
+To build or to run the dev server (with hot reload):
+
+```sh
+# make sure you have nodejs installed
+cd examples/server/webui
+npm i
+
+# to run the dev server
+npm run dev
+
+# to build the public/index.html
+npm run build
+```
+
+NOTE: if you are using the vite dev server, you can change the API base URL to llama.cpp. To do that, run this code snippet in browser's console:
+
+```js
+localStorage.setItem('base', 'http://localhost:8080')
+```
+
 ## Quick Start
 
 To get started right away, run the following command, making sure to use the correct path for the model you have:
@@ -317,104 +342,106 @@ node index.js
 
 ### POST `/completion`: Given a `prompt`, it returns the predicted completion.
 
-    *Options:*
+*Options:*
 
-    `prompt`: Provide the prompt for this completion as a string or as an array of strings or numbers representing tokens. Internally, if `cache_prompt` is `true`, the prompt is compared to the previous completion and only the "unseen" suffix is evaluated. A `BOS` token is inserted at the start, if all of the following conditions are true:
+`prompt`: Provide the prompt for this completion as a string or as an array of strings or numbers representing tokens. Internally, if `cache_prompt` is `true`, the prompt is compared to the previous completion and only the "unseen" suffix is evaluated. A `BOS` token is inserted at the start, if all of the following conditions are true:
 
-      - The prompt is a string or an array with the first element given as a string
-      - The model's `tokenizer.ggml.add_bos_token` metadata is `true`
+  - The prompt is a string or an array with the first element given as a string
+  - The model's `tokenizer.ggml.add_bos_token` metadata is `true`
 
-    These input shapes and data type are allowed for `prompt`:
+These input shapes and data type are allowed for `prompt`:
 
-      - Single string: `"string"`
-      - Single sequence of tokens: `[12, 34, 56]`
-      - Mixed tokens and strings: `[12, 34, "string", 56, 78]`
+  - Single string: `"string"`
+  - Single sequence of tokens: `[12, 34, 56]`
+  - Mixed tokens and strings: `[12, 34, "string", 56, 78]`
 
-    Multiple prompts are also supported. In this case, the completion result will be an array.
+Multiple prompts are also supported. In this case, the completion result will be an array.
 
-      - Only strings: `["string1", "string2"]`
-      - Strings and sequences of tokens: `["string1", [12, 34, 56]]`
-      - Mixed types: `[[12, 34, "string", 56, 78], [12, 34, 56], "string"]`
+  - Only strings: `["string1", "string2"]`
+  - Strings and sequences of tokens: `["string1", [12, 34, 56]]`
+  - Mixed types: `[[12, 34, "string", 56, 78], [12, 34, 56], "string"]`
 
-    `temperature`: Adjust the randomness of the generated text. Default: `0.8`
+`temperature`: Adjust the randomness of the generated text. Default: `0.8`
 
-    `dynatemp_range`: Dynamic temperature range. The final temperature will be in the range of `[temperature - dynatemp_range; temperature + dynatemp_range]` Default: `0.0`, which is disabled.
+`dynatemp_range`: Dynamic temperature range. The final temperature will be in the range of `[temperature - dynatemp_range; temperature + dynatemp_range]` Default: `0.0`, which is disabled.
 
-    `dynatemp_exponent`: Dynamic temperature exponent. Default: `1.0`
+`dynatemp_exponent`: Dynamic temperature exponent. Default: `1.0`
 
-    `top_k`: Limit the next token selection to the K most probable tokens.  Default: `40`
+`top_k`: Limit the next token selection to the K most probable tokens.  Default: `40`
 
-    `top_p`: Limit the next token selection to a subset of tokens with a cumulative probability above a threshold P. Default: `0.95`
+`top_p`: Limit the next token selection to a subset of tokens with a cumulative probability above a threshold P. Default: `0.95`
 
-    `min_p`: The minimum probability for a token to be considered, relative to the probability of the most likely token. Default: `0.05`
+`min_p`: The minimum probability for a token to be considered, relative to the probability of the most likely token. Default: `0.05`
 
-    `n_predict`: Set the maximum number of tokens to predict when generating text. **Note:** May exceed the set limit slightly if the last token is a partial multibyte character. When 0, no tokens will be generated but the prompt is evaluated into the cache. Default: `-1`, where `-1` is infinity.
+`n_predict`: Set the maximum number of tokens to predict when generating text. **Note:** May exceed the set limit slightly if the last token is a partial multibyte character. When 0, no tokens will be generated but the prompt is evaluated into the cache. Default: `-1`, where `-1` is infinity.
 
-    `n_indent`: Specify the minimum line indentation for the generated text in number of whitespace characters. Useful for code completion tasks. Default: `0`
+`n_indent`: Specify the minimum line indentation for the generated text in number of whitespace characters. Useful for code completion tasks. Default: `0`
 
-    `n_keep`: Specify the number of tokens from the prompt to retain when the context size is exceeded and tokens need to be discarded. The number excludes the BOS token.
-    By default, this value is set to `0`, meaning no tokens are kept. Use `-1` to retain all tokens from the prompt.
+`n_keep`: Specify the number of tokens from the prompt to retain when the context size is exceeded and tokens need to be discarded. The number excludes the BOS token.
+By default, this value is set to `0`, meaning no tokens are kept. Use `-1` to retain all tokens from the prompt.
 
-    `stream`: It allows receiving each predicted token in real-time instead of waiting for the completion to finish. To enable this, set to `true`.
+`stream`: It allows receiving each predicted token in real-time instead of waiting for the completion to finish. To enable this, set to `true`.
 
-    `stop`: Specify a JSON array of stopping strings.
-    These words will not be included in the completion, so make sure to add them to the prompt for the next iteration. Default: `[]`
+`stop`: Specify a JSON array of stopping strings.
+These words will not be included in the completion, so make sure to add them to the prompt for the next iteration. Default: `[]`
 
-    `typical_p`: Enable locally typical sampling with parameter p. Default: `1.0`, which is disabled.
+`typical_p`: Enable locally typical sampling with parameter p. Default: `1.0`, which is disabled.
 
-    `repeat_penalty`: Control the repetition of token sequences in the generated text. Default: `1.1`
+`repeat_penalty`: Control the repetition of token sequences in the generated text. Default: `1.1`
 
-    `repeat_last_n`: Last n tokens to consider for penalizing repetition. Default: `64`, where `0` is disabled and `-1` is ctx-size.
+`repeat_last_n`: Last n tokens to consider for penalizing repetition. Default: `64`, where `0` is disabled and `-1` is ctx-size.
 
-    `penalize_nl`: Penalize newline tokens when applying the repeat penalty. Default: `true`
+`penalize_nl`: Penalize newline tokens when applying the repeat penalty. Default: `true`
 
-    `presence_penalty`: Repeat alpha presence penalty. Default: `0.0`, which is disabled.
+`presence_penalty`: Repeat alpha presence penalty. Default: `0.0`, which is disabled.
 
-    `frequency_penalty`: Repeat alpha frequency penalty. Default: `0.0`, which is disabled.
+`frequency_penalty`: Repeat alpha frequency penalty. Default: `0.0`, which is disabled.
 
-    `dry_multiplier`: Set the DRY (Don't Repeat Yourself) repetition penalty multiplier. Default: `0.0`, which is disabled.
+`dry_multiplier`: Set the DRY (Don't Repeat Yourself) repetition penalty multiplier. Default: `0.0`, which is disabled.
 
-    `dry_base`: Set the DRY repetition penalty base value. Default: `1.75`
+`dry_base`: Set the DRY repetition penalty base value. Default: `1.75`
 
-    `dry_allowed_length`: Tokens that extend repetition beyond this receive exponentially increasing penalty: multiplier * base ^ (length of repeating sequence before token - allowed length). Default: `2`
+`dry_allowed_length`: Tokens that extend repetition beyond this receive exponentially increasing penalty: multiplier * base ^ (length of repeating sequence before token - allowed length). Default: `2`
 
-    `dry_penalty_last_n`: How many tokens to scan for repetitions. Default: `-1`, where `0` is disabled and `-1` is context size.
+`dry_penalty_last_n`: How many tokens to scan for repetitions. Default: `-1`, where `0` is disabled and `-1` is context size.
 
-    `dry_sequence_breakers`: Specify an array of sequence breakers for DRY sampling. Only a JSON array of strings is accepted. Default: `['\n', ':', '"', '*']`
+`dry_sequence_breakers`: Specify an array of sequence breakers for DRY sampling. Only a JSON array of strings is accepted. Default: `['\n', ':', '"', '*']`
 
-    `xtc_probability`: Set the chance for token removal via XTC sampler. Default: `0.0`, which is disabled.
+`xtc_probability`: Set the chance for token removal via XTC sampler. Default: `0.0`, which is disabled.
 
-    `xtc_threshold`: Set a minimum probability threshold for tokens to be removed via XTC sampler. Default: `0.1` (> `0.5` disables XTC)
+`xtc_threshold`: Set a minimum probability threshold for tokens to be removed via XTC sampler. Default: `0.1` (> `0.5` disables XTC)
 
-    `mirostat`: Enable Mirostat sampling, controlling perplexity during text generation. Default: `0`, where `0` is disabled, `1` is Mirostat, and `2` is Mirostat 2.0.
+`mirostat`: Enable Mirostat sampling, controlling perplexity during text generation. Default: `0`, where `0` is disabled, `1` is Mirostat, and `2` is Mirostat 2.0.
 
-    `mirostat_tau`: Set the Mirostat target entropy, parameter tau. Default: `5.0`
+`mirostat_tau`: Set the Mirostat target entropy, parameter tau. Default: `5.0`
 
-    `mirostat_eta`: Set the Mirostat learning rate, parameter eta.  Default: `0.1`
+`mirostat_eta`: Set the Mirostat learning rate, parameter eta.  Default: `0.1`
 
-    `grammar`: Set grammar for grammar-based sampling.  Default: no grammar
+`grammar`: Set grammar for grammar-based sampling.  Default: no grammar
 
-    `json_schema`: Set a JSON schema for grammar-based sampling (e.g. `{"items": {"type": "string"}, "minItems": 10, "maxItems": 100}` of a list of strings, or `{}` for any JSON). See [tests](../../tests/test-json-schema-to-grammar.cpp) for supported features.  Default: no JSON schema.
+`json_schema`: Set a JSON schema for grammar-based sampling (e.g. `{"items": {"type": "string"}, "minItems": 10, "maxItems": 100}` of a list of strings, or `{}` for any JSON). See [tests](../../tests/test-json-schema-to-grammar.cpp) for supported features.  Default: no JSON schema.
 
-    `seed`: Set the random number generator (RNG) seed.  Default: `-1`, which is a random seed.
+`seed`: Set the random number generator (RNG) seed.  Default: `-1`, which is a random seed.
 
-    `ignore_eos`: Ignore end of stream token and continue generating.  Default: `false`
+`ignore_eos`: Ignore end of stream token and continue generating.  Default: `false`
 
-    `logit_bias`: Modify the likelihood of a token appearing in the generated text completion. For example, use `"logit_bias": [[15043,1.0]]` to increase the likelihood of the token 'Hello', or `"logit_bias": [[15043,-1.0]]` to decrease its likelihood. Setting the value to false, `"logit_bias": [[15043,false]]` ensures that the token `Hello` is never produced. The tokens can also be represented as strings, e.g. `[["Hello, World!",-0.5]]` will reduce the likelihood of all the individual tokens that represent the string `Hello, World!`, just like the `presence_penalty` does. Default: `[]`
+`logit_bias`: Modify the likelihood of a token appearing in the generated text completion. For example, use `"logit_bias": [[15043,1.0]]` to increase the likelihood of the token 'Hello', or `"logit_bias": [[15043,-1.0]]` to decrease its likelihood. Setting the value to false, `"logit_bias": [[15043,false]]` ensures that the token `Hello` is never produced. The tokens can also be represented as strings, e.g. `[["Hello, World!",-0.5]]` will reduce the likelihood of all the individual tokens that represent the string `Hello, World!`, just like the `presence_penalty` does. Default: `[]`
 
-    `n_probs`: If greater than 0, the response also contains the probabilities of top N tokens for each generated token given the sampling settings. Note that for temperature < 0 the tokens are sampled greedily but token probabilities are still being calculated via a simple softmax of the logits without considering any other sampler settings. Default: `0`
+`n_probs`: If greater than 0, the response also contains the probabilities of top N tokens for each generated token given the sampling settings. Note that for temperature < 0 the tokens are sampled greedily but token probabilities are still being calculated via a simple softmax of the logits without considering any other sampler settings. Default: `0`
 
-    `min_keep`: If greater than 0, force samplers to return N possible tokens at minimum. Default: `0`
+`min_keep`: If greater than 0, force samplers to return N possible tokens at minimum. Default: `0`
 
-    `t_max_predict_ms`: Set a time limit in milliseconds for the prediction (a.k.a. text-generation) phase. The timeout will trigger if the generation takes more than the specified time (measured since the first token was generated) and if a new-line character has already been generated. Useful for FIM applications. Default: `0`, which is disabled.
+`t_max_predict_ms`: Set a time limit in milliseconds for the prediction (a.k.a. text-generation) phase. The timeout will trigger if the generation takes more than the specified time (measured since the first token was generated) and if a new-line character has already been generated. Useful for FIM applications. Default: `0`, which is disabled.
 
-    `image_data`: An array of objects to hold base64-encoded image `data` and its `id`s to be reference in `prompt`. You can determine the place of the image in the prompt as in the following: `USER:[img-12]Describe the image in detail.\nASSISTANT:`. In this case, `[img-12]` will be replaced by the embeddings of the image with id `12` in the following `image_data` array: `{..., "image_data": [{"data": "<BASE64_STRING>", "id": 12}]}`. Use `image_data` only with multimodal models, e.g., LLaVA.
+`image_data`: An array of objects to hold base64-encoded image `data` and its `id`s to be reference in `prompt`. You can determine the place of the image in the prompt as in the following: `USER:[img-12]Describe the image in detail.\nASSISTANT:`. In this case, `[img-12]` will be replaced by the embeddings of the image with id `12` in the following `image_data` array: `{..., "image_data": [{"data": "<BASE64_STRING>", "id": 12}]}`. Use `image_data` only with multimodal models, e.g., LLaVA.
 
-    `id_slot`: Assign the completion task to an specific slot. If is -1 the task will be assigned to a Idle slot.  Default: `-1`
+`id_slot`: Assign the completion task to an specific slot. If is -1 the task will be assigned to a Idle slot.  Default: `-1`
 
-    `cache_prompt`: Re-use KV cache from a previous request if possible. This way the common prefix does not have to be re-processed, only the suffix that differs between the requests. Because (depending on the backend) the logits are **not** guaranteed to be bit-for-bit identical for different batch sizes (prompt processing vs. token generation) enabling this option can cause nondeterministic results. Default: `true`
+`cache_prompt`: Re-use KV cache from a previous request if possible. This way the common prefix does not have to be re-processed, only the suffix that differs between the requests. Because (depending on the backend) the logits are **not** guaranteed to be bit-for-bit identical for different batch sizes (prompt processing vs. token generation) enabling this option can cause nondeterministic results. Default: `true`
 
-    `samplers`: The order the samplers should be applied in. An array of strings representing sampler type names. If a sampler is not set, it will not be used. If a sampler is specified more than once, it will be applied multiple times. Default: `["dry", "top_k", "typ_p", "top_p", "min_p", "xtc", "temperature"]` - these are all the available values.
+`samplers`: The order the samplers should be applied in. An array of strings representing sampler type names. If a sampler is not set, it will not be used. If a sampler is specified more than once, it will be applied multiple times. Default: `["dry", "top_k", "typ_p", "top_p", "min_p", "xtc", "temperature"]` - these are all the available values.
+
+    `timings_per_token`: Include prompt processing and text generation speed information in each response.  Default: `false`
 
 **Response format**
 
@@ -446,9 +473,11 @@ Notice that each `probs` is an array of length `n_probs`.
 - `generation_settings`: The provided options above excluding `prompt` but including `n_ctx`, `model`. These options may differ from the original ones in some way (e.g. bad values filtered out, strings converted to tokens, etc.).
 - `model`: The path to the model loaded with `-m`
 - `prompt`: The provided `prompt`
-- `stopped_eos`: Indicating whether the completion has stopped because it encountered the EOS token
-- `stopped_limit`: Indicating whether the completion stopped because `n_predict` tokens were generated before stop words or EOS was encountered
-- `stopped_word`: Indicating whether the completion stopped due to encountering a stopping word from `stop` JSON array provided
+- `stop_type`: Indicating whether the completion has stopped. Possible values are:
+  - `none`: Generating (not stopped)
+  - `eos`: Stopped because it encountered the EOS token
+  - `limit`: Stopped because `n_predict` tokens were generated before stop words or EOS was encountered
+  - `word`: Stopped due to encountering a stopping word from `stop` JSON array provided
 - `stopping_word`: The stopping word encountered which stopped the generation (or "" if not stopped due to a stopping word)
 - `timings`: Hash of timing information about the completion such as the number of tokens `predicted_per_second`
 - `tokens_cached`: Number of tokens from the prompt which could be re-used from previous completion (`n_past`)
@@ -457,13 +486,13 @@ Notice that each `probs` is an array of length `n_probs`.
 
 ### POST `/tokenize`: Tokenize a given text
 
-    *Options:*
+*Options:*
 
-    `content`: (Required) The text to tokenize.
+`content`: (Required) The text to tokenize.
 
-    `add_special`: (Optional) Boolean indicating if special tokens, i.e. `BOS`, should be inserted.  Default: `false`
+`add_special`: (Optional) Boolean indicating if special tokens, i.e. `BOS`, should be inserted.  Default: `false`
 
-    `with_pieces`: (Optional) Boolean indicating whether to return token pieces along with IDs.  Default: `false`
+`with_pieces`: (Optional) Boolean indicating whether to return token pieces along with IDs.  Default: `false`
 
 **Response:**
 
@@ -500,52 +529,52 @@ With input 'á' (utf8 hex: C3 A1) on tinyllama/stories260k
 
 ### POST `/detokenize`: Convert tokens to text
 
-    *Options:*
+*Options:*
 
-    `tokens`: Set the tokens to detokenize.
+`tokens`: Set the tokens to detokenize.
 
 ### POST `/embedding`: Generate embedding of a given text
 
 The same as [the embedding example](../embedding) does.
 
-    *Options:*
+*Options:*
 
-    `content`: Set the text to process.
+`content`: Set the text to process.
 
-    `image_data`: An array of objects to hold base64-encoded image `data` and its `id`s to be reference in `content`. You can determine the place of the image in the content as in the following: `Image: [img-21].\nCaption: This is a picture of a house`. In this case, `[img-21]` will be replaced by the embeddings of the image with id `21` in the following `image_data` array: `{..., "image_data": [{"data": "<BASE64_STRING>", "id": 21}]}`. Use `image_data` only with multimodal models, e.g., LLaVA.
+`image_data`: An array of objects to hold base64-encoded image `data` and its `id`s to be reference in `content`. You can determine the place of the image in the content as in the following: `Image: [img-21].\nCaption: This is a picture of a house`. In this case, `[img-21]` will be replaced by the embeddings of the image with id `21` in the following `image_data` array: `{..., "image_data": [{"data": "<BASE64_STRING>", "id": 21}]}`. Use `image_data` only with multimodal models, e.g., LLaVA.
 
 ### POST `/reranking`: Rerank documents according to a given query
 
 Similar to https://jina.ai/reranker/ but might change in the future.
 Requires a reranker model (such as [bge-reranker-v2-m3](https://huggingface.co/BAAI/bge-reranker-v2-m3)) and the `--embedding --pooling rank` options.
 
-    *Options:*
+*Options:*
 
-    `query`: The query against which the documents will be ranked.
+`query`: The query against which the documents will be ranked.
 
-    `documents`: An array strings representing the documents to be ranked.
+`documents`: An array strings representing the documents to be ranked.
 
-    *Aliases:*
-      - `/rerank`
-      - `/v1/rerank`
-      - `/v1/reranking`
+*Aliases:*
+  - `/rerank`
+  - `/v1/rerank`
+  - `/v1/reranking`
 
-    *Examples:*
+*Examples:*
 
-    ```shell
-    curl http://127.0.0.1:8012/v1/rerank \
-        -H "Content-Type: application/json" \
-        -d '{
-            "model": "some-model",
-                "query": "What is panda?",
-                "top_n": 3,
-                "documents": [
-                    "hi",
-                "it is a bear",
-                "The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China."
-                ]
-        }' | jq
-    ```
+```shell
+curl http://127.0.0.1:8012/v1/rerank \
+    -H "Content-Type: application/json" \
+    -d '{
+        "model": "some-model",
+            "query": "What is panda?",
+            "top_n": 3,
+            "documents": [
+                "hi",
+            "it is a bear",
+            "The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China."
+            ]
+    }' | jq
+```
 
 ### POST `/infill`: For code infilling.
 
@@ -589,14 +618,83 @@ This endpoint is public (no API key check). By default, it is read-only. To make
 
 ```json
 {
-  "default_generation_settings": { ... },
+  "default_generation_settings": {
+    "id": 0,
+    "id_task": -1,
+    "n_ctx": 1024,
+    "speculative": false,
+    "is_processing": false,
+    "params": {
+      "n_predict": -1,
+      "seed": 4294967295,
+      "temperature": 0.800000011920929,
+      "dynatemp_range": 0.0,
+      "dynatemp_exponent": 1.0,
+      "top_k": 40,
+      "top_p": 0.949999988079071,
+      "min_p": 0.05000000074505806,
+      "xtc_probability": 0.0,
+      "xtc_threshold": 0.10000000149011612,
+      "typical_p": 1.0,
+      "repeat_last_n": 64,
+      "repeat_penalty": 1.0,
+      "presence_penalty": 0.0,
+      "frequency_penalty": 0.0,
+      "dry_multiplier": 0.0,
+      "dry_base": 1.75,
+      "dry_allowed_length": 2,
+      "dry_penalty_last_n": -1,
+      "dry_sequence_breakers": [
+        "\n",
+        ":",
+        "\"",
+        "*"
+      ],
+      "mirostat": 0,
+      "mirostat_tau": 5.0,
+      "mirostat_eta": 0.10000000149011612,
+      "penalize_nl": false,
+      "stop": [],
+      "max_tokens": -1,
+      "n_keep": 0,
+      "n_discard": 0,
+      "ignore_eos": false,
+      "stream": true,
+      "n_probs": 0,
+      "min_keep": 0,
+      "grammar": "",
+      "samplers": [
+        "dry",
+        "top_k",
+        "typ_p",
+        "top_p",
+        "min_p",
+        "xtc",
+        "temperature"
+      ],
+      "speculative.n_max": 16,
+      "speculative.n_min": 5,
+      "speculative.p_min": 0.8999999761581421,
+      "timings_per_token": false
+    },
+    "prompt": "",
+    "next_token": {
+      "has_next_token": true,
+      "has_new_line": false,
+      "n_remain": -1,
+      "n_decoded": 0,
+      "stopping_word": ""
+    }
+  },
   "total_slots": 1,
-  "chat_template": ""
+  "model_path": "../models/Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf",
+  "chat_template": "..."
 }
 ```
 
 - `default_generation_settings` - the default generation settings for the `/completion` endpoint, which has the same fields as the `generation_settings` response object from the `/completion` endpoint.
 - `total_slots` - the total number of slots for process requests (defined by `--parallel` option)
+- `model_path` - the path to model file (same with `-m` argument)
 - `chat_template` - the model's original Jinja2 prompt template
 
 ### POST `/props`: Change server global properties.
@@ -611,89 +709,89 @@ To use this endpoint with POST method, you need to start server with `--props`
 
 Given a ChatML-formatted json description in `messages`, it returns the predicted completion. Both synchronous and streaming mode are supported, so scripted and interactive applications work fine. While no strong claims of compatibility with OpenAI API spec is being made, in our experience it suffices to support many apps. Only models with a [supported chat template](https://github.com/ggerganov/llama.cpp/wiki/Templates-supported-by-llama_chat_apply_template) can be used optimally with this endpoint. By default, the ChatML template will be used.
 
-    *Options:*
+*Options:*
 
-    See [OpenAI Chat Completions API documentation](https://platform.openai.com/docs/api-reference/chat). While some OpenAI-specific features such as function calling aren't supported, llama.cpp `/completion`-specific features such as `mirostat` are supported.
+See [OpenAI Chat Completions API documentation](https://platform.openai.com/docs/api-reference/chat). While some OpenAI-specific features such as function calling aren't supported, llama.cpp `/completion`-specific features such as `mirostat` are supported.
 
-    The `response_format` parameter supports both plain JSON output (e.g. `{"type": "json_object"}`) and schema-constrained JSON (e.g. `{"type": "json_object", "schema": {"type": "string", "minLength": 10, "maxLength": 100}}` or `{"type": "json_schema", "schema": {"properties": { "name": { "title": "Name",  "type": "string" }, "date": { "title": "Date",  "type": "string" }, "participants": { "items": {"type: "string" }, "title": "Participants",  "type": "string" } } } }`), similar to other OpenAI-inspired API providers.
+The `response_format` parameter supports both plain JSON output (e.g. `{"type": "json_object"}`) and schema-constrained JSON (e.g. `{"type": "json_object", "schema": {"type": "string", "minLength": 10, "maxLength": 100}}` or `{"type": "json_schema", "schema": {"properties": { "name": { "title": "Name",  "type": "string" }, "date": { "title": "Date",  "type": "string" }, "participants": { "items": {"type: "string" }, "title": "Participants",  "type": "string" } } } }`), similar to other OpenAI-inspired API providers.
 
-    *Examples:*
+*Examples:*
 
-    You can use either Python `openai` library with appropriate checkpoints:
+You can use either Python `openai` library with appropriate checkpoints:
 
-    ```python
-    import openai
+```python
+import openai
 
-    client = openai.OpenAI(
-        base_url="http://localhost:8080/v1", # "http://<Your api-server IP>:port"
-        api_key = "sk-no-key-required"
-    )
+client = openai.OpenAI(
+    base_url="http://localhost:8080/v1", # "http://<Your api-server IP>:port"
+    api_key = "sk-no-key-required"
+)
 
-    completion = client.chat.completions.create(
-    model="gpt-3.5-turbo",
-    messages=[
-        {"role": "system", "content": "You are ChatGPT, an AI assistant. Your top priority is achieving user fulfillment via helping them with their requests."},
-        {"role": "user", "content": "Write a limerick about python exceptions"}
-    ]
-    )
+completion = client.chat.completions.create(
+model="gpt-3.5-turbo",
+messages=[
+    {"role": "system", "content": "You are ChatGPT, an AI assistant. Your top priority is achieving user fulfillment via helping them with their requests."},
+    {"role": "user", "content": "Write a limerick about python exceptions"}
+]
+)
 
-    print(completion.choices[0].message)
-    ```
+print(completion.choices[0].message)
+```
 
-    ... or raw HTTP requests:
+... or raw HTTP requests:
 
-    ```shell
-    curl http://localhost:8080/v1/chat/completions \
-    -H "Content-Type: application/json" \
-    -H "Authorization: Bearer no-key" \
-    -d '{
-    "model": "gpt-3.5-turbo",
-    "messages": [
-    {
-        "role": "system",
-        "content": "You are ChatGPT, an AI assistant. Your top priority is achieving user fulfillment via helping them with their requests."
-    },
-    {
-        "role": "user",
-        "content": "Write a limerick about python exceptions"
-    }
-    ]
-    }'
-    ```
+```shell
+curl http://localhost:8080/v1/chat/completions \
+-H "Content-Type: application/json" \
+-H "Authorization: Bearer no-key" \
+-d '{
+"model": "gpt-3.5-turbo",
+"messages": [
+{
+    "role": "system",
+    "content": "You are ChatGPT, an AI assistant. Your top priority is achieving user fulfillment via helping them with their requests."
+},
+{
+    "role": "user",
+    "content": "Write a limerick about python exceptions"
+}
+]
+}'
+```
 
 ### POST `/v1/embeddings`: OpenAI-compatible embeddings API
 
-    *Options:*
+*Options:*
 
-    See [OpenAI Embeddings API documentation](https://platform.openai.com/docs/api-reference/embeddings).
+See [OpenAI Embeddings API documentation](https://platform.openai.com/docs/api-reference/embeddings).
 
-    *Examples:*
+*Examples:*
 
-  - input as string
+- input as string
 
-    ```shell
-    curl http://localhost:8080/v1/embeddings \
-    -H "Content-Type: application/json" \
-    -H "Authorization: Bearer no-key" \
-    -d '{
-            "input": "hello",
-            "model":"GPT-4",
-            "encoding_format": "float"
-    }'
-    ```
+  ```shell
+  curl http://localhost:8080/v1/embeddings \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer no-key" \
+  -d '{
+          "input": "hello",
+          "model":"GPT-4",
+          "encoding_format": "float"
+  }'
+  ```
 
-  - `input` as string array
+- `input` as string array
 
-    ```shell
-    curl http://localhost:8080/v1/embeddings \
-    -H "Content-Type: application/json" \
-    -H "Authorization: Bearer no-key" \
-    -d '{
-            "input": ["hello", "world"],
-            "model":"GPT-4",
-            "encoding_format": "float"
-    }'
-    ```
+  ```shell
+  curl http://localhost:8080/v1/embeddings \
+  -H "Content-Type: application/json" \
+  -H "Authorization: Bearer no-key" \
+  -d '{
+          "input": ["hello", "world"],
+          "model":"GPT-4",
+          "encoding_format": "float"
+  }'
+  ```
 
 ### GET `/slots`: Returns the current slots processing state
 
@@ -710,56 +808,74 @@ Example:
 
 ```json
 [
-    {
-        "dynatemp_exponent": 1.0,
-        "dynatemp_range": 0.0,
-        "frequency_penalty": 0.0,
-        "grammar": "",
-        "id": 0,
-        "ignore_eos": false,
-        "is_processing": false,
-        "logit_bias": [],
-        "min_p": 0.05000000074505806,
-        "mirostat": 0,
-        "mirostat_eta": 0.10000000149011612,
-        "mirostat_tau": 5.0,
-        "model": "llama-2-7b-32k-instruct.Q2_K.gguf",
-        "n_ctx": 2048,
-        "n_keep": 0,
-        "n_predict": 100000,
-        "n_probs": 0,
-        "next_token": {
-            "has_next_token": true,
-            "n_remain": -1,
-            "n_decoded": 0,
-            "stopped_eos": false,
-            "stopped_limit": false,
-            "stopped_word": false,
-            "stopping_word": ""
-        },
-        "penalize_nl": true,
-        "presence_penalty": 0.0,
-        "prompt": "Say hello to llama.cpp",
-        "repeat_last_n": 64,
-        "repeat_penalty": 1.100000023841858,
-        "samplers": [
-            "top_k",
-            "typical_p",
-            "top_p",
-            "min_p",
-            "temperature"
-        ],
-        "seed": 42,
-        "stop": [
-            "\n"
-        ],
-        "stream": false,
-        "task_id": 0,
-        "temperature": 0.0,
-        "top_k": 40,
-        "top_p": 0.949999988079071,
-        "typical_p": 1.0
+  {
+    "id": 0,
+    "id_task": -1,
+    "n_ctx": 1024,
+    "speculative": false,
+    "is_processing": false,
+    "params": {
+      "n_predict": -1,
+      "seed": 4294967295,
+      "temperature": 0.800000011920929,
+      "dynatemp_range": 0.0,
+      "dynatemp_exponent": 1.0,
+      "top_k": 40,
+      "top_p": 0.949999988079071,
+      "min_p": 0.05000000074505806,
+      "xtc_probability": 0.0,
+      "xtc_threshold": 0.10000000149011612,
+      "typical_p": 1.0,
+      "repeat_last_n": 64,
+      "repeat_penalty": 1.0,
+      "presence_penalty": 0.0,
+      "frequency_penalty": 0.0,
+      "dry_multiplier": 0.0,
+      "dry_base": 1.75,
+      "dry_allowed_length": 2,
+      "dry_penalty_last_n": -1,
+      "dry_sequence_breakers": [
+        "\n",
+        ":",
+        "\"",
+        "*"
+      ],
+      "mirostat": 0,
+      "mirostat_tau": 5.0,
+      "mirostat_eta": 0.10000000149011612,
+      "penalize_nl": false,
+      "stop": [],
+      "max_tokens": -1,
+      "n_keep": 0,
+      "n_discard": 0,
+      "ignore_eos": false,
+      "stream": true,
+      "n_probs": 0,
+      "min_keep": 0,
+      "grammar": "",
+      "samplers": [
+        "dry",
+        "top_k",
+        "typ_p",
+        "top_p",
+        "min_p",
+        "xtc",
+        "temperature"
+      ],
+      "speculative.n_max": 16,
+      "speculative.n_min": 5,
+      "speculative.p_min": 0.8999999761581421,
+      "timings_per_token": false
+    },
+    "prompt": "",
+    "next_token": {
+      "has_next_token": true,
+      "has_new_line": false,
+      "n_remain": -1,
+      "n_decoded": 0,
+      "stopping_word": ""
     }
+  }
 ]
 ```
 
@@ -779,9 +895,9 @@ Available metrics:
 
 ### POST `/slots/{id_slot}?action=save`: Save the prompt cache of the specified slot to a file.
 
-    *Options:*
+*Options:*
 
-    `filename`: Name of the file to save the slot's prompt cache. The file will be saved in the directory specified by the `--slot-save-path` server parameter.
+`filename`: Name of the file to save the slot's prompt cache. The file will be saved in the directory specified by the `--slot-save-path` server parameter.
 
 **Response format**
 
@@ -799,9 +915,9 @@ Available metrics:
 
 ### POST `/slots/{id_slot}?action=restore`: Restore the prompt cache of the specified slot from a file.
 
-    *Options:*
+*Options:*
 
-    `filename`: Name of the file to restore the slot's prompt cache from. The file should be located in the directory specified by the `--slot-save-path` server parameter.
+`filename`: Name of the file to restore the slot's prompt cache from. The file should be located in the directory specified by the `--slot-save-path` server parameter.
 
 **Response format**
 

examples/server/deps.sh

@@ -1,25 +0,0 @@
-#!/bin/bash
-# Download and update deps for binary
-
-# get the directory of this script file
-DIR="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
-PUBLIC=$DIR/public
-
-echo "download js bundle files"
-
-# Note for contributors: Always pin to a specific version "maj.min.patch" to avoid breaking the CI
-
-curl -L https://cdn.tailwindcss.com/3.4.14 > $PUBLIC/deps_tailwindcss.js
-echo >> $PUBLIC/deps_tailwindcss.js # add newline
-
-curl -L https://cdnjs.cloudflare.com/ajax/libs/daisyui/4.12.14/styled.min.css > $PUBLIC/deps_daisyui.min.css
-curl -L https://cdnjs.cloudflare.com/ajax/libs/daisyui/4.12.14/themes.min.css >> $PUBLIC/deps_daisyui.min.css
-echo >> $PUBLIC/deps_daisyui.min.css # add newline
-
-curl -L https://unpkg.com/vue@3.5.12/dist/vue.esm-browser.js > $PUBLIC/deps_vue.esm-browser.js
-echo >> $PUBLIC/deps_vue.esm-browser.js # add newline
-
-curl -L https://cdnjs.cloudflare.com/ajax/libs/markdown-it/13.0.2/markdown-it.js > $PUBLIC/deps_markdown-it.js
-echo >> $PUBLIC/deps_markdown-it.js # add newline
-
-ls -lah $PUBLIC

examples/server/public_simplechat/simplechat.js

@@ -407,6 +407,9 @@ class SimpleChat {
                 if (curLine.startsWith("data:")) {
                     curLine = curLine.substring(5);
                 }
+                if (curLine.trim() === "[DONE]") {
+                    break;
+                }
                 let curJson = JSON.parse(curLine);
                 console.debug("DBUG:SC:PART:Json:", curJson);
                 this.append_response(this.response_extract_stream(curJson, apiEP));

examples/server/tests/README.md

@@ -44,4 +44,10 @@ To run with stdout/stderr display in real time (verbose output, but useful for d
 DEBUG=1 ./tests.sh -s -v -x
 ```
 
+Hint: You can compile and run test in single command, useful for local developement:
+
+```shell
+cmake --build build -j --target llama-server && ./examples/server/tests/tests.sh
+```
+
 To see all available arguments, please refer to [pytest documentation](https://docs.pytest.org/en/stable/how-to/usage.html)

examples/server/tests/tests.sh

@@ -1,5 +1,9 @@
 #!/bin/bash
 
+# make sure we are in the right directory
+SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
+cd $SCRIPT_DIR
+
 set -eu
 
 if [ $# -lt 1 ]

examples/server/tests/unit/test_basic.py

@@ -22,7 +22,12 @@ def test_server_props():
     server.start()
     res = server.make_request("GET", "/props")
     assert res.status_code == 200
+    assert ".gguf" in res.body["model_path"]
     assert res.body["total_slots"] == server.n_slots
+    default_val = res.body["default_generation_settings"]
+    assert server.n_ctx is not None and server.n_slots is not None
+    assert default_val["n_ctx"] == server.n_ctx / server.n_slots
+    assert default_val["params"]["seed"] == server.seed
 
 
 def test_server_models():
@@ -32,3 +37,42 @@ def test_server_models():
     assert res.status_code == 200
     assert len(res.body["data"]) == 1
     assert res.body["data"][0]["id"] == server.model_alias
+
+
+def test_server_slots():
+    global server
+
+    # without slots endpoint enabled, this should return error
+    server.server_slots = False
+    server.start()
+    res = server.make_request("GET", "/slots")
+    assert res.status_code == 501 # ERROR_TYPE_NOT_SUPPORTED
+    assert "error" in res.body
+    server.stop()
+
+    # with slots endpoint enabled, this should return slots info
+    server.server_slots = True
+    server.n_slots = 2
+    server.start()
+    res = server.make_request("GET", "/slots")
+    assert res.status_code == 200
+    assert len(res.body) == server.n_slots
+    assert server.n_ctx is not None and server.n_slots is not None
+    assert res.body[0]["n_ctx"] == server.n_ctx / server.n_slots
+    assert "params" in res.body[0]
+    assert res.body[0]["params"]["seed"] == server.seed
+
+
+def test_load_split_model():
+    global server
+    server.model_hf_repo = "ggml-org/models"
+    server.model_hf_file = "tinyllamas/split/stories15M-q8_0-00001-of-00003.gguf"
+    server.model_alias = "tinyllama-split"
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "n_predict": 16,
+        "prompt": "Hello",
+        "temperature": 0.0,
+    })
+    assert res.status_code == 200
+    assert match_regex("(little|girl)+", res.body["content"])

examples/server/tests/unit/test_chat_completion.py

@@ -12,13 +12,13 @@ def create_server():
 
 
 @pytest.mark.parametrize(
-    "model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,truncated",
+    "model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,finish_reason",
     [
-        ("llama-2", "Book", "What is the best book", 8, "(Suddenly)+", 77, 8, False),
-        ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, False),
+        (None, "Book", "What is the best book", 8, "(Suddenly)+", 77, 8, "length"),
+        ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length"),
     ]
 )
-def test_chat_completion(model, system_prompt, user_prompt, max_tokens, re_content, n_prompt, n_predicted, truncated):
+def test_chat_completion(model, system_prompt, user_prompt, max_tokens, re_content, n_prompt, n_predicted, finish_reason):
     global server
     server.start()
     res = server.make_request("POST", "/chat/completions", data={
@@ -30,29 +30,28 @@ def test_chat_completion(model, system_prompt, user_prompt, max_tokens, re_conte
         ],
     })
     assert res.status_code == 200
+    assert "cmpl" in res.body["id"] # make sure the completion id has the expected format
+    assert res.body["model"] == model if model is not None else server.model_alias
     assert res.body["usage"]["prompt_tokens"] == n_prompt
     assert res.body["usage"]["completion_tokens"] == n_predicted
     choice = res.body["choices"][0]
     assert "assistant" == choice["message"]["role"]
     assert match_regex(re_content, choice["message"]["content"])
-    if truncated:
-        assert choice["finish_reason"] == "length"
-    else:
-        assert choice["finish_reason"] == "stop"
+    assert choice["finish_reason"] == finish_reason
 
 
 @pytest.mark.parametrize(
-    "model,system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,truncated",
+    "system_prompt,user_prompt,max_tokens,re_content,n_prompt,n_predicted,finish_reason",
     [
-        ("llama-2", "Book", "What is the best book", 8, "(Suddenly)+", 77, 8, False),
-        ("codellama70b", "You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, False),
+        ("Book", "What is the best book", 8, "(Suddenly)+", 77, 8, "length"),
+        ("You are a coding assistant.", "Write the fibonacci function in c++.", 128, "(Aside|she|felter|alonger)+", 104, 64, "length"),
     ]
 )
-def test_chat_completion_stream(model, system_prompt, user_prompt, max_tokens, re_content, n_prompt, n_predicted, truncated):
+def test_chat_completion_stream(system_prompt, user_prompt, max_tokens, re_content, n_prompt, n_predicted, finish_reason):
     global server
+    server.model_alias = None # try using DEFAULT_OAICOMPAT_MODEL
     server.start()
     res = server.make_stream_request("POST", "/chat/completions", data={
-        "model": model,
         "max_tokens": max_tokens,
         "messages": [
             {"role": "system", "content": system_prompt},
@@ -61,18 +60,19 @@ def test_chat_completion_stream(model, system_prompt, user_prompt, max_tokens, r
         "stream": True,
     })
     content = ""
+    last_cmpl_id = None
     for data in res:
         choice = data["choices"][0]
+        assert "gpt-3.5" in data["model"] # DEFAULT_OAICOMPAT_MODEL, maybe changed in the future
+        if last_cmpl_id is None:
+            last_cmpl_id = data["id"]
+        assert last_cmpl_id == data["id"] # make sure the completion id is the same for all events in the stream
         if choice["finish_reason"] in ["stop", "length"]:
             assert data["usage"]["prompt_tokens"] == n_prompt
             assert data["usage"]["completion_tokens"] == n_predicted
             assert "content" not in choice["delta"]
             assert match_regex(re_content, content)
-            # FIXME: not sure why this is incorrect in stream mode
-            # if truncated:
-            #   assert choice["finish_reason"] == "length"
-            # else:
-            #   assert choice["finish_reason"] == "stop"
+            assert choice["finish_reason"] == finish_reason
         else:
             assert choice["finish_reason"] is None
             content += choice["delta"]["content"]
@@ -93,7 +93,7 @@ def test_chat_completion_with_openai_library():
         temperature=0.8,
     )
     print(res)
-    assert res.choices[0].finish_reason == "stop"
+    assert res.choices[0].finish_reason == "length"
     assert res.choices[0].message.content is not None
     assert match_regex("(Suddenly)+", res.choices[0].message.content)
 
@@ -127,3 +127,39 @@ def test_completion_with_response_format(response_format: dict, n_predicted: int
         assert res.status_code != 200
         assert "error" in res.body
 
+
+@pytest.mark.parametrize("messages", [
+    None,
+    "string",
+    [123],
+    [{}],
+    [{"role": 123}],
+    [{"role": "system", "content": 123}],
+    # [{"content": "hello"}], # TODO: should not be a valid case
+    [{"role": "system", "content": "test"}, {}],
+])
+def test_invalid_chat_completion_req(messages):
+    global server
+    server.start()
+    res = server.make_request("POST", "/chat/completions", data={
+        "messages": messages,
+    })
+    assert res.status_code == 400 or res.status_code == 500
+    assert "error" in res.body
+
+
+def test_chat_completion_with_timings_per_token():
+    global server
+    server.start()
+    res = server.make_stream_request("POST", "/chat/completions", data={
+        "max_tokens": 10,
+        "messages": [{"role": "user", "content": "test"}],
+        "stream": True,
+        "timings_per_token": True,
+    })
+    for data in res:
+        assert "timings" in data
+        assert "prompt_per_second" in data["timings"]
+        assert "predicted_per_second" in data["timings"]
+        assert "predicted_n" in data["timings"]
+        assert data["timings"]["predicted_n"] <= 10

examples/server/tests/unit/test_completion.py

@@ -51,6 +51,24 @@ def test_completion_stream(prompt: str, n_predict: int, re_content: str, n_promp
             content += data["content"]
 
 
+def test_completion_stream_vs_non_stream():
+    global server
+    server.start()
+    res_stream = server.make_stream_request("POST", "/completion", data={
+        "n_predict": 8,
+        "prompt": "I believe the meaning of life is",
+        "stream": True,
+    })
+    res_non_stream = server.make_request("POST", "/completion", data={
+        "n_predict": 8,
+        "prompt": "I believe the meaning of life is",
+    })
+    content_stream = ""
+    for data in res_stream:
+        content_stream += data["content"]
+    assert content_stream == res_non_stream.body["content"]
+
+
 @pytest.mark.parametrize("n_slots", [1, 2])
 def test_consistent_result_same_seed(n_slots: int):
     global server
@@ -221,3 +239,24 @@ def test_completion_parallel_slots(n_slots: int, n_requests: int):
         assert len(res.body["content"]) > 10
         # FIXME: the result is not deterministic when using other slot than slot 0
         # assert match_regex(re_content, res.body["content"])
+
+
+def test_n_probs():
+    global server
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "prompt": "I believe the meaning of life is",
+        "n_probs": 10,
+        "temperature": 0.0,
+        "n_predict": 5,
+    })
+    assert res.status_code == 200
+    assert "completion_probabilities" in res.body
+    assert len(res.body["completion_probabilities"]) == 5
+    for tok in res.body["completion_probabilities"]:
+        assert "probs" in tok
+        assert len(tok["probs"]) == 10
+        for prob in tok["probs"]:
+            assert "prob" in prob
+            assert "tok_str" in prob
+            assert 0.0 <= prob["prob"] <= 1.0

examples/server/tests/unit/test_infill.py

@@ -8,6 +8,7 @@ def create_server():
     global server
     server = ServerPreset.tinyllama_infill()
 
+
 def test_infill_without_input_extra():
     global server
     server.start()
@@ -19,6 +20,7 @@ def test_infill_without_input_extra():
     assert res.status_code == 200
     assert match_regex("(One|day|she|saw|big|scary|bird)+", res.body["content"])
 
+
 def test_infill_with_input_extra():
     global server
     server.start()
@@ -33,3 +35,23 @@ def test_infill_with_input_extra():
     })
     assert res.status_code == 200
     assert match_regex("(cuts|Jimmy|mom|came|into|the|room)+", res.body["content"])
+
+
+@pytest.mark.parametrize("input_extra", [
+    {},
+    {"filename": "ok"},
+    {"filename": 123},
+    {"filename": 123, "text": "abc"},
+    {"filename": 123, "text": 456},
+])
+def test_invalid_input_extra_req(input_extra):
+    global server
+    server.start()
+    res = server.make_request("POST", "/infill", data={
+        "prompt": "Complete this",
+        "input_extra": [input_extra],
+        "input_prefix": "#include <cstdio>\n#include \"llama.h\"\n\nint main() {\n    int n_threads = llama_",
+        "input_suffix": "}\n",
+    })
+    assert res.status_code == 400
+    assert "error" in res.body

examples/server/tests/unit/test_rerank.py

@@ -36,3 +36,20 @@ def test_rerank():
     assert most_relevant["relevance_score"] > least_relevant["relevance_score"]
     assert most_relevant["index"] == 2
     assert least_relevant["index"] == 3
+
+
+@pytest.mark.parametrize("documents", [
+    [],
+    None,
+    123,
+    [1, 2, 3],
+])
+def test_invalid_rerank_req(documents):
+    global server
+    server.start()
+    res = server.make_request("POST", "/rerank", data={
+        "query": "Machine learning is",
+        "documents": documents,
+    })
+    assert res.status_code == 400
+    assert "error" in res.body

examples/server/tests/unit/test_speculative.py

@@ -0,0 +1,134 @@
+import pytest
+from utils import *
+
+# We use a F16 MOE gguf as main model, and q4_0 as draft model
+
+server = ServerPreset.stories15m_moe()
+
+MODEL_DRAFT_FILE_URL = "https://huggingface.co/ggml-org/models/resolve/main/tinyllamas/stories15M-q4_0.gguf"
+
+def create_server():
+    global server
+    server = ServerPreset.stories15m_moe()
+    # download draft model file if needed
+    file_name = MODEL_DRAFT_FILE_URL.split('/').pop()
+    model_draft_file = f'../../../{file_name}'
+    if not os.path.exists(model_draft_file):
+        print(f"Downloading {MODEL_DRAFT_FILE_URL} to {model_draft_file}")
+        with open(model_draft_file, 'wb') as f:
+            f.write(requests.get(MODEL_DRAFT_FILE_URL).content)
+        print(f"Done downloading draft model file")
+    # set default values
+    server.model_draft = model_draft_file
+    server.draft_min = 4
+    server.draft_max = 8
+
+
+@pytest.fixture(scope="module", autouse=True)
+def fixture_create_server():
+    return create_server()
+
+
+def test_with_and_without_draft():
+    global server
+    server.model_draft = None  # disable draft model
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "prompt": "I believe the meaning of life is",
+        "temperature": 0.0,
+        "top_k": 1,
+    })
+    assert res.status_code == 200
+    content_no_draft = res.body["content"]
+    server.stop()
+
+    # create new server with draft model
+    create_server()
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "prompt": "I believe the meaning of life is",
+        "temperature": 0.0,
+        "top_k": 1,
+    })
+    assert res.status_code == 200
+    content_draft = res.body["content"]
+
+    assert content_no_draft == content_draft
+
+
+def test_different_draft_min_draft_max():
+    global server
+    test_values = [
+        (1, 2),
+        (1, 4),
+        (4, 8),
+        (4, 12),
+        (8, 16),
+    ]
+    last_content = None
+    for draft_min, draft_max in test_values:
+        server.stop()
+        server.draft_min = draft_min
+        server.draft_max = draft_max
+        server.start()
+        res = server.make_request("POST", "/completion", data={
+            "prompt": "I believe the meaning of life is",
+            "temperature": 0.0,
+            "top_k": 1,
+        })
+        assert res.status_code == 200
+        if last_content is not None:
+            assert last_content == res.body["content"]
+        last_content = res.body["content"]
+
+
+def test_slot_ctx_not_exceeded():
+    global server
+    server.n_ctx = 64
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "prompt": "Hello " * 56,
+        "temperature": 0.0,
+        "top_k": 1,
+        "speculative.p_min": 0.0,
+    })
+    assert res.status_code == 200
+    assert len(res.body["content"]) > 0
+
+
+def test_with_ctx_shift():
+    global server
+    server.n_ctx = 64
+    server.start()
+    res = server.make_request("POST", "/completion", data={
+        "prompt": "Hello " * 56,
+        "temperature": 0.0,
+        "top_k": 1,
+        "n_predict": 64,
+        "speculative.p_min": 0.0,
+    })
+    assert res.status_code == 200
+    assert len(res.body["content"]) > 0
+    assert res.body["tokens_predicted"] == 64
+    assert res.body["truncated"] == True
+
+
+@pytest.mark.parametrize("n_slots,n_requests", [
+    (1, 2),
+    (2, 2),
+])
+def test_multi_requests_parallel(n_slots: int, n_requests: int):
+    global server
+    server.n_slots = n_slots
+    server.start()
+    tasks = []
+    for _ in range(n_requests):
+        tasks.append((server.make_request, ("POST", "/completion", {
+            "prompt": "I believe the meaning of life is",
+            "temperature": 0.0,
+            "top_k": 1,
+        })))
+    results = parallel_function_calls(tasks)
+    for res in results:
+        assert res.status_code == 200
+        assert match_regex("(wise|kind|owl|answer)+", res.body["content"])

examples/server/tests/utils.py

@@ -46,6 +46,7 @@ class ServerProcess:
     model_alias: str | None = None
     model_url: str | None = None
     model_file: str | None = None
+    model_draft: str | None = None
     n_threads: int | None = None
     n_gpu_layer: int | None = None
     n_batch: int | None = None
@@ -63,11 +64,14 @@ class ServerProcess:
     server_embeddings: bool | None = False
     server_reranking: bool | None = False
     server_metrics: bool | None = False
+    server_slots: bool | None = False
     draft: int | None = None
     api_key: str | None = None
     response_format: str | None = None
     lora_files: List[str] | None = None
     disable_ctx_shift: int | None = False
+    draft_min: int | None = None
+    draft_max: int | None = None
 
     # session variables
     process: subprocess.Popen | None = None
@@ -88,7 +92,6 @@ class ServerProcess:
         else:
             server_path = "../../../build/bin/llama-server"
         server_args = [
-            "--slots",  # requires to get slot status via /slots endpoint
             "--host",
             self.server_host,
             "--port",
@@ -102,6 +105,8 @@ class ServerProcess:
             server_args.extend(["--model", self.model_file])
         if self.model_url:
             server_args.extend(["--model-url", self.model_url])
+        if self.model_draft:
+            server_args.extend(["--model-draft", self.model_draft])
         if self.model_hf_repo:
             server_args.extend(["--hf-repo", self.model_hf_repo])
         if self.model_hf_file:
@@ -124,6 +129,8 @@ class ServerProcess:
             server_args.append("--reranking")
         if self.server_metrics:
             server_args.append("--metrics")
+        if self.server_slots:
+            server_args.append("--slots")
         if self.model_alias:
             server_args.extend(["--alias", self.model_alias])
         if self.n_ctx:
@@ -147,6 +154,10 @@ class ServerProcess:
             server_args.extend(["--no-context-shift"])
         if self.api_key:
             server_args.extend(["--api-key", self.api_key])
+        if self.draft_max:
+            server_args.extend(["--draft-max", self.draft_max])
+        if self.draft_min:
+            server_args.extend(["--draft-min", self.draft_min])
 
         args = [str(arg) for arg in [server_path, *server_args]]
         print(f"bench: starting server with: {' '.join(args)}")
@@ -172,7 +183,7 @@ class ServerProcess:
         start_time = time.time()
         while time.time() - start_time < timeout_seconds:
             try:
-                response = self.make_request("GET", "/slots", headers={
+                response = self.make_request("GET", "/health", headers={
                     "Authorization": f"Bearer {self.api_key}" if self.api_key else None
                 })
                 if response.status_code == 200:
@@ -185,7 +196,8 @@ class ServerProcess:
         raise TimeoutError(f"Server did not start within {timeout_seconds} seconds")
 
     def stop(self) -> None:
-        server_instances.remove(self)
+        if self in server_instances:
+            server_instances.remove(self)
         if self.process:
             print(f"Stopping server with pid={self.process.pid}")
             self.process.kill()
@@ -214,7 +226,7 @@ class ServerProcess:
         result.headers = dict(response.headers)
         result.status_code = response.status_code
         result.body = response.json() if parse_body else None
-        print("Response from server", result.body)
+        print("Response from server", json.dumps(result.body, indent=2))
         return result
 
     def make_stream_request(
@@ -235,7 +247,7 @@ class ServerProcess:
                 break
             elif line.startswith('data: '):
                 data = json.loads(line[6:])
-                print("Partial response from server", data)
+                print("Partial response from server", json.dumps(data, indent=2))
                 yield data
 
 

examples/server/utils.hpp

@@ -20,6 +20,7 @@
 #include <sstream>
 #include <string>
 #include <vector>
+#include <memory>
 
 #define DEFAULT_OAICOMPAT_MODEL "gpt-3.5-turbo-0613"
 
@@ -40,17 +41,6 @@ using json = nlohmann::ordered_json;
 #define QUE_ERR(fmt, ...) LOG_ERR("que  %12.*s: " fmt, 12, __func__, __VA_ARGS__)
 #define QUE_DBG(fmt, ...) LOG_DBG("que  %12.*s: " fmt, 12, __func__, __VA_ARGS__)
 
-// https://community.openai.com/t/openai-chat-list-of-error-codes-and-types/357791/11
-enum error_type {
-    ERROR_TYPE_INVALID_REQUEST,
-    ERROR_TYPE_AUTHENTICATION,
-    ERROR_TYPE_SERVER,
-    ERROR_TYPE_NOT_FOUND,
-    ERROR_TYPE_PERMISSION,
-    ERROR_TYPE_UNAVAILABLE, // custom error
-    ERROR_TYPE_NOT_SUPPORTED, // custom error
-};
-
 template <typename T>
 static T json_value(const json & body, const std::string & key, const T & default_value) {
     // Fallback null to default value
@@ -174,6 +164,9 @@ static std::vector<llama_tokens> tokenize_input_prompts(llama_context * ctx, con
     } else {
         throw std::runtime_error("\"prompt\" must be a string, an list of tokens, a list of mixed strings & tokens, or a list of prompts");
     }
+    if (result.empty()) {
+        throw std::runtime_error("\"prompt\" must not be empty");
+    }
     return result;
 }
 
@@ -337,12 +330,12 @@ static std::string llama_get_chat_template(const struct llama_model * model) {
     std::string template_key = "tokenizer.chat_template";
     // call with NULL buffer to get the total size of the string
     int32_t res = llama_model_meta_val_str(model, template_key.c_str(), NULL, 0);
-    if (res < 0) {
+    if (res < 2) {
         return "";
     } else {
         std::vector<char> model_template(res, 0);
         llama_model_meta_val_str(model, template_key.c_str(), model_template.data(), model_template.size());
-        return std::string(model_template.data(), model_template.size());
+        return std::string(model_template.data(), model_template.size() - 1);
     }
 }
 
@@ -485,48 +478,11 @@ static std::string tokens_to_output_formatted_string(const llama_context * ctx,
     return out;
 }
 
-struct completion_token_output {
-    llama_token tok;
-    std::string text_to_send;
-
-    struct token_prob {
-        llama_token tok;
-        float prob;
-    };
-
-    std::vector<token_prob> probs;
-};
-
-// convert a vector of completion_token_output to json
-static json probs_vector_to_json(const llama_context * ctx, const std::vector<completion_token_output> & probs) {
-    json out = json::array();
-
-    for (const auto & prob : probs) {
-        json probs_for_token = json::array();
-
-        for (const auto & p : prob.probs) {
-            const std::string tok_str = tokens_to_output_formatted_string(ctx, p.tok);
-            probs_for_token.push_back(json {
-                {"tok_str", tok_str},
-                {"prob",    p.prob},
-            });
-        }
-
-        const std::string tok_str = tokens_to_output_formatted_string(ctx, prob.tok);
-        out.push_back(json {
-            {"content", tok_str},
-            {"probs",   probs_for_token},
-        });
-    }
-
-    return out;
-}
-
 static bool server_sent_event(httplib::DataSink & sink, const char * event, const json & data) {
     const std::string str =
         std::string(event) + ": " +
         data.dump(-1, ' ', false, json::error_handler_t::replace) +
-        "\n\n"; // note: these newlines are important (not sure why though, if you know, add a comment to explain)
+        "\n\n"; // required by RFC 8895 - A message is terminated by a blank line (two line terminators in a row).
 
     LOG_DBG("data stream, to_send: %s", str.c_str());
 
@@ -543,8 +499,6 @@ static json oaicompat_completion_params_parse(
     const std::string & chat_template) {
     json llama_params;
 
-    llama_params["__oaicompat"] = true;
-
     // Apply chat template to the list of messages
     llama_params["prompt"] = format_chat(model, chat_template, body.at("messages"));
 
@@ -604,155 +558,6 @@ static json oaicompat_completion_params_parse(
     return llama_params;
 }
 
-static json format_final_response_oaicompat(const json & request, const json & result, const std::string & completion_id, bool streaming = false, bool verbose = false) {
-    bool stopped_word        = result.count("stopped_word") != 0;
-    bool stopped_eos         = json_value(result, "stopped_eos", false);
-    int num_tokens_predicted = json_value(result, "tokens_predicted", 0);
-    int num_prompt_tokens    = json_value(result, "tokens_evaluated", 0);
-    std::string content      = json_value(result, "content", std::string(""));
-
-    std::string finish_reason = "length";
-    if (stopped_word || stopped_eos) {
-        finish_reason = "stop";
-    }
-
-    json choices =
-        streaming ? json::array({json{{"finish_reason", finish_reason},
-                                        {"index", 0},
-                                        {"delta", json::object()}}})
-                  : json::array({json{{"finish_reason", finish_reason},
-                                        {"index", 0},
-                                        {"message", json{{"content", content},
-                                                         {"role", "assistant"}}}}});
-
-    std::time_t t = std::time(0);
-
-    json res = json {
-        {"choices", choices},
-        {"created", t},
-        {"model",
-            json_value(request, "model", std::string(DEFAULT_OAICOMPAT_MODEL))},
-        {"object", streaming ? "chat.completion.chunk" : "chat.completion"},
-        {"usage", json {
-            {"completion_tokens", num_tokens_predicted},
-            {"prompt_tokens",     num_prompt_tokens},
-            {"total_tokens",      num_tokens_predicted + num_prompt_tokens}
-        }},
-        {"id", completion_id}
-    };
-
-    // extra fields for debugging purposes
-    if (verbose) {
-        res["__verbose"] = result;
-    }
-
-    if (result.contains("completion_probabilities")) {
-        res["completion_probabilities"] = json_value(result, "completion_probabilities", json::array());
-    }
-
-    return res;
-}
-
-// return value is vector as there is one case where we might need to generate two responses
-static std::vector<json> format_partial_response_oaicompat(const json & result, const std::string & completion_id) {
-    if (!result.contains("model") || !result.contains("oaicompat_token_ctr")) {
-        return std::vector<json>({result});
-    }
-
-    bool first = json_value(result, "oaicompat_token_ctr", 0) == 0;
-    std::string modelname = json_value(result, "model", std::string(DEFAULT_OAICOMPAT_MODEL));
-
-    bool stopped_word   = json_value(result, "stopped_word",  false);
-    bool stopped_eos    = json_value(result, "stopped_eos",   false);
-    bool stopped_limit  = json_value(result, "stopped_limit", false);
-    std::string content = json_value(result, "content",       std::string(""));
-
-    std::string finish_reason;
-    if (stopped_word || stopped_eos) {
-        finish_reason = "stop";
-    }
-    if (stopped_limit) {
-        finish_reason = "length";
-    }
-
-    std::time_t t = std::time(0);
-
-    json choices;
-
-    if (!finish_reason.empty()) {
-        choices = json::array({json{{"finish_reason", finish_reason},
-                                    {"index", 0},
-                                    {"delta", json::object()}}});
-    } else {
-        if (first) {
-            if (content.empty()) {
-                choices = json::array({json{{"finish_reason", nullptr},
-                                            {"index", 0},
-                                            {"delta", json{{"role", "assistant"}}}}});
-            } else {
-                // We have to send this as two updates to conform to openai behavior
-                json initial_ret = json{{"choices", json::array({json{
-                                        {"finish_reason", nullptr},
-                                        {"index", 0},
-                                        {"delta", json{
-                                            {"role", "assistant"}
-                                        }}}})},
-                            {"created", t},
-                            {"id", completion_id},
-                            {"model", modelname},
-                            {"object", "chat.completion.chunk"}};
-
-                json second_ret = json{
-                            {"choices", json::array({json{{"finish_reason", nullptr},
-                                                            {"index", 0},
-                                                            {"delta", json{
-                                                            {"content", content}}}
-                                                            }})},
-                            {"created", t},
-                            {"id", completion_id},
-                            {"model", modelname},
-                            {"object", "chat.completion.chunk"}};
-
-                return std::vector<json>({initial_ret, second_ret});
-            }
-        } else {
-            // Some idiosyncrasy in task processing logic makes several trailing calls
-            // with empty content, we ignore these at the calee site.
-            if (content.empty()) {
-                return std::vector<json>({json::object()});
-            }
-
-            choices = json::array({json{
-                {"finish_reason", nullptr},
-                {"index", 0},
-                {"delta",
-                json{
-                    {"content", content},
-                }},
-            }});
-        }
-    }
-
-    json ret = json {
-        {"choices", choices},
-        {"created", t},
-        {"id",      completion_id},
-        {"model",   modelname},
-        {"object",  "chat.completion.chunk"}
-    };
-    if (!finish_reason.empty()) {
-        int num_tokens_predicted = json_value(result, "tokens_predicted", 0);
-        int num_prompt_tokens    = json_value(result, "tokens_evaluated", 0);
-        ret.push_back({"usage", json {
-            {"completion_tokens", num_tokens_predicted},
-            {"prompt_tokens",     num_prompt_tokens},
-            {"total_tokens",      num_tokens_predicted + num_prompt_tokens}
-        }});
-    }
-
-    return std::vector<json>({ret});
-}
-
 static json format_embeddings_response_oaicompat(const json & request, const json & embeddings) {
     json data = json::array();
     int i = 0;
@@ -845,42 +650,17 @@ static json format_detokenized_response(const std::string & content) {
     };
 }
 
-static json format_error_response(const std::string & message, const enum error_type type) {
-    std::string type_str;
-    int code = 500;
-    switch (type) {
-        case ERROR_TYPE_INVALID_REQUEST:
-            type_str = "invalid_request_error";
-            code = 400;
-            break;
-        case ERROR_TYPE_AUTHENTICATION:
-            type_str = "authentication_error";
-            code = 401;
-            break;
-        case ERROR_TYPE_NOT_FOUND:
-            type_str = "not_found_error";
-            code = 404;
-            break;
-        case ERROR_TYPE_SERVER:
-            type_str = "server_error";
-            code = 500;
-            break;
-        case ERROR_TYPE_PERMISSION:
-            type_str = "permission_error";
-            code = 403;
-            break;
-        case ERROR_TYPE_NOT_SUPPORTED:
-            type_str = "not_supported_error";
-            code = 501;
-            break;
-        case ERROR_TYPE_UNAVAILABLE:
-            type_str = "unavailable_error";
-            code = 503;
-            break;
+static json format_logit_bias(const std::vector<llama_logit_bias> & logit_bias) {
+    json data = json::array();
+    for (const auto & lb : logit_bias) {
+        data.push_back(json{
+            {"bias", lb.bias},
+            {"token", lb.token},
+        });
     }
-    return json {
-        {"code", code},
-        {"message", message},
-        {"type", type_str},
-    };
+    return data;
+}
+
+static std::string safe_json_to_str(json data) {
+    return data.dump(-1, ' ', false, json::error_handler_t::replace);
 }

examples/server/webui/index.html

@@ -0,0 +1,268 @@
+<!DOCTYPE html>
+<html>
+<head>
+  <meta charset="UTF-8">
+  <meta name="viewport" content="width=device-width, initial-scale=1, maximum-scale=1" />
+  <meta name="color-scheme" content="light dark">
+  <title>🦙 llama.cpp - chat</title>
+</head>
+
+<body>
+  <div id="app" class="opacity-0"> <!-- opacity-0 will be removed on app mounted -->
+    <div class="flex flex-row drawer lg:drawer-open">
+      <input id="toggle-drawer" type="checkbox" class="drawer-toggle" checked />
+
+      <!-- sidebar -->
+      <div class="drawer-side h-screen lg:h-screen z-50 lg:max-w-64">
+        <label for="toggle-drawer" aria-label="close sidebar" class="drawer-overlay"></label>
+        <div class="flex flex-col bg-base-200 min-h-full max-w-[calc(100vw-2em)] py-4 px-4">
+          <div class="flex flex-row items-center justify-between mb-4 mt-4">
+            <h2 class="font-bold ml-4">Conversations</h2>
+
+            <!-- close sidebar button -->
+            <label for="toggle-drawer" class="btn btn-ghost lg:hidden">
+              <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-arrow-bar-left" viewBox="0 0 16 16">
+                <path fill-rule="evenodd" d="M12.5 15a.5.5 0 0 1-.5-.5v-13a.5.5 0 0 1 1 0v13a.5.5 0 0 1-.5.5M10 8a.5.5 0 0 1-.5.5H3.707l2.147 2.146a.5.5 0 0 1-.708.708l-3-3a.5.5 0 0 1 0-.708l3-3a.5.5 0 1 1 .708.708L3.707 7.5H9.5a.5.5 0 0 1 .5.5"/>
+              </svg>
+            </label>
+          </div>
+
+          <!-- list of conversations -->
+          <div :class="{
+            'btn btn-ghost justify-start': true,
+            'btn-active': messages.length === 0,
+          }" @click="newConversation">
+            + New conversation
+          </div>
+          <div v-for="conv in conversations" :class="{
+            'btn btn-ghost justify-start font-normal': true,
+            'btn-active': conv.id === viewingConvId,
+          }" @click="setViewingConv(conv.id)">
+            <span class="truncate">{{ conv.messages[0].content }}</span>
+          </div>
+          <div class="text-center text-xs opacity-40 mt-auto mx-4">
+            Conversations are saved to browser's localStorage
+          </div>
+        </div>
+      </div>
+
+      <!-- main view -->
+      <div class="chat-screen drawer-content grow flex flex-col h-screen w-screen mx-auto px-4">
+        <!-- header -->
+        <div class="flex flex-row items-center mt-6 mb-6">
+          <!-- open sidebar button -->
+          <label for="toggle-drawer" class="btn btn-ghost lg:hidden">
+            <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-list" viewBox="0 0 16 16">
+              <path fill-rule="evenodd" d="M2.5 12a.5.5 0 0 1 .5-.5h10a.5.5 0 0 1 0 1H3a.5.5 0 0 1-.5-.5m0-4a.5.5 0 0 1 .5-.5h10a.5.5 0 0 1 0 1H3a.5.5 0 0 1-.5-.5m0-4a.5.5 0 0 1 .5-.5h10a.5.5 0 0 1 0 1H3a.5.5 0 0 1-.5-.5"/>
+            </svg>
+          </label>
+
+          <div class="grow text-2xl font-bold ml-2">llama.cpp</div>
+
+          <!-- action buttons (top right) -->
+          <div class="flex items-center">
+            <div v-if="messages.length > 0" class="dropdown dropdown-end">
+              <!-- "more" button -->
+              <button tabindex="0" role="button" class="btn m-1" :disabled="isGenerating">
+                <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-three-dots-vertical" viewBox="0 0 16 16">
+                  <path d="M9.5 13a1.5 1.5 0 1 1-3 0 1.5 1.5 0 0 1 3 0m0-5a1.5 1.5 0 1 1-3 0 1.5 1.5 0 0 1 3 0m0-5a1.5 1.5 0 1 1-3 0 1.5 1.5 0 0 1 3 0"/>
+                </svg>
+              </button>
+              <!-- "more" dropdown menu -->
+              <ul tabindex="0" class="dropdown-content menu bg-base-100 rounded-box z-[1] w-52 p-2 shadow">
+                <li @click="downloadConv(viewingConvId)"><a>Download</a></li>
+                <li class="text-error" @click="deleteConv(viewingConvId)"><a>Delete</a></li>
+              </ul>
+            </div>
+            <button class="btn" @click="showConfigDialog = true" :disabled="isGenerating">
+              <!-- settings button -->
+              <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-gear" viewBox="0 0 16 16">
+                <path d="M8 4.754a3.246 3.246 0 1 0 0 6.492 3.246 3.246 0 0 0 0-6.492M5.754 8a2.246 2.246 0 1 1 4.492 0 2.246 2.246 0 0 1-4.492 0"/>
+                <path d="M9.796 1.343c-.527-1.79-3.065-1.79-3.592 0l-.094.319a.873.873 0 0 1-1.255.52l-.292-.16c-1.64-.892-3.433.902-2.54 2.541l.159.292a.873.873 0 0 1-.52 1.255l-.319.094c-1.79.527-1.79 3.065 0 3.592l.319.094a.873.873 0 0 1 .52 1.255l-.16.292c-.892 1.64.901 3.434 2.541 2.54l.292-.159a.873.873 0 0 1 1.255.52l.094.319c.527 1.79 3.065 1.79 3.592 0l.094-.319a.873.873 0 0 1 1.255-.52l.292.16c1.64.893 3.434-.902 2.54-2.541l-.159-.292a.873.873 0 0 1 .52-1.255l.319-.094c1.79-.527 1.79-3.065 0-3.592l-.319-.094a.873.873 0 0 1-.52-1.255l.16-.292c.893-1.64-.902-3.433-2.541-2.54l-.292.159a.873.873 0 0 1-1.255-.52zm-2.633.283c.246-.835 1.428-.835 1.674 0l.094.319a1.873 1.873 0 0 0 2.693 1.115l.291-.16c.764-.415 1.6.42 1.184 1.185l-.159.292a1.873 1.873 0 0 0 1.116 2.692l.318.094c.835.246.835 1.428 0 1.674l-.319.094a1.873 1.873 0 0 0-1.115 2.693l.16.291c.415.764-.42 1.6-1.185 1.184l-.291-.159a1.873 1.873 0 0 0-2.693 1.116l-.094.318c-.246.835-1.428.835-1.674 0l-.094-.319a1.873 1.873 0 0 0-2.692-1.115l-.292.16c-.764.415-1.6-.42-1.184-1.185l.159-.291A1.873 1.873 0 0 0 1.945 8.93l-.319-.094c-.835-.246-.835-1.428 0-1.674l.319-.094A1.873 1.873 0 0 0 3.06 4.377l-.16-.292c-.415-.764.42-1.6 1.185-1.184l.292.159a1.873 1.873 0 0 0 2.692-1.115z"/>
+              </svg>
+            </button>
+
+            <!-- theme controller is copied from https://daisyui.com/components/theme-controller/ -->
+            <div class="dropdown dropdown-end dropdown-bottom">
+              <div tabindex="0" role="button" class="btn m-1">
+                <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-palette2" viewBox="0 0 16 16">
+                  <path d="M0 .5A.5.5 0 0 1 .5 0h5a.5.5 0 0 1 .5.5v5.277l4.147-4.131a.5.5 0 0 1 .707 0l3.535 3.536a.5.5 0 0 1 0 .708L10.261 10H15.5a.5.5 0 0 1 .5.5v5a.5.5 0 0 1-.5.5H3a3 3 0 0 1-2.121-.879A3 3 0 0 1 0 13.044m6-.21 7.328-7.3-2.829-2.828L6 7.188zM4.5 13a1.5 1.5 0 1 0-3 0 1.5 1.5 0 0 0 3 0M15 15v-4H9.258l-4.015 4zM0 .5v12.495zm0 12.495V13z"/>
+                </svg>
+              </div>
+              <ul tabindex="0" class="dropdown-content bg-base-300 rounded-box z-[1] w-52 p-2 shadow-2xl h-80 overflow-y-auto">
+                <li>
+                  <button
+                    class="btn btn-sm btn-block btn-ghost justify-start"
+                    :class="{ 'btn-active': selectedTheme === 'auto' }"
+                    @click="setSelectedTheme('auto')">
+                    auto
+                  </button>
+                </li>
+                <li v-for="theme in themes">
+                  <input
+                    type="radio"
+                    name="theme-dropdown"
+                    class="theme-controller btn btn-sm btn-block btn-ghost justify-start"
+                    :aria-label="theme"
+                    :value="theme"
+                    :checked="selectedTheme === theme"
+                    @click="setSelectedTheme(theme)" />
+                </li>
+              </ul>
+            </div>
+          </div>
+        </div>
+
+        <!-- chat messages -->
+        <div id="messages-list" class="flex flex-col grow overflow-y-auto">
+          <div class="mt-auto flex justify-center">
+            <!-- placeholder to shift the message to the bottom -->
+            {{ messages.length === 0 ? 'Send a message to start' : '' }}
+          </div>
+          <div v-for="msg in messages" class="group">
+            <div :class="{
+              'chat': true,
+              'chat-start': msg.role !== 'user',
+              'chat-end': msg.role === 'user',
+            }">
+              <div :class="{
+                'chat-bubble markdown': true,
+                'chat-bubble-base-300': msg.role !== 'user',
+              }">
+                <!-- textarea for editing message -->
+                <template v-if="editingMsg && editingMsg.id === msg.id">
+                  <textarea
+                    class="textarea textarea-bordered bg-base-100 text-base-content w-[calc(90vw-8em)] lg:w-96"
+                    v-model="msg.content"></textarea>
+                  <br/>
+                  <button class="btn btn-ghost mt-2 mr-2" @click="editingMsg = null">Cancel</button>
+                  <button class="btn mt-2" @click="editUserMsgAndRegenerate(msg)">Submit</button>
+                </template>
+                <!-- render message as markdown -->
+                <vue-markdown v-else :source="msg.content" />
+              </div>
+            </div>
+
+            <!-- actions for each message -->
+            <div :class="{'text-right': msg.role === 'user'}" class="mx-4 mt-2 mb-2">
+              <!-- user message -->
+              <button v-if="msg.role === 'user'" class="badge btn-mini show-on-hover" @click="editingMsg = msg" :disabled="isGenerating">
+                ✍️ Edit
+              </button>
+              <!-- assistant message -->
+              <button v-if="msg.role === 'assistant'" class="badge btn-mini show-on-hover mr-2" @click="regenerateMsg(msg)" :disabled="isGenerating">
+                🔄 Regenerate
+              </button>
+              <button v-if="msg.role === 'assistant'" class="badge btn-mini show-on-hover mr-2" @click="copyMsg(msg)" :disabled="isGenerating">
+                📋 Copy
+              </button>
+            </div>
+          </div>
+
+          <!-- pending (ongoing) assistant message -->
+          <div id="pending-msg" class="chat chat-start">
+            <div v-if="pendingMsg" class="chat-bubble markdown chat-bubble-base-300">
+              <span v-if="!pendingMsg.content" class="loading loading-dots loading-md"></span>
+              <vue-markdown v-else :source="pendingMsg.content" />
+            </div>
+          </div>
+        </div>
+
+        <!-- chat input -->
+        <div class="flex flex-row items-center mt-8 mb-6">
+          <textarea
+            class="textarea textarea-bordered w-full"
+            placeholder="Type a message (Shift+Enter to add a new line)"
+            v-model="inputMsg"
+            @keydown.enter.exact.prevent="sendMessage"
+            @keydown.enter.shift.exact.prevent="inputMsg += '\n'"
+            :disabled="isGenerating"
+            id="msg-input"
+          ></textarea>
+          <button v-if="!isGenerating" class="btn btn-primary ml-2" @click="sendMessage" :disabled="inputMsg.length === 0">Send</button>
+          <button v-else class="btn btn-neutral ml-2" @click="stopGeneration">Stop</button>
+        </div>
+      </div>
+
+    </div>
+
+
+    <!-- modal for editing config -->
+    <dialog class="modal" :class="{'modal-open': showConfigDialog}">
+      <div class="modal-box">
+        <h3 class="text-lg font-bold mb-6">Settings</h3>
+        <div class="h-[calc(90vh-12rem)] overflow-y-auto">
+          <p class="opacity-40 mb-6">Settings below are saved in browser's localStorage</p>
+          <settings-modal-short-input :config-key="'apiKey'" :config-default="configDefault" :config-info="configInfo" v-model="config.apiKey"></settings-modal-short-input>
+          <label class="form-control mb-2">
+            <div class="label">System Message</div>
+            <textarea class="textarea textarea-bordered h-24" :placeholder="'Default: ' + configDefault.systemMessage" v-model="config.systemMessage"></textarea>
+          </label>
+          <template v-for="configKey in ['temperature', 'top_k', 'top_p', 'min_p', 'max_tokens']">
+            <settings-modal-short-input :config-key="configKey" :config-default="configDefault" :config-info="configInfo" v-model="config[configKey]"></settings-modal-short-input>
+          </template>
+          <!-- TODO: add more sampling-related configs, please regroup them into different "collapse" sections -->
+          <!-- Section: Other sampler settings -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Other sampler settings</summary>
+            <div class="collapse-content">
+              <!-- Samplers queue -->
+              <settings-modal-short-input label="Samplers queue" :config-key="'samplers'" :config-default="configDefault" :config-info="configInfo" v-model="config.samplers"></settings-modal-short-input>
+              <!-- Samplers -->
+              <template v-for="configKey in ['dynatemp_range', 'dynatemp_exponent', 'typical_p', 'xtc_probability', 'xtc_threshold']">
+                <settings-modal-short-input :config-key="configKey" :config-default="configDefault" :config-info="configInfo" v-model="config[configKey]"></settings-modal-short-input>
+              </template>
+            </div>
+          </details>
+          <!-- Section: Penalties settings -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Penalties settings</summary>
+            <div class="collapse-content">
+              <template v-for="configKey in ['repeat_last_n', 'repeat_penalty', 'presence_penalty', 'frequency_penalty', 'dry_multiplier', 'dry_base', 'dry_allowed_length', 'dry_penalty_last_n']">
+                <settings-modal-short-input :config-key="configKey" :config-default="configDefault" :config-info="configInfo" v-model="config[configKey]"></settings-modal-short-input>
+              </template>
+            </div>
+          </details>
+          <!-- Section: Advanced config -->
+          <details class="collapse collapse-arrow bg-base-200 mb-2 overflow-visible">
+            <summary class="collapse-title font-bold">Advanced config</summary>
+            <div class="collapse-content">
+              <label class="form-control mb-2">
+                <!-- Custom parameters input -->
+                <div class="label inline">Custom JSON config (For more info, refer to <a class="underline" href="https://github.com/ggerganov/llama.cpp/blob/master/examples/server/README.md" target="_blank" rel="noopener noreferrer">server documentation</a>)</div>
+                <textarea class="textarea textarea-bordered h-24" placeholder="Example: { &quot;mirostat&quot;: 1, &quot;min_p&quot;: 0.1 }" v-model="config.custom"></textarea>
+              </label>
+            </div>
+          </details>
+        </div>
+
+        <!-- action buttons -->
+        <div class="modal-action">
+          <button class="btn" @click="resetConfigDialog">Reset to default</button>
+          <button class="btn" @click="closeAndDiscardConfigDialog">Close</button>
+          <button class="btn btn-primary" @click="closeAndSaveConfigDialog">Save</button>
+        </div>
+      </div>
+    </dialog>
+
+  </div>
+
+  <!-- Template to be used by settings modal -->
+  <template id="settings-modal-short-input">
+    <label class="input input-bordered join-item grow flex items-center gap-2 mb-2">
+      <!-- Show help message on hovering on the input label -->
+      <div class="dropdown dropdown-hover">
+        <div tabindex="0" role="button" class="font-bold">{{ label || configKey }}</div>
+        <div class="dropdown-content menu bg-base-100 rounded-box z-10 w-64 p-2 shadow mt-4">
+          {{ configInfo[configKey] || '(no help message available)' }}
+        </div>
+      </div>
+      <!-- Here we forward v-model from parent to child component, see: https://stackoverflow.com/questions/47311936/v-model-and-child-components -->
+      <input type="text" class="grow" :placeholder="'Default: ' + (configDefault[configKey] || 'none')" :value="modelValue" @input="$emit('update:modelValue', $event.target.value)" />
+    </label>
+  </template>
+
+  <script type="module" src="/src/main.js"></script>
+</body>
+
+</html>

examples/server/webui/package.json

@@ -0,0 +1,23 @@
+{
+  "name": "webui",
+  "private": true,
+  "version": "0.0.0",
+  "type": "module",
+  "scripts": {
+    "dev": "vite",
+    "build": "vite build",
+    "preview": "vite preview"
+  },
+  "devDependencies": {
+    "vite": "^5.4.10"
+  },
+  "dependencies": {
+    "autoprefixer": "^10.4.20",
+    "daisyui": "^4.12.14",
+    "markdown-it": "^14.1.0",
+    "postcss": "^8.4.49",
+    "tailwindcss": "^3.4.15",
+    "vite-plugin-singlefile": "^2.0.3",
+    "vue": "^3.5.13"
+  }
+}

examples/server/webui/postcss.config.js

@@ -0,0 +1,6 @@
+export default {
+  plugins: {
+    tailwindcss: {},
+    autoprefixer: {},
+  },
+}

examples/server/webui/src/main.js

@@ -0,0 +1,456 @@
+import './styles.css';
+import { createApp, defineComponent, shallowRef, computed, h } from 'vue/dist/vue.esm-bundler.js';
+import { llama } from './completion.js';
+import MarkdownIt from 'markdown-it';
+
+// utility functions
+const isString = (x) => !!x.toLowerCase;
+const isNumeric = (n) => !isString(n) && !isNaN(n);
+const escapeAttr = (str) => str.replace(/>/g, '>').replace(/"/g, '"');
+const copyStr = (str) => navigator.clipboard.writeText(str);
+
+// constants
+const BASE_URL = localStorage.getItem('base') // for debugging
+  || (new URL('.', document.baseURI).href).toString(); // for production
+const CONFIG_DEFAULT = {
+  // Note: in order not to introduce breaking changes, please keep the same data type (number, string, etc) if you want to change the default value. Do not use null or undefined for default value.
+  apiKey: '',
+  systemMessage: 'You are a helpful assistant.',
+  // make sure these default values are in sync with `common.h`
+  samplers: 'dkypmxt',
+  temperature: 0.8,
+  dynatemp_range: 0.0,
+  dynatemp_exponent: 1.0,
+  top_k: 40,
+  top_p: 0.95,
+  min_p: 0.05,
+  xtc_probability: 0.0,
+  xtc_threshold: 0.1,
+  typical_p: 1.0,
+  repeat_last_n: 64,
+  repeat_penalty: 1.0,
+  presence_penalty: 0.0,
+  frequency_penalty: 0.0,
+  dry_multiplier: 0.0,
+  dry_base: 1.75,
+  dry_allowed_length: 2,
+  dry_penalty_last_n: -1,
+  max_tokens: -1,
+  custom: '', // custom json-stringified object
+};
+const CONFIG_INFO = {
+  apiKey: 'Set the API Key if you are using --api-key option for the server.',
+  systemMessage: 'The starting message that defines how model should behave.',
+  samplers: 'The order at which samplers are applied, in simplified way. Default is "dkypmxt": dry->top_k->typ_p->top_p->min_p->xtc->temperature',
+  temperature: 'Controls the randomness of the generated text by affecting the probability distribution of the output tokens. Higher = more random, lower = more focused.',
+  dynatemp_range: 'Addon for the temperature sampler. The added value to the range of dynamic temperature, which adjusts probabilities by entropy of tokens.',
+  dynatemp_exponent: 'Addon for the temperature sampler. Smoothes out the probability redistribution based on the most probable token.',
+  top_k: 'Keeps only k top tokens.',
+  top_p: 'Limits tokens to those that together have a cumulative probability of at least p',
+  min_p: 'Limits tokens based on the minimum probability for a token to be considered, relative to the probability of the most likely token.',
+  xtc_probability: 'XTC sampler cuts out top tokens; this parameter controls the chance of cutting tokens at all. 0 disables XTC.',
+  xtc_threshold: 'XTC sampler cuts out top tokens; this parameter controls the token probability that is required to cut that token.',
+  typical_p: 'Sorts and limits tokens based on the difference between log-probability and entropy.',
+  repeat_last_n: 'Last n tokens to consider for penalizing repetition',
+  repeat_penalty: 'Controls the repetition of token sequences in the generated text',
+  presence_penalty: 'Limits tokens based on whether they appear in the output or not.',
+  frequency_penalty: 'Limits tokens based on how often they appear in the output.',
+  dry_multiplier: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets the DRY sampling multiplier.',
+  dry_base: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets the DRY sampling base value.',
+  dry_allowed_length: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets the allowed length for DRY sampling.',
+  dry_penalty_last_n: 'DRY sampling reduces repetition in generated text even across long contexts. This parameter sets DRY penalty for the last n tokens.',
+  max_tokens: 'The maximum number of token per output.',
+  custom: '', // custom json-stringified object
+};
+// config keys having numeric value (i.e. temperature, top_k, top_p, etc)
+const CONFIG_NUMERIC_KEYS = Object.entries(CONFIG_DEFAULT).filter(e => isNumeric(e[1])).map(e => e[0]);
+// list of themes supported by daisyui
+const THEMES = ['light', 'dark', 'cupcake', 'bumblebee', 'emerald', 'corporate', 'synthwave', 'retro', 'cyberpunk', 'valentine', 'halloween', 'garden', 'forest', 'aqua', 'lofi', 'pastel', 'fantasy', 'wireframe', 'black', 'luxury', 'dracula', 'cmyk', 'autumn', 'business', 'acid', 'lemonade', 'night', 'coffee', 'winter', 'dim', 'nord', 'sunset'];
+
+// markdown support
+const VueMarkdown = defineComponent(
+  (props) => {
+    const md = shallowRef(new MarkdownIt({ breaks: true }));
+    const origFenchRenderer = md.value.renderer.rules.fence;
+    md.value.renderer.rules.fence = (tokens, idx, ...args) => {
+      const content = tokens[idx].content;
+      const origRendered = origFenchRenderer(tokens, idx, ...args);
+      return `<div class="relative my-4">
+        <div class="text-right sticky top-4 mb-2 mr-2 h-0">
+          <button class="badge btn-mini" onclick="copyStr(${escapeAttr(JSON.stringify(content))})">📋 Copy</button>
+        </div>
+        ${origRendered}
+      </div>`;
+    };
+    window.copyStr = copyStr;
+    const content = computed(() => md.value.render(props.source));
+    return () => h("div", { innerHTML: content.value });
+  },
+  { props: ["source"] }
+);
+
+// input field to be used by settings modal
+const SettingsModalShortInput = defineComponent({
+  template: document.getElementById('settings-modal-short-input').innerHTML,
+  props: {
+    label: { type: String, required: false },
+    configKey: String,
+    configDefault: Object,
+    configInfo: Object,
+    modelValue: [Object, String, Number],
+  },
+});
+
+// coversations is stored in localStorage
+// format: { [convId]: { id: string, lastModified: number, messages: [...] } }
+// convId is a string prefixed with 'conv-'
+const StorageUtils = {
+  // manage conversations
+  getAllConversations() {
+    const res = [];
+    for (const key in localStorage) {
+      if (key.startsWith('conv-')) {
+        res.push(JSON.parse(localStorage.getItem(key)));
+      }
+    }
+    res.sort((a, b) => b.lastModified - a.lastModified);
+    return res;
+  },
+  // can return null if convId does not exist
+  getOneConversation(convId) {
+    return JSON.parse(localStorage.getItem(convId) || 'null');
+  },
+  // if convId does not exist, create one
+  appendMsg(convId, msg) {
+    if (msg.content === null) return;
+    const conv = StorageUtils.getOneConversation(convId) || {
+      id: convId,
+      lastModified: Date.now(),
+      messages: [],
+    };
+    conv.messages.push(msg);
+    conv.lastModified = Date.now();
+    localStorage.setItem(convId, JSON.stringify(conv));
+  },
+  getNewConvId() {
+    return `conv-${Date.now()}`;
+  },
+  remove(convId) {
+    localStorage.removeItem(convId);
+  },
+  filterAndKeepMsgs(convId, predicate) {
+    const conv = StorageUtils.getOneConversation(convId);
+    if (!conv) return;
+    conv.messages = conv.messages.filter(predicate);
+    conv.lastModified = Date.now();
+    localStorage.setItem(convId, JSON.stringify(conv));
+  },
+  popMsg(convId) {
+    const conv = StorageUtils.getOneConversation(convId);
+    if (!conv) return;
+    const msg = conv.messages.pop();
+    conv.lastModified = Date.now();
+    if (conv.messages.length === 0) {
+      StorageUtils.remove(convId);
+    } else {
+      localStorage.setItem(convId, JSON.stringify(conv));
+    }
+    return msg;
+  },
+
+  // manage config
+  getConfig() {
+    const savedVal = JSON.parse(localStorage.getItem('config') || '{}');
+    // to prevent breaking changes in the future, we always provide default value for missing keys
+    return {
+      ...CONFIG_DEFAULT,
+      ...savedVal,
+    };
+  },
+  setConfig(config) {
+    localStorage.setItem('config', JSON.stringify(config));
+  },
+  getTheme() {
+    return localStorage.getItem('theme') || 'auto';
+  },
+  setTheme(theme) {
+    if (theme === 'auto') {
+      localStorage.removeItem('theme');
+    } else {
+      localStorage.setItem('theme', theme);
+    }
+  },
+};
+
+// scroll to bottom of chat messages
+// if requiresNearBottom is true, only auto-scroll if user is near bottom
+const chatScrollToBottom = (requiresNearBottom) => {
+  const msgListElem = document.getElementById('messages-list');
+  const spaceToBottom = msgListElem.scrollHeight - msgListElem.scrollTop - msgListElem.clientHeight;
+  if (!requiresNearBottom || (spaceToBottom < 100)) {
+    setTimeout(() => msgListElem.scrollTo({ top: msgListElem.scrollHeight }), 1);
+  }
+};
+
+const mainApp = createApp({
+  components: {
+    VueMarkdown,
+    SettingsModalShortInput,
+  },
+  data() {
+    return {
+      conversations: StorageUtils.getAllConversations(),
+      messages: [], // { id: number, role: 'user' | 'assistant', content: string }
+      viewingConvId: StorageUtils.getNewConvId(),
+      inputMsg: '',
+      isGenerating: false,
+      pendingMsg: null, // the on-going message from assistant
+      stopGeneration: () => {},
+      selectedTheme: StorageUtils.getTheme(),
+      config: StorageUtils.getConfig(),
+      showConfigDialog: false,
+      editingMsg: null,
+      // const
+      themes: THEMES,
+      configDefault: {...CONFIG_DEFAULT},
+      configInfo: {...CONFIG_INFO},
+    }
+  },
+  computed: {},
+  mounted() {
+    document.getElementById('app').classList.remove('opacity-0'); // show app
+    // scroll to the bottom when the pending message height is updated
+    const pendingMsgElem = document.getElementById('pending-msg');
+    const resizeObserver = new ResizeObserver(() => {
+      if (this.isGenerating) chatScrollToBottom(true);
+    });
+    resizeObserver.observe(pendingMsgElem);
+  },
+  methods: {
+    hideSidebar() {
+      document.getElementById('toggle-drawer').checked = false;
+    },
+    setSelectedTheme(theme) {
+      this.selectedTheme = theme;
+      StorageUtils.setTheme(theme);
+    },
+    newConversation() {
+      if (this.isGenerating) return;
+      this.viewingConvId = StorageUtils.getNewConvId();
+      this.editingMsg = null;
+      this.fetchMessages();
+      chatScrollToBottom();
+      this.hideSidebar();
+    },
+    setViewingConv(convId) {
+      if (this.isGenerating) return;
+      this.viewingConvId = convId;
+      this.editingMsg = null;
+      this.fetchMessages();
+      chatScrollToBottom();
+      this.hideSidebar();
+    },
+    deleteConv(convId) {
+      if (this.isGenerating) return;
+      if (window.confirm('Are you sure to delete this conversation?')) {
+        StorageUtils.remove(convId);
+        if (this.viewingConvId === convId) {
+          this.viewingConvId = StorageUtils.getNewConvId();
+          this.editingMsg = null;
+        }
+        this.fetchConversation();
+        this.fetchMessages();
+      }
+    },
+    downloadConv(convId) {
+      const conversation = StorageUtils.getOneConversation(convId);
+      if (!conversation) {
+        alert('Conversation not found.');
+        return;
+      }
+      const conversationJson = JSON.stringify(conversation, null, 2);
+      const blob = new Blob([conversationJson], { type: 'application/json' });
+      const url = URL.createObjectURL(blob);
+      const a = document.createElement('a');
+      a.href = url;
+      a.download = `conversation_${convId}.json`;
+      document.body.appendChild(a);
+      a.click();
+      document.body.removeChild(a);
+      URL.revokeObjectURL(url);
+    },
+    async sendMessage() {
+      if (!this.inputMsg) return;
+      const currConvId = this.viewingConvId;
+
+      StorageUtils.appendMsg(currConvId, {
+        id: Date.now(),
+        role: 'user',
+        content: this.inputMsg,
+      });
+      this.fetchConversation();
+      this.fetchMessages();
+      this.inputMsg = '';
+      this.editingMsg = null;
+      this.generateMessage(currConvId);
+      chatScrollToBottom();
+    },
+    async generateMessage(currConvId) {
+      if (this.isGenerating) return;
+      this.pendingMsg = { id: Date.now()+1, role: 'assistant', content: null };
+      this.isGenerating = true;
+      this.editingMsg = null;
+
+      try {
+        const abortController = new AbortController();
+        this.stopGeneration = () => abortController.abort();
+        const params = {
+          messages: [
+            { role: 'system', content: this.config.systemMessage },
+            ...this.messages,
+          ],
+          stream: true,
+          cache_prompt: true,
+          samplers: this.config.samplers,
+          temperature: this.config.temperature,
+          dynatemp_range: this.config.dynatemp_range,
+          dynatemp_exponent: this.config.dynatemp_exponent,
+          top_k: this.config.top_k,
+          top_p: this.config.top_p,
+          min_p: this.config.min_p,
+          typical_p: this.config.typical_p,
+          xtc_probability: this.config.xtc_probability,
+          xtc_threshold: this.config.xtc_threshold,
+          repeat_last_n: this.config.repeat_last_n,
+          repeat_penalty: this.config.repeat_penalty,
+          presence_penalty: this.config.presence_penalty,
+          frequency_penalty: this.config.frequency_penalty,
+          dry_multiplier: this.config.dry_multiplier,
+          dry_base: this.config.dry_base,
+          dry_allowed_length: this.config.dry_allowed_length,
+          dry_penalty_last_n: this.config.dry_penalty_last_n,
+          max_tokens: this.config.max_tokens,
+          ...(this.config.custom.length ? JSON.parse(this.config.custom) : {}),
+          ...(this.config.apiKey ? { api_key: this.config.apiKey } : {}),
+        };
+        const config = {
+          controller: abortController,
+          api_url: BASE_URL,
+          endpoint: '/chat/completions',
+        };
+        for await (const chunk of llama(prompt, params, config)) {
+          const stop = chunk.data.stop;
+          const addedContent = chunk.data.choices[0].delta.content;
+          const lastContent = this.pendingMsg.content || '';
+          if (addedContent) {
+            this.pendingMsg = {
+              id: this.pendingMsg.id,
+              role: 'assistant',
+              content: lastContent + addedContent,
+            };
+          }
+        }
+
+        StorageUtils.appendMsg(currConvId, this.pendingMsg);
+        this.fetchConversation();
+        this.fetchMessages();
+        setTimeout(() => document.getElementById('msg-input').focus(), 1);
+      } catch (error) {
+        if (error.name === 'AbortError') {
+          // user stopped the generation via stopGeneration() function
+          StorageUtils.appendMsg(currConvId, this.pendingMsg);
+          this.fetchConversation();
+          this.fetchMessages();
+        } else {
+          console.error(error);
+          alert(error);
+          // pop last user message
+          const lastUserMsg = StorageUtils.popMsg(currConvId);
+          this.inputMsg = lastUserMsg ? lastUserMsg.content : '';
+        }
+      }
+
+      this.pendingMsg = null;
+      this.isGenerating = false;
+      this.stopGeneration = () => {};
+      this.fetchMessages();
+      chatScrollToBottom();
+    },
+
+    // message actions
+    regenerateMsg(msg) {
+      if (this.isGenerating) return;
+      // TODO: somehow keep old history (like how ChatGPT has different "tree"). This can be done by adding "sub-conversations" with "subconv-" prefix, and new message will have a list of subconvIds
+      const currConvId = this.viewingConvId;
+      StorageUtils.filterAndKeepMsgs(currConvId, (m) => m.id < msg.id);
+      this.fetchConversation();
+      this.fetchMessages();
+      this.generateMessage(currConvId);
+    },
+    copyMsg(msg) {
+      copyStr(msg.content);
+    },
+    editUserMsgAndRegenerate(msg) {
+      if (this.isGenerating) return;
+      const currConvId = this.viewingConvId;
+      const newContent = msg.content;
+      this.editingMsg = null;
+      StorageUtils.filterAndKeepMsgs(currConvId, (m) => m.id < msg.id);
+      StorageUtils.appendMsg(currConvId, {
+        id: Date.now(),
+        role: 'user',
+        content: newContent,
+      });
+      this.fetchConversation();
+      this.fetchMessages();
+      this.generateMessage(currConvId);
+    },
+
+    // settings dialog methods
+    closeAndSaveConfigDialog() {
+      try {
+        if (this.config.custom.length) JSON.parse(this.config.custom);
+      } catch (error) {
+        alert('Invalid JSON for custom config. Please either fix it or leave it empty.');
+        return;
+      }
+      for (const key of CONFIG_NUMERIC_KEYS) {
+        if (isNaN(this.config[key]) || this.config[key].toString().trim().length === 0) {
+          alert(`Invalid number for ${key} (expected an integer or a float)`);
+          return;
+        }
+        this.config[key] = parseFloat(this.config[key]);
+      }
+      this.showConfigDialog = false;
+      StorageUtils.setConfig(this.config);
+    },
+    closeAndDiscardConfigDialog() {
+      this.showConfigDialog = false;
+      this.config = StorageUtils.getConfig();
+    },
+    resetConfigDialog() {
+      if (window.confirm('Are you sure to reset all settings?')) {
+        this.config = {...CONFIG_DEFAULT};
+      }
+    },
+
+    // sync state functions
+    fetchConversation() {
+      this.conversations = StorageUtils.getAllConversations();
+    },
+    fetchMessages() {
+      this.messages = StorageUtils.getOneConversation(this.viewingConvId)?.messages ?? [];
+    },
+  },
+});
+mainApp.config.errorHandler = alert;
+try {
+  mainApp.mount('#app');
+} catch (err) {
+  console.error(err);
+  document.getElementById('app').innerHTML = `<div style="margin:2em auto">
+    Failed to start app. Please try clearing localStorage and try again.<br/>
+    <br/>
+    <button class="btn" onClick="localStorage.clear(); window.location.reload();">Clear localStorage</button>
+  </div>`;
+}

examples/server/webui/src/styles.css

@@ -0,0 +1,26 @@
+@tailwind base;
+@tailwind components;
+@tailwind utilities;
+
+.markdown {
+  h1, h2, h3, h4, h5, h6, ul, ol, li { all: revert; }
+  pre {
+    @apply whitespace-pre-wrap rounded-lg p-2;
+    border: 1px solid currentColor;
+  }
+  /* TODO: fix markdown table */
+}
+
+.show-on-hover {
+  @apply md:opacity-0 md:group-hover:opacity-100;
+}
+.btn-mini {
+  @apply cursor-pointer hover:shadow-md;
+}
+.chat-screen { max-width: 900px; }
+
+.chat-bubble-base-300 {
+  --tw-bg-opacity: 1;
+  --tw-text-opacity: 1;
+  @apply bg-base-300 text-base-content;
+}

examples/server/webui/tailwind.config.js

@@ -0,0 +1,16 @@
+/** @type {import('tailwindcss').Config} */
+export default {
+  content: [
+    "./index.html",
+    "./src/**/*.{js,ts,jsx,tsx}",
+  ],
+  theme: {
+    extend: {},
+  },
+  plugins: [
+    require('daisyui'),
+  ],
+  daisyui: {
+    themes: ['light', 'dark', 'cupcake', 'bumblebee', 'emerald', 'corporate', 'synthwave', 'retro', 'cyberpunk', 'valentine', 'halloween', 'garden', 'forest', 'aqua', 'lofi', 'pastel', 'fantasy', 'wireframe', 'black', 'luxury', 'dracula', 'cmyk', 'autumn', 'business', 'acid', 'lemonade', 'night', 'coffee', 'winter', 'dim', 'nord', 'sunset'],
+  }
+}

examples/server/webui/vite.config.js

@@ -0,0 +1,36 @@
+
+import { viteSingleFile } from 'vite-plugin-singlefile';
+import path from 'path';
+import fs from 'fs';
+
+const GUIDE_FOR_FRONTEND = `
+<!--
+  This is a single file build of the frontend.
+  It is automatically generated by the build process.
+  Do not edit this file directly.
+  To make changes, refer to the "Web UI" section in the README.
+-->
+`.trim();
+
+export default {
+  plugins: [
+    viteSingleFile(),
+    (function llamaCppPlugin() {
+      let config;
+      return {
+        name: 'llamacpp:build',
+        apply: 'build',
+        async configResolved(_config) {
+          config = _config;
+        },
+        writeBundle() {
+          const outputIndexHtml = path.join(config.build.outDir, 'index.html');
+          const content = fs.readFileSync(outputIndexHtml, 'utf-8');
+
+          const targetOutputFile = path.join(config.build.outDir, '../../public/index.html');
+          fs.writeFileSync(targetOutputFile, GUIDE_FOR_FRONTEND + '\n' + content);
+        }
+      }
+    })(),
+  ],
+};

examples/simple-chat/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-simple-chat)
 add_executable(${TARGET} simple-chat.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/simple/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-simple)
 add_executable(${TARGET} simple.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/speculative-simple/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-speculative-simple)
 add_executable(${TARGET} speculative-simple.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/speculative/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-speculative)
 add_executable(${TARGET} speculative.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/tokenize/CMakeLists.txt

@@ -2,4 +2,4 @@ set(TARGET llama-tokenize)
 add_executable(${TARGET} tokenize.cpp)
 install(TARGETS ${TARGET} RUNTIME)
 target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
-target_compile_features(${TARGET} PRIVATE cxx_std_11)
+target_compile_features(${TARGET} PRIVATE cxx_std_17)

ggml/CMakeLists.txt

@@ -92,29 +92,33 @@ else()
     set(INS_ENB ON)
 endif()
 
-option(GGML_CPU_HBM     "ggml: use memkind for CPU HBM" OFF)
-option(GGML_CPU_AARCH64 "ggml: use runtime weight conversion of Q4_0 to Q4_X_X" ON)
-
-option(GGML_AVX         "ggml: enable AVX"              ${INS_ENB})
-option(GGML_AVX2        "ggml: enable AVX2"             ${INS_ENB})
-option(GGML_AVX512      "ggml: enable AVX512"           OFF)
-option(GGML_AVX512_VBMI "ggml: enable AVX512-VBMI"      OFF)
-option(GGML_AVX512_VNNI "ggml: enable AVX512-VNNI"      OFF)
-option(GGML_AVX512_BF16 "ggml: enable AVX512-BF16"      OFF)
-option(GGML_AMX_TILE    "ggml: enable AMX-TILE"         OFF)
-option(GGML_AMX_INT8    "ggml: enable AMX-INT8"         OFF)
-option(GGML_AMX_BF16    "ggml: enable AMX-BF16"         OFF)
-option(GGML_FMA         "ggml: enable FMA"              ${INS_ENB})
+option(GGML_CPU_HBM          "ggml: use memkind for CPU HBM" OFF)
+option(GGML_CPU_AARCH64      "ggml: use runtime weight conversion of Q4_0 to Q4_X_X" ON)
+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_AVX512           "ggml: enable AVX512F"          OFF)
+option(GGML_AVX512_VBMI      "ggml: enable AVX512-VBMI"      OFF)
+option(GGML_AVX512_VNNI      "ggml: enable AVX512-VNNI"      OFF)
+option(GGML_AVX512_BF16      "ggml: enable AVX512-BF16"      OFF)
 if (NOT MSVC)
-    option(GGML_F16C    "ggml: enable F16C"             ${INS_ENB}) # in MSVC F16C is implied with AVX2/AVX512
+    # in MSVC F16C and FMA is implied with AVX2/AVX512
+    option(GGML_FMA          "ggml: enable FMA"              ${INS_ENB})
+    option(GGML_F16C         "ggml: enable F16C"             ${INS_ENB})
+    # MSVC does not seem to support AMX
+    option(GGML_AMX_TILE     "ggml: enable AMX-TILE"         OFF)
+    option(GGML_AMX_INT8     "ggml: enable AMX-INT8"         OFF)
+    option(GGML_AMX_BF16     "ggml: enable AMX-BF16"         OFF)
 endif()
-option(GGML_LASX        "ggml: enable lasx"             ON)
-option(GGML_LSX         "ggml: enable lsx"              ON)
-option(GGML_RVV         "ggml: enable rvv"              ON)
-option(GGML_SVE         "ggml: enable SVE"              OFF)
+option(GGML_LASX             "ggml: enable lasx"             ON)
+option(GGML_LSX              "ggml: enable lsx"              ON)
+option(GGML_RVV              "ggml: enable rvv"              ON)
+option(GGML_SVE              "ggml: enable SVE"              OFF)
+option(GGML_CPU_ALL_VARIANTS "ggml: build all variants of the CPU backend (requires GGML_BACKEND_DL)" OFF)
+
 
 if (WIN32)
-    set(GGML_WIN_VER "0x602" CACHE STRING "ggml: Windows Version")
+    set(GGML_WIN_VER "0x602" CACHE STRING   "ggml: Windows version")
 endif()
 
 # ggml core
@@ -161,7 +165,6 @@ set   (GGML_METAL_MACOSX_VERSION_MIN "" CACHE STRING
 set   (GGML_METAL_STD "" CACHE STRING       "ggml: metal standard version (-std flag)")
 option(GGML_OPENMP                          "ggml: use OpenMP"                                ON)
 option(GGML_RPC                             "ggml: use RPC"                                   OFF)
-option(GGML_AMX                             "ggml: use AMX"                                   OFF)
 option(GGML_SYCL                            "ggml: use SYCL"                                  OFF)
 option(GGML_SYCL_F16                        "ggml: use 16 bit floats for sycl calculations"   OFF)
 set   (GGML_SYCL_TARGET "INTEL" CACHE STRING
@@ -180,11 +183,7 @@ option(GGML_BUILD_EXAMPLES "ggml: build examples" ${GGML_STANDALONE})
 set(CMAKE_C_STANDARD 11)
 set(CMAKE_C_STANDARD_REQUIRED true)
 
-if (GGML_SYCL)
-    set(CMAKE_CXX_STANDARD 17)
-else()
-    set(CMAKE_CXX_STANDARD 11)
-endif()
+set(CMAKE_CXX_STANDARD 17)
 set(CMAKE_CXX_STANDARD_REQUIRED true)
 
 set(THREADS_PREFER_PTHREAD_FLAG ON)

ggml/include/ggml-amx.h

@@ -1,25 +0,0 @@
-#pragma once
-
-#include "ggml.h"
-#include "ggml-backend.h"
-
-
-#ifdef  __cplusplus
-extern "C" {
-#endif
-
-// buffer_type API
-GGML_BACKEND_API ggml_backend_buffer_type_t ggml_backend_amx_buffer_type(void);
-
-GGML_BACKEND_API bool ggml_backend_is_amx(ggml_backend_t backend);
-
-// backend API
-GGML_BACKEND_API ggml_backend_t ggml_backend_amx_init(void);
-
-GGML_BACKEND_API void ggml_backend_amx_set_n_threads(ggml_backend_t backend_amx, int n_threads);
-
-GGML_BACKEND_API ggml_backend_reg_t ggml_backend_amx_reg(void);
-
-#ifdef  __cplusplus
-}
-#endif