mostly style fixes; fix KQ_mask comment

* Prevent control characters from being served in json string

* Prevent control characters from being served in json string (array)

.gitignore

@@ -45,6 +45,7 @@ models-mnt
 /embedding
 /gguf
 /gguf-llama-simple
+/gritlm
 /imatrix
 /infill
 /libllama.so

Makefile

@@ -2,7 +2,7 @@
 BUILD_TARGETS = \
 	main quantize quantize-stats perplexity imatrix embedding vdot q8dot train-text-from-scratch convert-llama2c-to-ggml \
 	simple batched batched-bench save-load-state server gguf llama-bench libllava.a llava-cli baby-llama beam-search  \
-	speculative infill tokenize benchmark-matmult parallel finetune export-lora lookahead lookup passkey tests/test-c.o
+	speculative infill tokenize benchmark-matmult parallel finetune export-lora lookahead lookup passkey gritlm tests/test-c.o
 
 # Binaries only useful for tests
 TEST_TARGETS = \
@@ -720,6 +720,10 @@ embedding: examples/embedding/embedding.cpp                   ggml.o llama.o $(C
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
 
+gritlm: examples/gritlm/gritlm.cpp                         ggml.o llama.o $(COMMON_DEPS) $(OBJS)
+	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
+	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)
+
 save-load-state: examples/save-load-state/save-load-state.cpp ggml.o llama.o $(COMMON_DEPS) $(OBJS)
 	$(CXX) $(CXXFLAGS) -c $< -o $(call GET_OBJ_FILE, $<)
 	$(CXX) $(CXXFLAGS) $(filter-out %.h $<,$^) $(call GET_OBJ_FILE, $<) -o $@ $(LDFLAGS)

examples/CMakeLists.txt

@@ -20,6 +20,7 @@ else()
     add_subdirectory(convert-llama2c-to-ggml)
     add_subdirectory(embedding)
     add_subdirectory(finetune)
+    add_subdirectory(gritlm)
     add_subdirectory(infill)
     add_subdirectory(llama-bench)
     add_subdirectory(llava)

examples/gritlm/CMakeLists.txt

@@ -0,0 +1,5 @@
+set(TARGET 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)

examples/gritlm/gritlm.cpp

@@ -0,0 +1,243 @@
+#include "common.h"
+#include "llama.h"
+
+#include <string>
+#include <vector>
+
+// #define GRIT_DEBUG
+
+static float dot_product(const std::vector<float> & v1, const std::vector<float> & v2) {
+    float dot = 0.0f;
+    for (uint64_t i = 0; i < v1.size(); ++i) {
+        dot += v1[i] * v2[i];
+    }
+    return dot;
+}
+
+static float norm(const std::vector<float> & v) {
+    return std::sqrt(dot_product(v, v));
+}
+
+static float cosine_similarity(const std::vector<float> & v1, const std::vector<float> & v2) {
+    return dot_product(v1, v2) / (norm(v1) * norm(v2));
+}
+
+static void normalize(const std::vector<float> & in, float * out) {
+    float inorm = norm(in);
+    for (uint64_t i = 0; i < in.size(); i++) {
+        out[i] = in[i] / inorm;
+    }
+}
+
+static std::vector<std::vector<float>> encode(llama_context * ctx, const std::vector<std::string> & sentences, const std::string & instruction) {
+    auto result = std::vector<std::vector<float>>{};
+
+    auto mdl = llama_get_model(ctx);
+    auto batch = llama_batch_init(llama_n_batch(ctx), 0, 1);
+
+    for (uint64_t i = 0; i < sentences.size(); i++) {
+        llama_batch_clear(batch);
+
+        std::string input_string = instruction + sentences[i];
+        std::vector<llama_token> inputs = llama_tokenize(mdl, input_string, true, false);
+        auto n_toks = (int32_t)inputs.size();
+
+        // GritLM seems to have embed EOS = ""
+        // https://github.com/ContextualAI/gritlm/blob/92025b16534712b31b3c4aaaf069350e222bd5f8/gritlm/gritlm.py#L18
+        // inputs.push_back(llama_token_eos(mdl));
+
+        // we want to ignore instruction tokens for mean pooling
+        std::vector<llama_token> inputs_instruct = llama_tokenize(mdl, instruction, true, false);
+        auto n_inst = (int32_t)inputs_instruct.size();
+
+#ifdef GRIT_DEBUG
+        // debug tokens - should be matching as referenced in the GritLM sample
+        std::for_each(inputs.begin(), inputs.end(), [&ctx](llama_token t) {
+            std::printf("[%u:%s]", t, llama_token_to_piece(ctx, t).c_str());
+        });
+        std::printf("\n");
+#endif
+
+        // add input to batch (this increments n_tokens)
+        for (int32_t j = 0; j < n_toks; j++) {
+            llama_batch_add(batch, inputs[j], j, { 0 }, j >= n_inst);
+        }
+
+        // clear previous kv_cache values (irrelevant for embeddings)
+        llama_kv_cache_clear(ctx);
+
+        // run model
+        llama_decode(ctx, batch);
+
+        // get embedding dimensions
+        uint64_t n_embd = llama_n_embd(mdl);
+
+        // allocate embedding output
+        std::vector<float> emb_unorm(n_embd, 0.0f);
+
+        // sum up all token embeddings
+        for (int32_t k = n_inst; k < n_toks; k++) {
+            float * emb = llama_get_embeddings_ith(ctx, k);
+            for (uint64_t j = 0; j < n_embd; j++) {
+                emb_unorm[j] += emb[j];
+            }
+        }
+
+        // divide by number of tokens (mean pooling)
+        uint64_t n_sent = n_toks - n_inst;
+        for (uint64_t j = 0; j < n_embd; j++) {
+            emb_unorm[j] /= n_sent;
+        }
+
+        auto emb_norm = std::vector<float>(emb_unorm.size());
+        normalize(emb_unorm, emb_norm.data());
+        result.push_back(emb_norm);
+
+#ifdef GRIT_DEBUG
+        // print out emb_norm
+        std::printf("embedding %ld: ", i);
+        for (uint64_t j = 0; j < n_embd; j++) {
+            std::printf("%.5f ", emb_norm[j]);
+        }
+        std::printf("\n\n");
+#endif
+    }
+
+    llama_batch_free(batch);
+    return result;
+}
+
+static std::string aggregate_pieces(const std::vector<std::string> & pieces) {
+    // calculate total length required
+    size_t length = 0;
+    for (const auto & str : pieces) {
+        length += str.size();
+    }
+
+    // reserve memory
+    std::string result;
+    result.reserve(length);
+
+    // append pieces
+    for (const auto & str : pieces) {
+        result += str;
+    }
+
+    return result;
+}
+
+static std::string generate(llama_context * ctx, const std::string & prompt, bool stream) {
+    std::vector<std::string> pieces;
+
+    const llama_model * mdl = llama_get_model(ctx);
+    llama_token eos_token = llama_token_eos(mdl);
+    llama_batch bat = llama_batch_init(llama_n_batch(ctx), 0, 1);
+
+    std::vector<llama_token> inputs = llama_tokenize(mdl, prompt, false, true);
+    int32_t i_current_token = 0;
+
+    while (true) {
+        llama_batch_clear(bat);
+        for (auto i = 0; i < inputs.size(); i++) {
+            llama_batch_add(bat, inputs[i], i_current_token++, { 0 }, i == inputs.size() - 1);
+        }
+        inputs.clear();
+
+        llama_decode(ctx, bat);
+        auto logits = llama_get_logits_ith(ctx, bat.n_tokens - 1);
+
+        auto candidates = std::vector<llama_token_data>(llama_n_vocab(mdl));
+        for (auto token = 0; token < candidates.size(); token++) {
+            candidates[token] = llama_token_data{ token, logits[token], 0.0f };
+        }
+        auto candidates_p = llama_token_data_array{ candidates.data(), candidates.size(), false };
+
+        llama_token token = llama_sample_token_greedy(ctx, &candidates_p);
+        if (token == eos_token) {
+            break;
+        }
+
+        std::string piece = llama_token_to_piece(ctx, token);
+        if (stream) {
+            std::printf("%s", piece.c_str());
+        }
+
+        pieces.push_back(piece);
+        inputs.push_back(token);
+    }
+
+    llama_batch_free(bat);
+
+    return aggregate_pieces(pieces);
+}
+
+static std::string gritlm_instruction(const std::string & instruction) {
+    return !instruction.empty() ? "<|user|>\n" + instruction + "\n<|embed|>\n" : "<|embed|>\n";
+}
+
+int main(int argc, char * argv[])
+{
+    gpt_params params;
+    if (!gpt_params_parse(argc, argv, params)) {
+        return 1;
+    }
+
+    llama_model_params mparams = llama_model_params_from_gpt_params(params);
+    llama_context_params cparams = llama_context_params_from_gpt_params(params);
+
+    llama_backend_init();
+
+    llama_model * mdl = llama_load_model_from_file(params.model.c_str(), mparams);
+
+    // create new context - set to embedding mode
+    llama_context * embd_ctx = llama_new_context_with_model(mdl, cparams);
+    llama_set_embeddings(embd_ctx, true);
+
+    // create new context - default mode is causal
+    llama_context * causal_ctx = llama_new_context_with_model(mdl, cparams);
+
+    // samples taken from here: https://github.com/ContextualAI/gritlm#basic
+    // Embedding/Representation
+    {
+        std::string instruction = "Given a scientific paper title, retrieve the paper's abstract";
+
+        std::vector<std::string> queries = {
+            "Bitcoin: A Peer-to-Peer Electronic Cash System",
+            "Generative Representational Instruction Tuning",
+        };
+
+        std::vector<std::string> documents = {
+            "A purely peer-to-peer version of electronic cash would allow online payments to be sent directly from one party to another without going through a financial institution. Digital signatures provide part of the solution, but the main benefits are lost if a trusted third party is still required to prevent double-spending. We propose a solution to the double-spending problem using a peer-to-peer network. The network timestamps transactions by hashing them into an ongoing chain of hash-based proof-of-work, forming a record that cannot be changed without redoing the proof-of-work. The longest chain not only serves as proof of the sequence of events witnessed, but proof that it came from the largest pool of CPU power. As long as a majority of CPU power is controlled by nodes that are not cooperating to attack the network, they'll generate the longest chain and outpace attackers. The network itself requires minimal structure. Messages are broadcast on a best effort basis, and nodes can leave and rejoin the network at will, accepting the longest proof-of-work chain as proof of what happened while they were gone.",
+            "All text-based language problems can be reduced to either generation or embedding. Current models only perform well at one or the other. We introduce generative representational instruction tuning (GRIT) whereby a large language model is trained to handle both generative and embedding tasks by distinguishing between them through instructions. Compared to other open models, our resulting GritLM 7B sets a new state of the art on the Massive Text Embedding Benchmark (MTEB) and outperforms all models up to its size on a range of generative tasks. By scaling up further, GritLM 8X7B outperforms all open generative language models that we tried while still being among the best embedding models. Notably, we find that GRIT matches training on only generative or embedding data, thus we can unify both at no performance loss. Among other benefits, the unification via GRIT speeds up Retrieval-Augmented Generation (RAG) by > 60% for long documents, by no longer requiring separate retrieval and generation models. Models, code, etc. are freely available at https://github.com/ContextualAI/gritlm.",
+        };
+
+        // No need to add instruction for retrieval documents
+        std::vector<std::vector<float>> d_rep = encode(embd_ctx, documents, gritlm_instruction(""));
+        std::vector<std::vector<float>> q_rep = encode(embd_ctx, queries, gritlm_instruction(instruction));
+
+        float cosine_sim_q0_d0 = cosine_similarity(q_rep[0], d_rep[0]);
+        float cosine_sim_q0_d1 = cosine_similarity(q_rep[0], d_rep[1]);
+        float cosine_sim_q1_d0 = cosine_similarity(q_rep[1], d_rep[0]);
+        float cosine_sim_q1_d1 = cosine_similarity(q_rep[1], d_rep[1]);
+
+        std::printf("Cosine similarity between \"%.50s\" and \"%.50s\" is: %.3f\n", queries[0].c_str(), documents[0].c_str(), cosine_sim_q0_d0);
+        std::printf("Cosine similarity between \"%.50s\" and \"%.50s\" is: %.3f\n", queries[0].c_str(), documents[1].c_str(), cosine_sim_q0_d1);
+        std::printf("Cosine similarity between \"%.50s\" and \"%.50s\" is: %.3f\n", queries[1].c_str(), documents[0].c_str(), cosine_sim_q1_d0);
+        std::printf("Cosine similarity between \"%.50s\" and \"%.50s\" is: %.3f\n", queries[1].c_str(), documents[1].c_str(), cosine_sim_q1_d1);
+    }
+
+    // Generation
+    // GritLM models are not finetuned with system prompts, as you can just include system-like instructions together with your user instruction
+    {
+        const std::string prompt = "<|user|>\nPlease write me a poem about my recent hike of Mt. Fuji at midnight in the style of Shakespeare.\n<|assistant|>\n";
+        std::string response = generate(causal_ctx, prompt, true);
+    }
+
+    llama_free(embd_ctx);
+    llama_free(causal_ctx);
+
+    llama_free_model(mdl);
+    llama_backend_free();
+
+    return 0;
+}

ggml-quants.c

@@ -51,6 +51,7 @@
 
 #define UNUSED GGML_UNUSED
 
+// some compilers don't provide _mm256_set_m128i, e.g. gcc 7
 #define MM256_SET_M128I(a, b) _mm256_insertf128_si256(_mm256_castsi128_si256(b), (a), 1)
 
 #if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__)
@@ -9563,7 +9564,7 @@ void ggml_vec_dot_iq2_xs_q8_K(int n, float * restrict s, size_t bs, const void *
 
         const __m128i odd_bits = _mm_shuffle_epi8(bit_helper, partial_sign_bits_for_counting);
         const __m128i full_sign_bits = _mm_or_si128(partial_sign_bits, odd_bits);
-        const __m256i full_signs = _mm256_set_m128i(full_sign_bits, full_sign_bits);
+        const __m256i full_signs = MM256_SET_M128I(full_sign_bits, full_sign_bits);
 
         const __m256i q8_1 = _mm256_loadu_si256((const __m256i *)y[i].qs);
         const __m256i q8_2 = _mm256_loadu_si256((const __m256i *)(y[i].qs+32));
@@ -9585,8 +9586,8 @@ void ggml_vec_dot_iq2_xs_q8_K(int n, float * restrict s, size_t bs, const void *
         const __m256i dot1  = _mm256_maddubs_epi16(q2_1, q8s_1);
         const __m256i dot2  = _mm256_maddubs_epi16(q2_2, q8s_2);
 
-        const __m256i sc1 = _mm256_set_m128i(_mm_set1_epi16(2*(x[i].scales[0] >> 4)+1), _mm_set1_epi16(2*(x[i].scales[0] & 0xf)+1));
-        const __m256i sc2 = _mm256_set_m128i(_mm_set1_epi16(2*(x[i].scales[1] >> 4)+1), _mm_set1_epi16(2*(x[i].scales[1] & 0xf)+1));
+        const __m256i sc1 = MM256_SET_M128I(_mm_set1_epi16(2*(x[i].scales[0] >> 4)+1), _mm_set1_epi16(2*(x[i].scales[0] & 0xf)+1));
+        const __m256i sc2 = MM256_SET_M128I(_mm_set1_epi16(2*(x[i].scales[1] >> 4)+1), _mm_set1_epi16(2*(x[i].scales[1] & 0xf)+1));
 
         const __m256i sum = _mm256_add_epi32(_mm256_madd_epi16(sc1, dot1), _mm256_madd_epi16(sc2, dot2));
 
@@ -9653,8 +9654,8 @@ void ggml_vec_dot_iq2_xs_q8_K(int n, float * restrict s, size_t bs, const void *
 
             const __m128i full_signs_l = _mm256_castsi256_si128(full_sign_bits);
             const __m128i full_signs_h = _mm256_extractf128_si256(full_sign_bits, 1);
-            const __m256i full_signs_1 = _mm256_set_m128i(full_signs_l, full_signs_l);
-            const __m256i full_signs_2 = _mm256_set_m128i(full_signs_h, full_signs_h);
+            const __m256i full_signs_1 = MM256_SET_M128I(full_signs_l, full_signs_l);
+            const __m256i full_signs_2 = MM256_SET_M128I(full_signs_h, full_signs_h);
 
             __m256i signs;
             signs = _mm256_shuffle_epi8(full_signs_1, block_sign_shuffle_1);
@@ -10551,10 +10552,10 @@ void ggml_vec_dot_iq4_nl_q8_0(int n, float * restrict s, size_t bs, const void *
         const __m128i q4bits_2 = _mm_loadu_si128((const __m128i*)x[1].qs);
         const __m256i q8b_1 = _mm256_loadu_si256((const __m256i *)y[0].qs);
         const __m256i q8b_2 = _mm256_loadu_si256((const __m256i *)y[1].qs);
-        const __m256i q4b_1 = _mm256_set_m128i(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_1, 4), m4b)),
-                                               _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_1, m4b)));
-        const __m256i q4b_2 = _mm256_set_m128i(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_2, 4), m4b)),
-                                               _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_2, m4b)));
+        const __m256i q4b_1 = MM256_SET_M128I(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_1, 4), m4b)),
+                                              _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_1, m4b)));
+        const __m256i q4b_2 = MM256_SET_M128I(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_2, 4), m4b)),
+                                              _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_2, m4b)));
         const __m256i p16_1 = mul_add_epi8(q4b_1, q8b_1);
         const __m256i p16_2 = mul_add_epi8(q4b_2, q8b_2);
         const __m256i p_1 = _mm256_madd_epi16(p16_1, mone);
@@ -10661,10 +10662,10 @@ void ggml_vec_dot_iq4_xs_q8_K(int n, float * restrict s, size_t bs, const void *
             const __m128i q4bits_2 = _mm_loadu_si128((const __m128i*)qs);  qs += 16;
             const __m256i q8b_1 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
             const __m256i q8b_2 = _mm256_loadu_si256((const __m256i *)q8); q8 += 32;
-            const __m256i q4b_1 = _mm256_set_m128i(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_1, 4), m4b)),
-                                                   _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_1, m4b)));
-            const __m256i q4b_2 = _mm256_set_m128i(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_2, 4), m4b)),
-                                                   _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_2, m4b)));
+            const __m256i q4b_1 = MM256_SET_M128I(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_1, 4), m4b)),
+                                                  _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_1, m4b)));
+            const __m256i q4b_2 = MM256_SET_M128I(_mm_shuffle_epi8(values128, _mm_and_si128(_mm_srli_epi16(q4bits_2, 4), m4b)),
+                                                  _mm_shuffle_epi8(values128, _mm_and_si128(q4bits_2, m4b)));
             const __m256i p16_1 = mul_add_epi8(q4b_1, q8b_1);
             const __m256i p16_2 = mul_add_epi8(q4b_2, q8b_2);
             const int16_t ls1 = ((x[ibl].scales_l[ib/2] & 0xf) | ((sh << 4) & 0x30)) - 32;

ggml-vulkan.h

@@ -10,6 +10,7 @@ extern "C" {
 #define GGML_VK_NAME "Vulkan"
 #define GGML_VK_MAX_DEVICES 16
 
+GGML_API void ggml_vk_instance_init(void);
 GGML_API void ggml_vk_init_cpu_assist(void);
 
 GGML_API void ggml_vk_preallocate_buffers_graph_cpu_assist(struct ggml_tensor * node);

grammars/json.gbnf

@@ -15,7 +15,7 @@ array  ::=
 
 string ::=
   "\"" (
-    [^"\\] |
+    [^"\\\x7F\x00-\x1F] |
     "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
   )* "\"" ws
 

grammars/json_arr.gbnf

@@ -24,7 +24,7 @@ array  ::=
 
 string ::=
   "\"" (
-    [^"\\] |
+    [^"\\\x7F\x00-\x1F] |
     "\\" (["\\/bfnrt] | "u" [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F] [0-9a-fA-F]) # escapes
   )* "\"" ws
 

llama.cpp

@@ -1684,7 +1684,6 @@ struct llama_cparams {
 
     bool embeddings;
     bool offload_kqv;
-
     enum llama_pooling_type pooling_type;
 
     ggml_backend_sched_eval_callback cb_eval;
@@ -5014,8 +5013,8 @@ static struct ggml_tensor * llm_build_kqv(
         ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
     }
 
-#if defined(GGML_USE_VULKAN) || defined(GGML_USE_KOMPUTE)
-#pragma message("TODO: ALiBi support in ggml_soft_max_ext is not implemented for Vulkan, and Kompute")
+#if defined(GGML_USE_KOMPUTE)
+#pragma message("TODO: ALiBi support in ggml_soft_max_ext is not implemented for Kompute")
 #pragma message("      Falling back to ggml_alibi(). Will become an error in Mar 2024")
 #pragma message("ref:  https://github.com/ggerganov/llama.cpp/pull/5488")
     if (hparams.f_max_alibi_bias > 0.0f) {
@@ -8030,7 +8029,14 @@ static void llama_set_inputs(llama_context & lctx, const llama_batch & batch) {
         ggml_backend_tensor_set(lctx.inp_pos, batch.pos, 0, n_tokens*ggml_element_size(lctx.inp_pos));
     }
 
-    if (hparams.causal_attn) {
+    GGML_ASSERT(
+        (hparams.causal_attn || cparams.embeddings) &&
+        "non-causal attention with generative models is not supported"
+    );
+
+    // NOTE: hparams.causal_attn indicates the model is capable of generation and uses the kv cache.
+    // But if cparams.embeddings is set, the attention will be non-causal nonetheless.
+    if (!cparams.embeddings) {
         const int64_t n_kv     = kv_self.n;
         const int64_t n_tokens = batch.n_tokens;
 
@@ -8055,8 +8061,9 @@ static void llama_set_inputs(llama_context & lctx, const llama_batch & batch) {
             }
         }
     } else {
-        // non-causal attention attends only the tokens within the batch (i.e. the KV cache is not used)
+        // for models using the kv cache, the mask needs to match the kv cache size
         const int64_t n_tokens = batch.n_tokens;
+        const int64_t n_stride = hparams.causal_attn ? kv_self.n : n_tokens;
 
         assert(ggml_backend_buffer_is_host(lctx.inp_KQ_mask->buffer));
 
@@ -8075,7 +8082,11 @@ static void llama_set_inputs(llama_context & lctx, const llama_batch & batch) {
                         }
                     }
 
-                    data[h*(n_tokens*n_tokens) + j*n_tokens + i] = f;
+                    data[h*(n_tokens*n_tokens) + j*n_stride + i] = f;
+                }
+
+                for (int i = n_tokens; i < n_stride; ++i) {
+                    data[h*(n_tokens*n_tokens) + j*n_stride + i] = -INFINITY;
                 }
             }
         }
@@ -13158,6 +13169,10 @@ void llama_set_abort_callback(struct llama_context * ctx, bool (*abort_callback)
     ctx->abort_callback_data = abort_callback_data;
 }
 
+void llama_set_embeddings(struct llama_context * ctx, bool embeddings) {
+    ctx->cparams.embeddings = embeddings;
+}
+
 struct llama_batch llama_batch_get_one(
              llama_token * tokens,
                  int32_t   n_tokens,

llama.h

@@ -641,6 +641,10 @@ extern "C" {
     // n_threads_batch is the number of threads used for prompt and batch processing (multiple tokens)
     LLAMA_API void llama_set_n_threads(struct llama_context * ctx, uint32_t n_threads, uint32_t n_threads_batch);
 
+    // Set whether to use causal attention or not
+    // If set to true, the model will only attend to the past tokens
+    LLAMA_API void llama_set_embeddings(struct llama_context * ctx, bool embeddings);
+
     // Set abort callback
     LLAMA_API void llama_set_abort_callback(struct llama_context * ctx, ggml_abort_callback abort_callback, void * abort_callback_data);