npm - @fugood/llama.node - Versions diffs - 0.3.0 → 0.3.2 - Mend

@fugood/llama.node 0.3.0 → 0.3.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (187) hide show

package/CMakeLists.txt +1 -10
package/bin/darwin/arm64/llama-node.node +0 -0
package/bin/darwin/x64/llama-node.node +0 -0
package/bin/linux/arm64/llama-node.node +0 -0
package/bin/linux/x64/llama-node.node +0 -0
package/bin/linux-vulkan/arm64/llama-node.node +0 -0
package/bin/linux-vulkan/x64/llama-node.node +0 -0
package/bin/win32/arm64/llama-node.node +0 -0
package/bin/win32/arm64/node.lib +0 -0
package/bin/win32/x64/llama-node.node +0 -0
package/bin/win32/x64/node.lib +0 -0
package/bin/win32-vulkan/arm64/llama-node.node +0 -0
package/bin/win32-vulkan/arm64/node.lib +0 -0
package/bin/win32-vulkan/x64/llama-node.node +0 -0
package/bin/win32-vulkan/x64/node.lib +0 -0
package/package.json +6 -4
package/src/LlamaCompletionWorker.cpp +6 -6
package/src/LlamaContext.cpp +7 -9
package/src/common.hpp +2 -1
package/src/llama.cpp/.github/workflows/build.yml +98 -24
package/src/llama.cpp/.github/workflows/close-issue.yml +5 -0
package/src/llama.cpp/.github/workflows/docker.yml +43 -34
package/src/llama.cpp/.github/workflows/nix-ci-aarch64.yml +7 -0
package/src/llama.cpp/.github/workflows/nix-ci.yml +7 -0
package/src/llama.cpp/.github/workflows/python-check-requirements.yml +2 -4
package/src/llama.cpp/.github/workflows/python-type-check.yml +3 -1
package/src/llama.cpp/.github/workflows/server.yml +7 -0
package/src/llama.cpp/CMakeLists.txt +20 -8
package/src/llama.cpp/common/CMakeLists.txt +12 -10
package/src/llama.cpp/common/arg.cpp +2006 -0
package/src/llama.cpp/common/arg.h +77 -0
package/src/llama.cpp/common/common.cpp +496 -1632
package/src/llama.cpp/common/common.h +161 -63
package/src/llama.cpp/common/console.cpp +3 -0
package/src/llama.cpp/common/log.cpp +401 -0
package/src/llama.cpp/common/log.h +66 -698
package/src/llama.cpp/common/ngram-cache.cpp +3 -0
package/src/llama.cpp/common/sampling.cpp +348 -350
package/src/llama.cpp/common/sampling.h +62 -139
package/src/llama.cpp/common/stb_image.h +5990 -6398
package/src/llama.cpp/common/train.cpp +2 -0
package/src/llama.cpp/docs/build.md +36 -1
package/src/llama.cpp/examples/CMakeLists.txt +0 -1
package/src/llama.cpp/examples/baby-llama/baby-llama.cpp +1 -2
package/src/llama.cpp/examples/batched/batched.cpp +39 -55
package/src/llama.cpp/examples/batched-bench/batched-bench.cpp +34 -44
package/src/llama.cpp/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp +55 -52
package/src/llama.cpp/examples/cvector-generator/cvector-generator.cpp +15 -15
package/src/llama.cpp/examples/cvector-generator/pca.hpp +3 -13
package/src/llama.cpp/examples/embedding/embedding.cpp +143 -87
package/src/llama.cpp/examples/eval-callback/eval-callback.cpp +33 -33
package/src/llama.cpp/examples/export-lora/export-lora.cpp +36 -35
package/src/llama.cpp/examples/gbnf-validator/gbnf-validator.cpp +14 -39
package/src/llama.cpp/examples/gen-docs/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/gen-docs/gen-docs.cpp +83 -0
package/src/llama.cpp/examples/gguf-split/gguf-split.cpp +58 -39
package/src/llama.cpp/examples/gritlm/gritlm.cpp +34 -27
package/src/llama.cpp/examples/imatrix/imatrix.cpp +59 -62
package/src/llama.cpp/examples/infill/infill.cpp +117 -132
package/src/llama.cpp/examples/llama-bench/llama-bench.cpp +265 -58
package/src/llama.cpp/examples/llama.android/llama/src/main/cpp/llama-android.cpp +29 -22
package/src/llama.cpp/examples/llava/CMakeLists.txt +7 -0
package/src/llama.cpp/examples/llava/clip.cpp +685 -150
package/src/llama.cpp/examples/llava/clip.h +11 -2
package/src/llama.cpp/examples/llava/llava-cli.cpp +47 -58
package/src/llama.cpp/examples/llava/llava.cpp +110 -24
package/src/llama.cpp/examples/llava/llava.h +2 -3
package/src/llama.cpp/examples/llava/minicpmv-cli.cpp +323 -0
package/src/llama.cpp/examples/llava/requirements.txt +1 -0
package/src/llama.cpp/examples/lookahead/lookahead.cpp +42 -43
package/src/llama.cpp/examples/lookup/lookup-create.cpp +10 -8
package/src/llama.cpp/examples/lookup/lookup-stats.cpp +23 -22
package/src/llama.cpp/examples/lookup/lookup.cpp +40 -43
package/src/llama.cpp/examples/main/main.cpp +210 -262
package/src/llama.cpp/examples/parallel/parallel.cpp +49 -49
package/src/llama.cpp/examples/passkey/passkey.cpp +42 -50
package/src/llama.cpp/examples/perplexity/perplexity.cpp +187 -200
package/src/llama.cpp/examples/quantize/CMakeLists.txt +1 -1
package/src/llama.cpp/examples/quantize/quantize.cpp +27 -9
package/src/llama.cpp/examples/quantize-stats/quantize-stats.cpp +2 -3
package/src/llama.cpp/examples/retrieval/retrieval.cpp +49 -44
package/src/llama.cpp/examples/rpc/rpc-server.cpp +24 -1
package/src/llama.cpp/examples/save-load-state/save-load-state.cpp +32 -35
package/src/llama.cpp/examples/server/CMakeLists.txt +3 -5
package/src/llama.cpp/examples/server/server.cpp +1027 -1073
package/src/llama.cpp/examples/server/tests/requirements.txt +2 -1
package/src/llama.cpp/examples/server/utils.hpp +107 -105
package/src/llama.cpp/examples/simple/simple.cpp +35 -41
package/src/llama.cpp/examples/speculative/speculative.cpp +129 -103
package/src/llama.cpp/examples/sycl/run-llama2.sh +10 -19
package/src/llama.cpp/examples/sycl/win-run-llama2.bat +1 -1
package/src/llama.cpp/examples/tokenize/tokenize.cpp +25 -27
package/src/llama.cpp/ggml/CMakeLists.txt +14 -3
package/src/llama.cpp/ggml/include/ggml-alloc.h +3 -3
package/src/llama.cpp/ggml/include/ggml-backend.h +145 -60
package/src/llama.cpp/ggml/include/ggml-blas.h +3 -3
package/src/llama.cpp/ggml/include/ggml-cann.h +15 -19
package/src/llama.cpp/ggml/include/ggml-cuda.h +16 -16
package/src/llama.cpp/ggml/include/ggml-metal.h +5 -8
package/src/llama.cpp/ggml/include/ggml-rpc.h +5 -5
package/src/llama.cpp/ggml/include/ggml-sycl.h +8 -8
package/src/llama.cpp/ggml/include/ggml-vulkan.h +7 -7
package/src/llama.cpp/ggml/include/ggml.h +293 -186
package/src/llama.cpp/ggml/src/CMakeLists.txt +86 -44
package/src/llama.cpp/ggml/src/ggml-aarch64.c +2135 -1119
package/src/llama.cpp/ggml/src/ggml-alloc.c +6 -0
package/src/llama.cpp/ggml/src/ggml-backend-impl.h +152 -70
package/src/llama.cpp/ggml/src/{ggml-backend.c → ggml-backend.cpp} +606 -286
package/src/llama.cpp/ggml/src/ggml-blas.cpp +9 -10
package/src/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +4 -27
package/src/llama.cpp/ggml/src/ggml-cann/acl_tensor.h +32 -4
package/src/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +179 -41
package/src/llama.cpp/ggml/src/ggml-cann/common.h +1 -0
package/src/llama.cpp/ggml/src/ggml-cann/kernels/CMakeLists.txt +2 -1
package/src/llama.cpp/ggml/src/ggml-cann/kernels/ascendc_kernels.h +2 -0
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +278 -0
package/src/llama.cpp/ggml/src/ggml-cann.cpp +215 -216
package/src/llama.cpp/ggml/src/ggml-common.h +20 -0
package/src/llama.cpp/ggml/src/ggml-cpu-impl.h +614 -0
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/cuda.h +14 -0
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +178 -0
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/musa.h +134 -0
package/src/llama.cpp/ggml/src/ggml-impl.h +49 -603
package/src/llama.cpp/ggml/src/ggml-kompute.cpp +4 -24
package/src/llama.cpp/ggml/src/ggml-quants.c +972 -92
package/src/llama.cpp/ggml/src/ggml-quants.h +15 -0
package/src/llama.cpp/ggml/src/ggml-rpc.cpp +116 -66
package/src/llama.cpp/ggml/src/ggml-sycl/backend.hpp +3 -0
package/src/llama.cpp/ggml/src/ggml-sycl/common.cpp +11 -0
package/src/llama.cpp/ggml/src/ggml-sycl/common.hpp +52 -0
package/src/llama.cpp/ggml/src/ggml-sycl/conv.cpp +99 -0
package/src/llama.cpp/ggml/src/ggml-sycl/conv.hpp +21 -0
package/src/llama.cpp/ggml/src/ggml-sycl/convert.cpp +57 -57
package/src/llama.cpp/ggml/src/ggml-sycl/convert.hpp +1 -1
package/src/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp +106 -106
package/src/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +4 -4
package/src/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +16 -3
package/src/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +101 -0
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.cpp +125 -0
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.hpp +23 -0
package/src/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +1 -1
package/src/llama.cpp/ggml/src/ggml-sycl/norm.cpp +6 -3
package/src/llama.cpp/ggml/src/ggml-sycl/presets.hpp +2 -0
package/src/llama.cpp/ggml/src/ggml-sycl/rope.cpp +1 -1
package/src/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp +71 -0
package/src/llama.cpp/ggml/src/ggml-sycl/tsembd.hpp +21 -0
package/src/llama.cpp/ggml/src/ggml-sycl.cpp +97 -169
package/src/llama.cpp/ggml/src/ggml-vulkan.cpp +1508 -1124
package/src/llama.cpp/ggml/src/ggml.c +3001 -1647
package/src/llama.cpp/ggml/src/llamafile/sgemm.cpp +192 -0
package/src/llama.cpp/ggml/src/vulkan-shaders/CMakeLists.txt +2 -0
package/src/llama.cpp/ggml/src/vulkan-shaders/vulkan-shaders-gen.cpp +88 -40
package/src/llama.cpp/include/llama.h +241 -264
package/src/llama.cpp/models/ggml-vocab-chameleon.gguf.inp +112 -0
package/src/llama.cpp/models/ggml-vocab-chameleon.gguf.out +46 -0
package/src/llama.cpp/requirements/requirements-convert_legacy_llama.txt +1 -1
package/src/llama.cpp/src/llama-grammar.cpp +721 -122
package/src/llama.cpp/src/llama-grammar.h +120 -15
package/src/llama.cpp/src/llama-impl.h +156 -1
package/src/llama.cpp/src/llama-sampling.cpp +1375 -303
package/src/llama.cpp/src/llama-sampling.h +20 -47
package/src/llama.cpp/src/llama-vocab.cpp +343 -120
package/src/llama.cpp/src/llama-vocab.h +33 -17
package/src/llama.cpp/src/llama.cpp +4247 -1525
package/src/llama.cpp/src/unicode-data.cpp +6 -4
package/src/llama.cpp/src/unicode-data.h +4 -4
package/src/llama.cpp/src/unicode.cpp +15 -7
package/src/llama.cpp/tests/CMakeLists.txt +3 -0
package/src/llama.cpp/tests/test-arg-parser.cpp +131 -0
package/src/llama.cpp/tests/test-backend-ops.cpp +1592 -289
package/src/llama.cpp/tests/test-barrier.cpp +93 -0
package/src/llama.cpp/tests/test-grad0.cpp +187 -70
package/src/llama.cpp/tests/test-grammar-integration.cpp +23 -38
package/src/llama.cpp/tests/test-grammar-parser.cpp +6 -4
package/src/llama.cpp/tests/test-json-schema-to-grammar.cpp +6 -4
package/src/llama.cpp/tests/test-llama-grammar.cpp +9 -8
package/src/llama.cpp/tests/test-log.cpp +39 -0
package/src/llama.cpp/tests/test-quantize-fns.cpp +6 -0
package/src/llama.cpp/tests/test-rope.cpp +1 -1
package/src/llama.cpp/tests/test-sampling.cpp +157 -98
package/src/llama.cpp/tests/test-tokenizer-0.cpp +55 -35
package/patches/llama.patch +0 -22
package/src/llama.cpp/.github/workflows/bench.yml +0 -310
package/src/llama.cpp/common/grammar-parser.cpp +0 -536
package/src/llama.cpp/common/grammar-parser.h +0 -29
package/src/llama.cpp/examples/benchmark/CMakeLists.txt +0 -6
package/src/llama.cpp/examples/benchmark/benchmark-matmult.cpp +0 -275

package/src/llama.cpp/src/llama-vocab.h CHANGED Viewed

@@ -6,6 +6,9 @@
 #include <vector>
 #include <unordered_map>
 #include <map>
+#include <set>
+struct llm_tokenizer;
 struct llama_vocab {
     using id    = llama_token;
@@ -18,6 +21,8 @@ struct llama_vocab {
         tattr attr;
     };
+    uint32_t n_vocab = 0; // TODO: not great because has to keep in sync with hparams.n_vocab
     enum llama_vocab_type     type     = LLAMA_VOCAB_TYPE_SPM;
     enum llama_vocab_pre_type type_pre = LLAMA_VOCAB_PRE_TYPE_DEFAULT;
@@ -35,33 +40,42 @@ struct llama_vocab {
     id special_bos_id  = 1;
     id special_eos_id  = 2;
     id special_unk_id  = 0;
-    id special_sep_id  = -1;
-    id special_pad_id  = -1;
-    id special_cls_id  = -1;
-    id special_mask_id = -1;
+    id special_sep_id  = LLAMA_TOKEN_NULL;
+    id special_pad_id  = LLAMA_TOKEN_NULL;
+    id special_cls_id  = LLAMA_TOKEN_NULL;
+    id special_mask_id = LLAMA_TOKEN_NULL;
     id linefeed_id       = 13;
-    id special_prefix_id = -1;
-    id special_suffix_id = -1;
-    id special_middle_id = -1;
-    id special_eot_id    = -1; // TODO: move above after "eos_id", and here add "file separator" token
+    id special_prefix_id = LLAMA_TOKEN_NULL;
+    id special_suffix_id = LLAMA_TOKEN_NULL;
+    id special_middle_id = LLAMA_TOKEN_NULL;
+    id special_eot_id    = LLAMA_TOKEN_NULL; // TODO: move above after "eos_id", and here add "file separator" token
+    id special_eom_id    = LLAMA_TOKEN_NULL;
+    // set of all tokens that cause "end of generation"
+    std::set<id> special_eog_ids;
     // tokenizer flags
-    bool tokenizer_add_space_prefix = false;
-    bool tokenizer_add_bos          = false;
-    bool tokenizer_add_eos          = false;
-    bool tokenizer_ignore_merges    = false;
-    bool tokenizer_clean_spaces     = false;  // clean_up_tokenization_spaces
+    bool tokenizer_add_space_prefix           = false;
+    bool tokenizer_add_bos                    = false;
+    bool tokenizer_add_eos                    = false;
+    bool tokenizer_ignore_merges              = false;
+    bool tokenizer_clean_spaces               = false;  // clean_up_tokenization_spaces
     bool tokenizer_remove_extra_whitespaces   = false;
     bool tokenizer_escape_whitespaces         = true;
     bool tokenizer_treat_whitespace_as_suffix = false;
     std::vector<char> precompiled_charsmap;
+    llm_tokenizer * tokenizer = nullptr;
+    llama_vocab() = default;
+    ~llama_vocab();
     int find_bpe_rank(const std::string & token_left, const std::string & token_right) const;
-};
-const struct llama_vocab * llama_get_vocab(const struct llama_context * ctx);
+    void init_tokenizer();
+};
 //
 // internal API
@@ -75,6 +89,7 @@ std::vector<llama_vocab::id> llama_tokenize_internal(
         bool add_special,
         bool parse_special = false);
+// TODO: move the API below as member functions of llama_vocab
 llama_token llama_byte_to_token_impl(const llama_vocab & vocab, uint8_t ch);
 const char * llama_token_get_text_impl(const struct llama_vocab & vocab, llama_token token);
@@ -94,13 +109,14 @@ llama_token llama_token_sep_impl(const struct llama_vocab & vocab);
 llama_token llama_token_nl_impl (const struct llama_vocab & vocab);
 llama_token llama_token_pad_impl(const struct llama_vocab & vocab);
-int32_t llama_add_bos_token_impl(const struct llama_vocab & vocab);
-int32_t llama_add_eos_token_impl(const struct llama_vocab & vocab);
+bool llama_add_bos_token_impl(const struct llama_vocab & vocab);
+bool llama_add_eos_token_impl(const struct llama_vocab & vocab);
 llama_token llama_token_prefix_impl(const struct llama_vocab & vocab);
 llama_token llama_token_middle_impl(const struct llama_vocab & vocab);
 llama_token llama_token_suffix_impl(const struct llama_vocab & vocab);
 llama_token llama_token_eot_impl   (const struct llama_vocab & vocab);
+llama_token llama_token_eom_impl   (const struct llama_vocab & vocab);
 int32_t llama_tokenize_impl(
         const struct llama_vocab & vocab,