npm - @fugood/llama.node - Versions diffs - 0.2.0 → 0.2.2 - Mend

@fugood/llama.node 0.2.0 → 0.2.2

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 (94) hide show

package/CMakeLists.txt +9 -0
package/README.md +1 -1
package/bin/darwin/arm64/default.metallib +0 -0
package/bin/darwin/arm64/llama-node.node +0 -0
package/bin/darwin/x64/default.metallib +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/lib/binding.ts +1 -1
package/package.json +2 -1
package/patches/llama.patch +22 -0
package/src/LlamaContext.cpp +2 -2
package/src/TokenizeWorker.cpp +1 -1
package/src/llama.cpp/CMakeLists.txt +82 -54
package/src/llama.cpp/cmake/arm64-windows-llvm.cmake +16 -0
package/src/llama.cpp/cmake/arm64-windows-msvc.cmake +6 -0
package/src/llama.cpp/common/common.cpp +748 -754
package/src/llama.cpp/common/common.h +49 -41
package/src/llama.cpp/common/grammar-parser.cpp +10 -1
package/src/llama.cpp/common/json-schema-to-grammar.cpp +6 -6
package/src/llama.cpp/common/log.h +5 -5
package/src/llama.cpp/common/sampling.cpp +92 -10
package/src/llama.cpp/common/sampling.h +6 -1
package/src/llama.cpp/common/train.cpp +2 -2
package/src/llama.cpp/examples/CMakeLists.txt +3 -0
package/src/llama.cpp/examples/batched/batched.cpp +1 -1
package/src/llama.cpp/examples/convert-llama2c-to-ggml/convert-llama2c-to-ggml.cpp +1 -1
package/src/llama.cpp/examples/embedding/embedding.cpp +13 -4
package/src/llama.cpp/examples/eval-callback/eval-callback.cpp +2 -2
package/src/llama.cpp/examples/finetune/finetune.cpp +4 -3
package/src/llama.cpp/examples/imatrix/imatrix.cpp +2 -2
package/src/llama.cpp/examples/infill/infill.cpp +8 -8
package/src/llama.cpp/examples/llama-bench/llama-bench.cpp +57 -8
package/src/llama.cpp/examples/llama.android/llama/CMakeLists.txt +55 -0
package/src/llama.cpp/examples/llama.android/{app → llama}/src/main/cpp/CMakeLists.txt +7 -8
package/src/llama.cpp/examples/llama.android/{app → llama}/src/main/cpp/llama-android.cpp +14 -14
package/src/llama.cpp/examples/llava/clip.h +1 -1
package/src/llama.cpp/examples/llava/llava-cli.cpp +27 -7
package/src/llama.cpp/examples/llava/llava.cpp +0 -15
package/src/llama.cpp/examples/lookahead/lookahead.cpp +1 -1
package/src/llama.cpp/examples/lookup/lookup.cpp +1 -1
package/src/llama.cpp/examples/main/main.cpp +29 -17
package/src/llama.cpp/examples/parallel/parallel.cpp +1 -1
package/src/llama.cpp/examples/perplexity/perplexity.cpp +9 -9
package/src/llama.cpp/examples/quantize/quantize.cpp +2 -2
package/src/llama.cpp/examples/retrieval/retrieval.cpp +2 -2
package/src/llama.cpp/examples/rpc/CMakeLists.txt +2 -0
package/src/llama.cpp/examples/rpc/rpc-server.cpp +134 -0
package/src/llama.cpp/examples/server/server.cpp +33 -25
package/src/llama.cpp/examples/server/utils.hpp +1 -1
package/src/llama.cpp/examples/tokenize/tokenize.cpp +359 -9
package/src/llama.cpp/examples/train-text-from-scratch/train-text-from-scratch.cpp +4 -3
package/src/llama.cpp/ggml-backend.c +2 -3
package/src/llama.cpp/ggml-common.h +0 -54
package/src/llama.cpp/ggml-cuda.h +1 -0
package/src/llama.cpp/ggml-impl.h +51 -0
package/src/llama.cpp/ggml-kompute.cpp +13 -3
package/src/llama.cpp/ggml-opencl.cpp +4 -1
package/src/llama.cpp/ggml-quants.c +3715 -2050
package/src/llama.cpp/ggml-rpc.cpp +1155 -0
package/src/llama.cpp/ggml-rpc.h +24 -0
package/src/llama.cpp/ggml-sycl.cpp +119 -673
package/src/llama.cpp/ggml-vulkan-shaders.hpp +9351 -5627
package/src/llama.cpp/ggml-vulkan.cpp +203 -224
package/src/llama.cpp/ggml.c +1208 -1483
package/src/llama.cpp/ggml.h +71 -46
package/src/llama.cpp/llama.cpp +1374 -938
package/src/llama.cpp/llama.h +22 -6
package/src/llama.cpp/requirements.txt +0 -2
package/src/llama.cpp/tests/CMakeLists.txt +1 -1
package/src/llama.cpp/tests/test-backend-ops.cpp +120 -57
package/src/llama.cpp/tests/test-chat-template.cpp +16 -4
package/src/llama.cpp/tests/test-grad0.cpp +43 -83
package/src/llama.cpp/tests/test-grammar-integration.cpp +46 -0
package/src/llama.cpp/tests/test-tokenizer-1-bpe.cpp +27 -3
package/src/llama.cpp/unicode-data.cpp +6969 -2169
package/src/llama.cpp/unicode-data.h +15 -12
package/src/llama.cpp/unicode.cpp +89 -111
package/src/llama.cpp/unicode.h +44 -12
package/src/llama.cpp/build.zig +0 -172
package/src/llama.cpp/ggml-mpi.c +0 -216
package/src/llama.cpp/ggml-mpi.h +0 -39
package/src/llama.cpp/requirements/requirements-convert-lora-to-ggml.txt +0 -2
package/src/llama.cpp/requirements/requirements-convert-persimmon-to-gguf.txt +0 -2

package/src/llama.cpp/ggml-kompute.cpp CHANGED Viewed

@@ -1559,12 +1559,18 @@ static void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml
                 case GGML_OP_SOFT_MAX:
                     {
                         float scale;
-                        memcpy(&scale, dst->op_params, sizeof(float));
+                        float max_bias;
-#pragma message("TODO: add ggml_vk_soft_max() F16/F32 src1 and src2 support")
+                        memcpy(&scale,    (float *)dst->op_params + 0, sizeof(float));
+                        memcpy(&max_bias, (float *)dst->op_params + 1, sizeof(float));
+#pragma message("TODO: add ggml_vk_soft_max() F16 src1 support")
 #pragma message("ref:  https://github.com/ggerganov/llama.cpp/pull/5021")
                         GGML_ASSERT(!src1 || src1t == GGML_TYPE_F32);
-                        GGML_ASSERT(src2 == nullptr);
+#pragma message("TODO: add ALiBi support")
+#pragma message("ref:  https://github.com/ggerganov/llama.cpp/pull/7192")
+                        GGML_ASSERT(max_bias == 0.0f);
                         ggml_vk_soft_max(seq, id_src0, id_src1, id_dst, off_src0, off_src1, off_dst, ne00, ne01, ne02, ne03, scale);
                     } break;
@@ -1671,6 +1677,10 @@ static void ggml_vk_graph_compute(struct ggml_kompute_context * ctx, struct ggml
                     } break;
                 case GGML_OP_ROPE:
                     {
+#pragma message("TODO: implement phi3 frequency factors support")
+#pragma message("      https://github.com/ggerganov/llama.cpp/pull/7225")
+                        GGML_ASSERT(dst->src[2] == nullptr && "phi3 frequency factors not implemented yet");
                         GGML_ASSERT(ne10 == ne02);
                         GGML_ASSERT(src0t == dstt);
                         // const int n_past = ((int32_t *) dst->op_params)[0];

package/src/llama.cpp/ggml-opencl.cpp CHANGED Viewed

@@ -1835,7 +1835,10 @@ static void ggml_cl_mul_mat_q_f32(const ggml_tensor * src0, const ggml_tensor *
                     CL_CHECK(clEnqueueNDRangeKernel(queue, *to_fp32_cl, 1, &offset, &global, local > 0 ? &local : NULL, events.size(), !events.empty() ? events.data() : NULL, NULL));
                 }
-                for (int64_t i12 = i02 * r2, e12 = i12 + r2; i12 < e12; i12++) {
+                int64_t i12 = i02 * r2;
+                int64_t e12 = i12 + r2;
+                events.reserve(e12 - i12);
+                for (; i12 < e12; i12++) {
                     if (mul_mat_vec) { // specialized dequantize_mul_mat_vec kernel
                         // copy src1 to device
                         events.emplace_back();