@novastera-oss/llamarn 0.4.1 → 0.4.3-beta4
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.
- package/RNLlamaCpp.podspec +3 -0
- package/android/CMakeLists.txt +2 -0
- package/android/src/main/cpp/include/llama.h +44 -21
- package/android/src/main/jniLibs/arm64-v8a/libggml-base.so +0 -0
- package/android/src/main/jniLibs/arm64-v8a/libggml-cpu.so +0 -0
- package/android/src/main/jniLibs/arm64-v8a/libggml.so +0 -0
- package/android/src/main/jniLibs/arm64-v8a/libllama.so +0 -0
- package/android/src/main/jniLibs/armeabi-v7a/libggml-base.so +0 -0
- package/android/src/main/jniLibs/armeabi-v7a/libggml-cpu.so +0 -0
- package/android/src/main/jniLibs/armeabi-v7a/libggml.so +0 -0
- package/android/src/main/jniLibs/armeabi-v7a/libllama.so +0 -0
- package/android/src/main/jniLibs/x86/libggml-base.so +0 -0
- package/android/src/main/jniLibs/x86/libggml-cpu.so +0 -0
- package/android/src/main/jniLibs/x86/libggml.so +0 -0
- package/android/src/main/jniLibs/x86/libllama.so +0 -0
- package/android/src/main/jniLibs/x86_64/libggml-base.so +0 -0
- package/android/src/main/jniLibs/x86_64/libggml-cpu.so +0 -0
- package/android/src/main/jniLibs/x86_64/libggml.so +0 -0
- package/android/src/main/jniLibs/x86_64/libllama.so +0 -0
- package/cpp/build-info.cpp +2 -2
- package/cpp/llama.cpp/CMakeLists.txt +12 -0
- package/cpp/llama.cpp/CODEOWNERS +116 -10
- package/cpp/llama.cpp/CONTRIBUTING.md +30 -3
- package/cpp/llama.cpp/README.md +13 -5
- package/cpp/llama.cpp/build-xcframework.sh +5 -0
- package/cpp/llama.cpp/cmake/riscv64-spacemit-linux-gnu-gcc.cmake +29 -0
- package/cpp/llama.cpp/common/CMakeLists.txt +12 -2
- package/cpp/llama.cpp/common/arg.cpp +303 -795
- package/cpp/llama.cpp/common/arg.h +2 -3
- package/cpp/llama.cpp/common/chat-parser-xml-toolcall.cpp +861 -0
- package/cpp/llama.cpp/common/chat-parser-xml-toolcall.h +45 -0
- package/cpp/llama.cpp/common/chat-parser.cpp +156 -15
- package/cpp/llama.cpp/common/chat-parser.h +13 -0
- package/cpp/llama.cpp/common/chat.cpp +1147 -88
- package/cpp/llama.cpp/common/chat.h +16 -3
- package/cpp/llama.cpp/common/common.cpp +70 -15
- package/cpp/llama.cpp/common/common.h +57 -19
- package/cpp/llama.cpp/common/download.cpp +1072 -0
- package/cpp/llama.cpp/common/download.h +55 -0
- package/cpp/llama.cpp/common/http.h +73 -0
- package/cpp/llama.cpp/common/json-partial.cpp +70 -2
- package/cpp/llama.cpp/common/json-schema-to-grammar.cpp +61 -22
- package/cpp/llama.cpp/common/json-schema-to-grammar.h +2 -0
- package/cpp/llama.cpp/common/log.cpp +59 -2
- package/cpp/llama.cpp/common/log.h +12 -4
- package/cpp/llama.cpp/common/sampling.cpp +84 -8
- package/cpp/llama.cpp/common/sampling.h +3 -1
- package/cpp/llama.cpp/common/speculative.cpp +1 -1
- package/cpp/llama.cpp/convert_hf_to_gguf.py +1608 -233
- package/cpp/llama.cpp/convert_hf_to_gguf_update.py +6 -1
- package/cpp/llama.cpp/convert_lora_to_gguf.py +37 -5
- package/cpp/llama.cpp/ggml/CMakeLists.txt +47 -28
- package/cpp/llama.cpp/ggml/include/ggml-backend.h +19 -1
- package/cpp/llama.cpp/ggml/include/ggml-cpu.h +1 -1
- package/cpp/llama.cpp/ggml/include/ggml-hexagon.h +19 -0
- package/cpp/llama.cpp/ggml/include/ggml-metal.h +1 -6
- package/cpp/llama.cpp/ggml/include/ggml-rpc.h +7 -9
- package/cpp/llama.cpp/ggml/include/ggml-zdnn.h +2 -1
- package/cpp/llama.cpp/ggml/include/ggml.h +199 -6
- package/cpp/llama.cpp/ggml/src/CMakeLists.txt +38 -0
- package/cpp/llama.cpp/ggml/src/ggml-alloc.c +299 -130
- package/cpp/llama.cpp/ggml/src/ggml-backend-impl.h +4 -4
- package/cpp/llama.cpp/ggml/src/ggml-backend-reg.cpp +21 -5
- package/cpp/llama.cpp/ggml/src/ggml-backend.cpp +99 -2
- package/cpp/llama.cpp/ggml/src/ggml-blas/CMakeLists.txt +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-blas/ggml-blas.cpp +1 -0
- package/cpp/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +57 -45
- package/cpp/llama.cpp/ggml/src/ggml-cann/acl_tensor.h +138 -47
- package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +1584 -1773
- package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.h +201 -317
- package/cpp/llama.cpp/ggml/src/ggml-cann/common.h +146 -187
- package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +771 -713
- package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +135 -77
- package/cpp/llama.cpp/ggml/src/ggml-cpu/amx/amx.cpp +5 -2
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/arm/quants.c +428 -26
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/loongarch/quants.c +16 -17
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/riscv/quants.c +318 -145
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/s390/quants.c +155 -60
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/x86/repack.cpp +8 -8
- package/cpp/llama.cpp/ggml/src/ggml-cpu/arch-fallback.h +0 -1
- package/cpp/llama.cpp/ggml/src/ggml-cpu/common.h +14 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +10 -9
- package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +108 -64
- package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +14 -4
- package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +530 -87
- package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +37 -45
- package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +349 -127
- package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.cpp +947 -1218
- package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.h +5 -4
- package/cpp/llama.cpp/ggml/src/ggml-cpu/repack.cpp +143 -29
- package/cpp/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +82 -76
- package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime.cpp +1025 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/unary-ops.cpp +151 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/unary-ops.h +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.cpp +233 -28
- package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.h +326 -66
- package/cpp/llama.cpp/ggml/src/ggml-cuda/CMakeLists.txt +12 -3
- package/cpp/llama.cpp/ggml/src/ggml-cuda/argsort.cu +102 -6
- package/cpp/llama.cpp/ggml/src/ggml-cuda/binbcast.cu +110 -76
- package/cpp/llama.cpp/ggml/src/ggml-cuda/common.cuh +167 -38
- package/cpp/llama.cpp/ggml/src/ggml-cuda/conv2d.cu +6 -11
- package/cpp/llama.cpp/ggml/src/ggml-cuda/convert.cuh +12 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy-utils.cuh +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy.cu +245 -151
- package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy.cuh +1 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-common.cuh +341 -289
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile.cu +49 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile.cuh +1233 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec.cuh +586 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-wmma-f16.cu +6 -6
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +48 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn.cu +123 -220
- package/cpp/llama.cpp/ggml/src/ggml-cuda/getrows.cu +41 -39
- package/cpp/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu +715 -45
- package/cpp/llama.cpp/ggml/src/ggml-cuda/im2col.cu +150 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/im2col.cuh +1 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mma.cuh +321 -24
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmf.cu +93 -351
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmf.cuh +828 -1
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmid.cu +164 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmid.cuh +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cu +3 -166
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cuh +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvf.cu +371 -78
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvf.cuh +3 -2
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvq.cu +279 -147
- package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvq.cuh +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-cuda/norm.cu +97 -85
- package/cpp/llama.cpp/ggml/src/ggml-cuda/pad.cu +46 -23
- package/cpp/llama.cpp/ggml/src/ggml-cuda/pad_reflect_1d.cu +63 -54
- package/cpp/llama.cpp/ggml/src/ggml-cuda/quantize.cu +12 -10
- package/cpp/llama.cpp/ggml/src/ggml-cuda/rope.cu +192 -77
- package/cpp/llama.cpp/ggml/src/ggml-cuda/rope.cuh +2 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/scale.cu +10 -9
- package/cpp/llama.cpp/ggml/src/ggml-cuda/set-rows.cu +137 -75
- package/cpp/llama.cpp/ggml/src/ggml-cuda/set.cu +39 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/set.cuh +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq112-dv112.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq128-dv128.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq256-dv256.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq40-dv40.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq576-dv512.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq64-dv64.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq80-dv80.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq96-dv96.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +40 -19
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/topk-moe.cu +336 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/topk-moe.cuh +16 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/tsembd.cu +3 -3
- package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cu +105 -11
- package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cuh +36 -0
- package/cpp/llama.cpp/ggml/src/ggml-cuda/upscale.cu +87 -6
- package/cpp/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +28 -12
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/CMakeLists.txt +68 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp +3807 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/CMakeLists.txt +40 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/act-ops.c +442 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/binary-ops.c +360 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +157 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-ctx.h +40 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-dma.c +69 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-dma.h +119 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-msg.h +156 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-ops.h +64 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp_iface.idl +16 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-exp.c +93 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-inverse.c +60 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-sigmoid.c +49 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-utils.c +960 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-utils.h +1032 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/main.c +829 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/matmul-ops.c +2223 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/ops-utils.h +149 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/rope-ops.c +418 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/softmax-ops.c +402 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/unary-ops.c +255 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/worker-pool.c +297 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/worker-pool.h +57 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp-utils.c +448 -0
- package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp-utils.h +220 -0
- package/cpp/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +8 -13
- package/cpp/llama.cpp/ggml/src/ggml-impl.h +110 -12
- package/cpp/llama.cpp/ggml/src/ggml-metal/CMakeLists.txt +6 -5
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-common.cpp +446 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-context.h +33 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-context.m +599 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-device.cpp +1662 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-device.h +251 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-device.m +1527 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +244 -39
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-ops.cpp +3844 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-ops.h +90 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.cpp +723 -0
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.metal +3453 -1907
- package/cpp/llama.cpp/ggml/src/ggml-musa/CMakeLists.txt +3 -1
- package/cpp/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +10 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +1331 -109
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/cvt.cl +126 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +31 -4
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +35 -7
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +31 -4
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32.cl +162 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32.cl +156 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/get_rows.cl +36 -12
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_kq_kqv.cl +273 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +24 -10
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +24 -10
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl +154 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/pad.cl +29 -20
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/rms_norm.cl +25 -10
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/rope.cl +50 -24
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/set_rows.cl +123 -10
- package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/tsembd.cl +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-quants.c +1 -0
- package/cpp/llama.cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +341 -161
- package/cpp/llama.cpp/ggml/src/ggml-sycl/backend.hpp +6 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/binbcast.cpp +6 -5
- package/cpp/llama.cpp/ggml/src/ggml-sycl/common.hpp +74 -15
- package/cpp/llama.cpp/ggml/src/ggml-sycl/concat.cpp +50 -30
- package/cpp/llama.cpp/ggml/src/ggml-sycl/conv.cpp +10 -4
- package/cpp/llama.cpp/ggml/src/ggml-sycl/convert.cpp +166 -99
- package/cpp/llama.cpp/ggml/src/ggml-sycl/count-equal.cpp +79 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/count-equal.hpp +9 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/cpy.cpp +72 -94
- package/cpp/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +67 -49
- package/cpp/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +21 -31
- package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +252 -316
- package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +6 -2
- package/cpp/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +9 -6
- package/cpp/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +359 -142
- package/cpp/llama.cpp/ggml/src/ggml-sycl/gla.cpp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-sycl/im2col.cpp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-sycl/mmq.cpp +80 -60
- package/cpp/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +201 -132
- package/cpp/llama.cpp/ggml/src/ggml-sycl/norm.cpp +230 -55
- package/cpp/llama.cpp/ggml/src/ggml-sycl/norm.hpp +2 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/pad.cpp +97 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/pad.hpp +24 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/pad_reflect_1d.cpp +72 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/pad_reflect_1d.hpp +8 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/presets.hpp +2 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/repeat_back.cpp +76 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/repeat_back.hpp +8 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/roll.cpp +122 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/roll.hpp +20 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/rope.cpp +50 -41
- package/cpp/llama.cpp/ggml/src/ggml-sycl/set.cpp +73 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/set.hpp +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/set_rows.cpp +45 -36
- package/cpp/llama.cpp/ggml/src/ggml-sycl/softmax.cpp +330 -165
- package/cpp/llama.cpp/ggml/src/ggml-sycl/softmax.hpp +4 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/ssm_conv.cpp +127 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/ssm_conv.hpp +5 -0
- package/cpp/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp +12 -6
- package/cpp/llama.cpp/ggml/src/ggml-sycl/wkv.cpp +16 -12
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +38 -18
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +4184 -2159
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/abs.comp +21 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/add1.comp +28 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/arange.comp +20 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +33 -26
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/argsort_large.comp +114 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/ceil.comp +22 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/concat.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_dw.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +53 -30
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +3 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +13 -6
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_transpose.comp +67 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/cos.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/count_equal.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_f32.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs.comp → dequant_funcs.glsl} +138 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs_cm2.comp → dequant_funcs_cm2.glsl} +18 -4
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_head.comp → dequant_head.glsl} +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_s.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +3 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +7 -6
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +5 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_nl.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_xs.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +3 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +3 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_0.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +3 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_0.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_1.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +3 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q8_0.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/diag_mask_inf.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/div.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +3 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/fill.comp +19 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +52 -14
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{flash_attn_base.comp → flash_attn_base.glsl} +50 -12
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +61 -12
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +54 -12
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +5 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/floor.comp +22 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu_quick.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{generic_binary_head.comp → generic_binary_head.glsl} +10 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +21 -12
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +28 -18
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{glu_head.comp → glu_head.glsl} +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/group_norm.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +15 -7
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +125 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/leaky_relu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +18 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl +229 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +33 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_m.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_s.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +9 -7
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp +9 -7
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +3 -5
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q3_k.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +3 -5
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +3 -5
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q6_k.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +140 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +106 -634
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +118 -9
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +556 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +70 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +77 -214
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +589 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +78 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +97 -13
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/neg.comp +20 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/norm.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_adamw.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +25 -4
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/pool2d.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +55 -5
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/relu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/repeat_back.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +45 -3
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_back.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +227 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +20 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +5 -52
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +5 -35
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +5 -35
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +27 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +5 -41
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/round.comp +29 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sigmoid.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/silu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/silu_back.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sin.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +6 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/softplus.comp +23 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/square.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +44 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/ssm_scan.comp +140 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/step.comp +22 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sub.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +1 -1
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/tanh.comp +2 -2
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +5 -4
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp +171 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/trunc.comp +22 -0
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{types.comp → types.glsl} +79 -29
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +36 -12
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +471 -196
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/CMakeLists.txt +8 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1690 -383
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/bin_op.tmpl.wgsl +188 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +45 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +930 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/cpy.tmpl.wgsl +101 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +57 -10
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +874 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +323 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +25 -912
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +97 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.tmpl.wgsl +247 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.tmpl.wgsl +302 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl +267 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +123 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/rope.tmpl.wgsl +295 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl +90 -0
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/{set_rows.wgsl → set_rows.tmpl.wgsl} +38 -8
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/soft_max.tmpl.wgsl +345 -0
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/common.hpp +59 -0
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/ggml-zdnn.cpp +96 -314
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/mmf.cpp +80 -0
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/mmf.hpp +12 -0
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/utils.cpp +79 -0
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/utils.hpp +19 -0
- package/cpp/llama.cpp/ggml/src/ggml.c +440 -17
- package/cpp/llama.cpp/ggml/src/gguf.cpp +104 -29
- package/cpp/llama.cpp/gguf-py/gguf/constants.py +363 -13
- package/cpp/llama.cpp/gguf-py/gguf/gguf_writer.py +64 -0
- package/cpp/llama.cpp/gguf-py/gguf/lazy.py +8 -3
- package/cpp/llama.cpp/gguf-py/gguf/scripts/gguf_convert_endian.py +6 -0
- package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py +156 -18
- package/cpp/llama.cpp/gguf-py/gguf/utility.py +80 -0
- package/cpp/llama.cpp/gguf-py/gguf/vocab.py +4 -4
- package/cpp/llama.cpp/include/llama.h +44 -21
- package/cpp/llama.cpp/media/llama1-icon-transparent.png +0 -0
- package/cpp/llama.cpp/media/llama1-icon-transparent.svg +77 -0
- package/cpp/llama.cpp/media/llama1-icon.png +0 -0
- package/cpp/llama.cpp/media/llama1-icon.svg +87 -0
- package/cpp/llama.cpp/requirements/requirements-all.txt +2 -0
- package/cpp/llama.cpp/requirements/requirements-convert_hf_to_gguf.txt +3 -3
- package/cpp/llama.cpp/requirements/requirements-convert_legacy_llama.txt +3 -1
- package/cpp/llama.cpp/requirements/requirements-tool_bench.txt +1 -1
- package/cpp/llama.cpp/src/CMakeLists.txt +101 -0
- package/cpp/llama.cpp/src/llama-adapter.cpp +33 -0
- package/cpp/llama.cpp/src/llama-adapter.h +3 -0
- package/cpp/llama.cpp/src/llama-arch.cpp +344 -14
- package/cpp/llama.cpp/src/llama-arch.h +50 -0
- package/cpp/llama.cpp/src/llama-batch.cpp +63 -31
- package/cpp/llama.cpp/src/llama-batch.h +13 -2
- package/cpp/llama.cpp/src/llama-chat.cpp +85 -3
- package/cpp/llama.cpp/src/llama-chat.h +4 -0
- package/cpp/llama.cpp/src/llama-context.cpp +300 -45
- package/cpp/llama.cpp/src/llama-context.h +16 -6
- package/cpp/llama.cpp/src/llama-cparams.h +2 -1
- package/cpp/llama.cpp/src/llama-grammar.cpp +17 -9
- package/cpp/llama.cpp/src/llama-graph.cpp +226 -64
- package/cpp/llama.cpp/src/llama-graph.h +27 -5
- package/cpp/llama.cpp/src/llama-hparams.cpp +53 -2
- package/cpp/llama.cpp/src/llama-hparams.h +48 -8
- package/cpp/llama.cpp/src/llama-impl.cpp +3 -3
- package/cpp/llama.cpp/src/llama-impl.h +2 -0
- package/cpp/llama.cpp/src/llama-kv-cache-iswa.cpp +13 -3
- package/cpp/llama.cpp/src/llama-kv-cache-iswa.h +2 -0
- package/cpp/llama.cpp/src/llama-kv-cache.cpp +120 -62
- package/cpp/llama.cpp/src/llama-kv-cache.h +13 -4
- package/cpp/llama.cpp/src/llama-kv-cells.h +44 -2
- package/cpp/llama.cpp/src/llama-memory-hybrid.cpp +19 -9
- package/cpp/llama.cpp/src/llama-memory-hybrid.h +2 -0
- package/cpp/llama.cpp/src/llama-memory-recurrent.cpp +38 -17
- package/cpp/llama.cpp/src/llama-memory-recurrent.h +5 -2
- package/cpp/llama.cpp/src/llama-memory.h +3 -0
- package/cpp/llama.cpp/src/llama-model-loader.cpp +2 -0
- package/cpp/llama.cpp/src/llama-model.cpp +1070 -12614
- package/cpp/llama.cpp/src/llama-model.h +40 -4
- package/cpp/llama.cpp/src/llama-quant.cpp +14 -6
- package/cpp/llama.cpp/src/llama-sampling.cpp +243 -136
- package/cpp/llama.cpp/src/llama-vocab.cpp +43 -3
- package/cpp/llama.cpp/src/llama-vocab.h +43 -39
- package/cpp/llama.cpp/src/llama.cpp +69 -10
- package/cpp/llama.cpp/src/models/afmoe.cpp +187 -0
- package/cpp/llama.cpp/src/models/apertus.cpp +125 -0
- package/cpp/llama.cpp/src/models/arcee.cpp +135 -0
- package/cpp/llama.cpp/src/models/arctic.cpp +138 -0
- package/cpp/llama.cpp/src/models/arwkv7.cpp +86 -0
- package/cpp/llama.cpp/src/models/baichuan.cpp +122 -0
- package/cpp/llama.cpp/src/models/bailingmoe.cpp +144 -0
- package/cpp/llama.cpp/src/models/bailingmoe2.cpp +135 -0
- package/cpp/llama.cpp/src/models/bert.cpp +176 -0
- package/cpp/llama.cpp/src/models/bitnet.cpp +160 -0
- package/cpp/llama.cpp/src/models/bloom.cpp +101 -0
- package/cpp/llama.cpp/src/models/chameleon.cpp +178 -0
- package/cpp/llama.cpp/src/models/chatglm.cpp +132 -0
- package/cpp/llama.cpp/src/models/codeshell.cpp +111 -0
- package/cpp/llama.cpp/src/models/cogvlm.cpp +100 -0
- package/cpp/llama.cpp/src/models/cohere2-iswa.cpp +131 -0
- package/cpp/llama.cpp/src/models/command-r.cpp +122 -0
- package/cpp/llama.cpp/src/models/dbrx.cpp +123 -0
- package/cpp/llama.cpp/src/models/deci.cpp +135 -0
- package/cpp/llama.cpp/src/models/deepseek.cpp +144 -0
- package/cpp/llama.cpp/src/models/deepseek2.cpp +237 -0
- package/cpp/llama.cpp/src/models/dots1.cpp +134 -0
- package/cpp/llama.cpp/src/models/dream.cpp +105 -0
- package/cpp/llama.cpp/src/models/ernie4-5-moe.cpp +150 -0
- package/cpp/llama.cpp/src/models/ernie4-5.cpp +110 -0
- package/cpp/llama.cpp/src/models/exaone.cpp +114 -0
- package/cpp/llama.cpp/src/models/exaone4.cpp +123 -0
- package/cpp/llama.cpp/src/models/falcon-h1.cpp +113 -0
- package/cpp/llama.cpp/src/models/falcon.cpp +120 -0
- package/cpp/llama.cpp/src/models/gemma-embedding.cpp +120 -0
- package/cpp/llama.cpp/src/models/gemma.cpp +112 -0
- package/cpp/llama.cpp/src/models/gemma2-iswa.cpp +125 -0
- package/cpp/llama.cpp/src/models/gemma3-iswa.cpp +131 -0
- package/cpp/llama.cpp/src/models/gemma3n-iswa.cpp +377 -0
- package/cpp/llama.cpp/src/models/glm4-moe.cpp +153 -0
- package/cpp/llama.cpp/src/models/glm4.cpp +127 -0
- package/cpp/llama.cpp/src/models/gpt2.cpp +105 -0
- package/cpp/llama.cpp/src/models/gptneox.cpp +144 -0
- package/cpp/llama.cpp/src/models/granite-hybrid.cpp +196 -0
- package/cpp/llama.cpp/src/models/granite.cpp +211 -0
- package/cpp/llama.cpp/src/models/graph-context-mamba.cpp +283 -0
- package/cpp/llama.cpp/src/models/grok.cpp +159 -0
- package/cpp/llama.cpp/src/models/grovemoe.cpp +141 -0
- package/cpp/llama.cpp/src/models/hunyuan-dense.cpp +132 -0
- package/cpp/llama.cpp/src/models/hunyuan-moe.cpp +154 -0
- package/cpp/llama.cpp/src/models/internlm2.cpp +120 -0
- package/cpp/llama.cpp/src/models/jais.cpp +86 -0
- package/cpp/llama.cpp/src/models/jamba.cpp +106 -0
- package/cpp/llama.cpp/src/models/lfm2.cpp +173 -0
- package/cpp/llama.cpp/src/models/llada-moe.cpp +122 -0
- package/cpp/llama.cpp/src/models/llada.cpp +99 -0
- package/cpp/llama.cpp/src/models/llama-iswa.cpp +174 -0
- package/cpp/llama.cpp/src/models/llama.cpp +155 -0
- package/cpp/llama.cpp/src/models/mamba.cpp +55 -0
- package/cpp/llama.cpp/src/models/minicpm3.cpp +199 -0
- package/cpp/llama.cpp/src/models/minimax-m2.cpp +124 -0
- package/cpp/llama.cpp/src/models/models.h +485 -0
- package/cpp/llama.cpp/src/models/mpt.cpp +126 -0
- package/cpp/llama.cpp/src/models/nemotron-h.cpp +121 -0
- package/cpp/llama.cpp/src/models/nemotron.cpp +122 -0
- package/cpp/llama.cpp/src/models/neo-bert.cpp +104 -0
- package/cpp/llama.cpp/src/models/olmo.cpp +121 -0
- package/cpp/llama.cpp/src/models/olmo2.cpp +150 -0
- package/cpp/llama.cpp/src/models/olmoe.cpp +124 -0
- package/cpp/llama.cpp/src/models/openai-moe-iswa.cpp +124 -0
- package/cpp/llama.cpp/src/models/openelm.cpp +124 -0
- package/cpp/llama.cpp/src/models/orion.cpp +123 -0
- package/cpp/llama.cpp/src/models/pangu-embedded.cpp +121 -0
- package/cpp/llama.cpp/src/models/phi2.cpp +121 -0
- package/cpp/llama.cpp/src/models/phi3.cpp +152 -0
- package/cpp/llama.cpp/src/models/plamo.cpp +110 -0
- package/cpp/llama.cpp/src/models/plamo2.cpp +316 -0
- package/cpp/llama.cpp/src/models/plm.cpp +168 -0
- package/cpp/llama.cpp/src/models/qwen.cpp +108 -0
- package/cpp/llama.cpp/src/models/qwen2.cpp +117 -0
- package/cpp/llama.cpp/src/models/qwen2moe.cpp +151 -0
- package/cpp/llama.cpp/src/models/qwen2vl.cpp +117 -0
- package/cpp/llama.cpp/src/models/qwen3.cpp +117 -0
- package/cpp/llama.cpp/src/models/qwen3moe.cpp +124 -0
- package/cpp/llama.cpp/src/models/qwen3vl-moe.cpp +149 -0
- package/cpp/llama.cpp/src/models/qwen3vl.cpp +141 -0
- package/cpp/llama.cpp/src/models/refact.cpp +94 -0
- package/cpp/llama.cpp/src/models/rwkv6-base.cpp +162 -0
- package/cpp/llama.cpp/src/models/rwkv6.cpp +94 -0
- package/cpp/llama.cpp/src/models/rwkv6qwen2.cpp +86 -0
- package/cpp/llama.cpp/src/models/rwkv7-base.cpp +135 -0
- package/cpp/llama.cpp/src/models/rwkv7.cpp +90 -0
- package/cpp/llama.cpp/src/models/seed-oss.cpp +124 -0
- package/cpp/llama.cpp/src/models/smallthinker.cpp +120 -0
- package/cpp/llama.cpp/src/models/smollm3.cpp +128 -0
- package/cpp/llama.cpp/src/models/stablelm.cpp +146 -0
- package/cpp/llama.cpp/src/models/starcoder.cpp +100 -0
- package/cpp/llama.cpp/src/models/starcoder2.cpp +121 -0
- package/cpp/llama.cpp/src/models/t5-dec.cpp +166 -0
- package/cpp/llama.cpp/src/models/t5-enc.cpp +96 -0
- package/cpp/llama.cpp/src/models/wavtokenizer-dec.cpp +149 -0
- package/cpp/llama.cpp/src/models/xverse.cpp +108 -0
- package/cpp/llama.cpp/src/unicode.cpp +77 -0
- package/cpp/llama.cpp/src/unicode.h +43 -0
- package/cpp/llama.cpp/vendor/cpp-httplib/CMakeLists.txt +94 -0
- package/cpp/llama.cpp/vendor/cpp-httplib/httplib.cpp +9339 -0
- package/cpp/llama.cpp/vendor/cpp-httplib/httplib.h +433 -8222
- package/cpp/llama.cpp/vendor/cpp-httplib/patch-boringssl.cmake +6 -0
- package/cpp/llama.cpp/vendor/miniaudio/miniaudio.h +4179 -1900
- package/cpp/llama.cpp/vendor/minja/chat-template.hpp +9 -2
- package/cpp/llama.cpp/vendor/minja/minja.hpp +101 -22
- package/ios/include/chat.h +16 -3
- package/ios/include/common/minja/chat-template.hpp +9 -2
- package/ios/include/common/minja/minja.hpp +101 -22
- package/ios/include/common.h +57 -19
- package/ios/include/json-schema-to-grammar.h +2 -0
- package/ios/include/llama.h +44 -21
- package/ios/include/log.h +12 -4
- package/ios/include/sampling.h +3 -1
- package/ios/libs/llama.xcframework/Info.plist +20 -20
- package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +6399 -5557
- package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-backend.h +19 -1
- package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-cpu.h +1 -1
- package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-metal.h +1 -6
- package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h +199 -6
- package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/llama.h +44 -21
- package/ios/libs/llama.xcframework/ios-arm64/llama.framework/llama +0 -0
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +6362 -5520
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +4813 -4241
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +19 -1
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-cpu.h +1 -1
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-metal.h +1 -6
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +199 -6
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h +44 -21
- package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama +0 -0
- package/package.json +10 -4
- package/cpp/llama.cpp/ggml/src/ggml-cann/Doxyfile +0 -2579
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f16.cu +0 -371
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f16.cuh +0 -3
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f32.cu +0 -379
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f32.cuh +0 -3
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f16.cuh +0 -495
- package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f32.cuh +0 -486
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +0 -5
- package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.m +0 -6886
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +0 -154
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +0 -105
- package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +0 -55
- package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +0 -60
- package/cpp/llama.cpp/ggml/src/ggml-zdnn/ggml-zdnn-impl.h +0 -97
- package/cpp/llama.cpp/models/ggml-vocab-aquila.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-baichuan.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-gpt-neox.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-nomic-bert-moe.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-refact.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.out +0 -46
- package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf +0 -0
- package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.inp +0 -112
- package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.out +0 -46
- package/cpp/llama.cpp/models/templates/ByteDance-Seed-OSS.jinja +0 -171
- package/cpp/llama.cpp/models/templates/CohereForAI-c4ai-command-r-plus-tool_use.jinja +0 -202
- package/cpp/llama.cpp/models/templates/CohereForAI-c4ai-command-r7b-12-2024-tool_use.jinja +0 -156
- package/cpp/llama.cpp/models/templates/Mistral-Small-3.2-24B-Instruct-2506.jinja +0 -124
- package/cpp/llama.cpp/models/templates/NousResearch-Hermes-2-Pro-Llama-3-8B-tool_use.jinja +0 -152
- package/cpp/llama.cpp/models/templates/NousResearch-Hermes-3-Llama-3.1-8B-tool_use.jinja +0 -152
- package/cpp/llama.cpp/models/templates/Qwen-QwQ-32B.jinja +0 -62
- package/cpp/llama.cpp/models/templates/Qwen-Qwen2.5-7B-Instruct.jinja +0 -54
- package/cpp/llama.cpp/models/templates/Qwen-Qwen3-0.6B.jinja +0 -85
- package/cpp/llama.cpp/models/templates/README.md +0 -25
- package/cpp/llama.cpp/models/templates/deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja +0 -1
- package/cpp/llama.cpp/models/templates/deepseek-ai-DeepSeek-R1-Distill-Qwen-32B.jinja +0 -1
- package/cpp/llama.cpp/models/templates/fireworks-ai-llama-3-firefunction-v2.jinja +0 -57
- package/cpp/llama.cpp/models/templates/google-gemma-2-2b-it.jinja +0 -4
- package/cpp/llama.cpp/models/templates/ibm-granite-granite-3.3-2B-Instruct.jinja +0 -59
- package/cpp/llama.cpp/models/templates/llama-cpp-deepseek-r1.jinja +0 -76
- package/cpp/llama.cpp/models/templates/llama-cpp-rwkv-world.jinja +0 -34
- package/cpp/llama.cpp/models/templates/meetkai-functionary-medium-v3.1.jinja +0 -58
- package/cpp/llama.cpp/models/templates/meetkai-functionary-medium-v3.2.jinja +0 -287
- package/cpp/llama.cpp/models/templates/meta-llama-Llama-3.1-8B-Instruct.jinja +0 -109
- package/cpp/llama.cpp/models/templates/meta-llama-Llama-3.2-3B-Instruct.jinja +0 -93
- package/cpp/llama.cpp/models/templates/meta-llama-Llama-3.3-70B-Instruct.jinja +0 -109
- package/cpp/llama.cpp/models/templates/microsoft-Phi-3.5-mini-instruct.jinja +0 -8
- package/cpp/llama.cpp/models/templates/mistralai-Mistral-Nemo-Instruct-2407.jinja +0 -87
- package/cpp/llama.cpp/models/templates/moonshotai-Kimi-K2.jinja +0 -43
- package/cpp/llama.cpp/models/templates/openai-gpt-oss-120b.jinja +0 -331
- package/cpp/llama.cpp/models/templates/unsloth-mistral-Devstral-Small-2507.jinja +0 -105
- package/cpp/llama.cpp/prompts/LLM-questions.txt +0 -49
- package/cpp/llama.cpp/prompts/alpaca.txt +0 -1
- package/cpp/llama.cpp/prompts/assistant.txt +0 -31
- package/cpp/llama.cpp/prompts/chat-with-baichuan.txt +0 -4
- package/cpp/llama.cpp/prompts/chat-with-bob.txt +0 -7
- package/cpp/llama.cpp/prompts/chat-with-qwen.txt +0 -1
- package/cpp/llama.cpp/prompts/chat-with-vicuna-v0.txt +0 -7
- package/cpp/llama.cpp/prompts/chat-with-vicuna-v1.txt +0 -7
- package/cpp/llama.cpp/prompts/chat.txt +0 -28
- package/cpp/llama.cpp/prompts/dan-modified.txt +0 -1
- package/cpp/llama.cpp/prompts/dan.txt +0 -1
- package/cpp/llama.cpp/prompts/mnemonics.txt +0 -93
- package/cpp/llama.cpp/prompts/parallel-questions.txt +0 -43
- package/cpp/llama.cpp/prompts/reason-act.txt +0 -18
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5524
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +0 -4247
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Resources/Info.plist +0 -32
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Resources/Info.plist +0 -32
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/llama +0 -0
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Resources/Info.plist +0 -32
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/llama +0 -0
- package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/llama +0 -0
- package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
- package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5561
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Info.plist +0 -35
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/llama +0 -0
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5524
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +0 -4246
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Info.plist +0 -35
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/llama +0 -0
- package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
- package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5558
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Info.plist +0 -32
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/xros-arm64/llama.framework/llama +0 -0
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5520
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +0 -4243
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-alloc.h +0 -76
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +0 -354
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-blas.h +0 -25
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-cpu.h +0 -145
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-metal.h +0 -66
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +0 -256
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +0 -2492
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/gguf.h +0 -202
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -1391
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Info.plist +0 -32
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Modules/module.modulemap +0 -17
- package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/llama +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_bfloat16_support.comp → feature-tests/bfloat16.comp} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat_support.comp → feature-tests/coopmat.comp} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat2_support.comp → feature-tests/coopmat2.comp} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_integer_dot_support.comp → feature-tests/integer_dot.comp} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{generic_head.comp → generic_head.glsl} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{generic_unary_head.comp → generic_unary_head.glsl} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{glu_main.comp → glu_main.glsl} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{rte.comp → rte.glsl} +0 -0
- /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{utils.comp → utils.glsl} +0 -0
|
@@ -10,25 +10,27 @@
|
|
|
10
10
|
#extension GL_EXT_shader_explicit_arithmetic_types_float16 : require
|
|
11
11
|
#endif
|
|
12
12
|
|
|
13
|
-
#
|
|
14
|
-
#extension GL_KHR_cooperative_matrix : enable
|
|
15
|
-
#extension GL_KHR_memory_scope_semantics : enable
|
|
13
|
+
#if defined(MUL_MAT_ID_USE_SUBGROUPS)
|
|
16
14
|
#extension GL_KHR_shader_subgroup_basic : enable
|
|
15
|
+
#extension GL_KHR_shader_subgroup_ballot : enable
|
|
17
16
|
#endif
|
|
18
17
|
|
|
19
18
|
#ifdef MUL_MAT_ID
|
|
20
19
|
#extension GL_EXT_shader_explicit_arithmetic_types_int16 : require
|
|
21
20
|
#endif
|
|
22
21
|
|
|
23
|
-
#include "types.
|
|
22
|
+
#include "types.glsl"
|
|
24
23
|
|
|
25
24
|
layout(local_size_x_id = 0, local_size_y = 1, local_size_z = 1) in;
|
|
26
25
|
|
|
27
|
-
layout (binding = 0) readonly buffer A {
|
|
26
|
+
layout (binding = 0) readonly buffer A {A_TYPE data_a[];};
|
|
27
|
+
#if defined(A_TYPE_PACKED16)
|
|
28
|
+
layout (binding = 0) readonly buffer A_PACKED16 {A_TYPE_PACKED16 data_a_packed16[];};
|
|
29
|
+
#endif
|
|
28
30
|
#if defined(A_TYPE_PACKED32)
|
|
29
31
|
layout (binding = 0) readonly buffer A_PACKED32 {A_TYPE_PACKED32 data_a_packed32[];};
|
|
30
32
|
#endif
|
|
31
|
-
layout (binding = 1) readonly buffer B {
|
|
33
|
+
layout (binding = 1) readonly buffer B {block_q8_1_x4_packed128 data_b[];};
|
|
32
34
|
layout (binding = 2) writeonly buffer D {D_TYPE data_d[];};
|
|
33
35
|
|
|
34
36
|
#ifdef MUL_MAT_ID
|
|
@@ -76,41 +78,32 @@ layout (constant_id = 10) const uint WARP = 32;
|
|
|
76
78
|
|
|
77
79
|
#define BK 32
|
|
78
80
|
|
|
79
|
-
#
|
|
80
|
-
#define SHMEM_STRIDE (BK / 4 + 4)
|
|
81
|
-
#else
|
|
82
|
-
#define SHMEM_STRIDE (BK / 4 + 1)
|
|
83
|
-
#endif
|
|
81
|
+
#define MMQ_SHMEM
|
|
84
82
|
|
|
85
|
-
|
|
83
|
+
#include "mul_mmq_shmem_types.glsl"
|
|
86
84
|
|
|
87
|
-
#
|
|
88
|
-
#
|
|
89
|
-
shared FLOAT_TYPE buf_a_dm[BM];
|
|
85
|
+
#ifdef MUL_MAT_ID
|
|
86
|
+
#define BK_STEP 1
|
|
90
87
|
#else
|
|
91
|
-
|
|
88
|
+
#ifndef BK_STEP
|
|
89
|
+
#define BK_STEP 4
|
|
92
90
|
#endif
|
|
93
91
|
#endif
|
|
94
92
|
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
shared
|
|
98
|
-
|
|
93
|
+
// Shared memory cache
|
|
94
|
+
shared block_a_cache buf_a[BM * BK_STEP];
|
|
95
|
+
shared block_b_cache buf_b[BN * BK_STEP];
|
|
96
|
+
// Register cache
|
|
97
|
+
block_a_cache cache_a[WMITER * TM];
|
|
98
|
+
block_b_cache cache_b;
|
|
99
99
|
|
|
100
|
-
#define LOAD_VEC_A (4 *
|
|
101
|
-
#define LOAD_VEC_B
|
|
102
|
-
|
|
103
|
-
#ifdef MUL_MAT_ID
|
|
104
|
-
shared u16vec2 row_ids[4096];
|
|
105
|
-
#endif // MUL_MAT_ID
|
|
100
|
+
#define LOAD_VEC_A (4 * QUANT_R_MMQ)
|
|
101
|
+
#define LOAD_VEC_B 16
|
|
106
102
|
|
|
107
103
|
#define NUM_WARPS (BLOCK_SIZE / WARP)
|
|
108
104
|
|
|
109
|
-
#
|
|
110
|
-
|
|
111
|
-
#endif
|
|
112
|
-
|
|
113
|
-
#include "mul_mmq_funcs.comp"
|
|
105
|
+
#include "mul_mm_id_funcs.glsl"
|
|
106
|
+
#include "mul_mmq_funcs.glsl"
|
|
114
107
|
|
|
115
108
|
void main() {
|
|
116
109
|
#ifdef NEEDS_INIT_IQ_SHMEM
|
|
@@ -139,26 +132,12 @@ void main() {
|
|
|
139
132
|
const uint WNITER = (WM * WN) / (WARP * TM * TN * WMITER);
|
|
140
133
|
const uint WSUBM = WM / WMITER;
|
|
141
134
|
const uint WSUBN = WN / WNITER;
|
|
142
|
-
|
|
143
|
-
#ifdef COOPMAT
|
|
144
|
-
const uint warp_i = gl_SubgroupID;
|
|
145
|
-
|
|
146
|
-
const uint tiw = gl_SubgroupInvocationID;
|
|
147
|
-
|
|
148
|
-
const uint cms_per_row = WM / TM;
|
|
149
|
-
const uint cms_per_col = WN / TN;
|
|
150
|
-
|
|
151
|
-
const uint storestride = WARP / TM;
|
|
152
|
-
const uint store_r = tiw % TM;
|
|
153
|
-
const uint store_c = tiw / TM;
|
|
154
|
-
#else
|
|
155
135
|
const uint warp_i = gl_LocalInvocationID.x / WARP;
|
|
156
136
|
|
|
157
137
|
const uint tiw = gl_LocalInvocationID.x % WARP;
|
|
158
138
|
|
|
159
139
|
const uint tiwr = tiw % (WSUBM / TM);
|
|
160
140
|
const uint tiwc = tiw / (WSUBM / TM);
|
|
161
|
-
#endif
|
|
162
141
|
|
|
163
142
|
const uint warp_r = warp_i % (BM / WM);
|
|
164
143
|
const uint warp_c = warp_i / (BM / WM);
|
|
@@ -172,17 +151,27 @@ void main() {
|
|
|
172
151
|
const uint loadstride_b = BLOCK_SIZE * LOAD_VEC_B / BK;
|
|
173
152
|
|
|
174
153
|
#ifdef MUL_MAT_ID
|
|
175
|
-
|
|
176
|
-
|
|
177
|
-
|
|
154
|
+
#ifdef MUL_MAT_ID_USE_SUBGROUPS
|
|
155
|
+
if (bitCount(p.nei0) == 1) {
|
|
156
|
+
load_row_ids(expert_idx, true, ic);
|
|
157
|
+
} else {
|
|
158
|
+
load_row_ids(expert_idx, false, ic);
|
|
159
|
+
}
|
|
160
|
+
#else
|
|
161
|
+
_ne1 = 0;
|
|
162
|
+
for (uint ii1 = 0; ii1 < p.nei1 && _ne1 < (ic + 1) * BN; ii1++) {
|
|
163
|
+
for (uint ii0 = 0; ii0 < p.nei0 && _ne1 < (ic + 1) * BN; ii0++) {
|
|
178
164
|
if (data_ids[ii1*p.nbi1 + ii0] == expert_idx) {
|
|
179
|
-
|
|
165
|
+
if (_ne1 >= ic * BN) {
|
|
166
|
+
row_ids[_ne1 - ic * BN] = u16vec2(ii0, ii1);
|
|
167
|
+
}
|
|
180
168
|
_ne1++;
|
|
181
169
|
}
|
|
182
170
|
}
|
|
183
171
|
}
|
|
184
172
|
|
|
185
173
|
barrier();
|
|
174
|
+
#endif
|
|
186
175
|
|
|
187
176
|
// Workgroup has no work
|
|
188
177
|
if (ic * BN >= _ne1) return;
|
|
@@ -209,152 +198,72 @@ void main() {
|
|
|
209
198
|
uint pos_b_ib = (batch_idx * p.batch_stride_b + ic * BN * p.stride_b + start_k) / BK;
|
|
210
199
|
#endif
|
|
211
200
|
|
|
212
|
-
#ifdef COOPMAT
|
|
213
|
-
coopmat<int8_t, gl_ScopeSubgroup, TM, TK, gl_MatrixUseA> cache_a;
|
|
214
|
-
coopmat<int8_t, gl_ScopeSubgroup, TK, TN, gl_MatrixUseB> cache_b;
|
|
215
|
-
coopmat<int32_t, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator> cm_result;
|
|
216
|
-
|
|
217
|
-
coopmat<ACC_TYPE, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator> factors[cms_per_row * cms_per_col];
|
|
218
|
-
|
|
219
|
-
coopmat<ACC_TYPE, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator> sums[cms_per_row * cms_per_col];
|
|
220
|
-
|
|
221
|
-
[[unroll]] for (uint i = 0; i < cms_per_row * cms_per_col; i++) {
|
|
222
|
-
sums[i] = coopmat<ACC_TYPE, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator>(0.0f);
|
|
223
|
-
}
|
|
224
|
-
#else
|
|
225
|
-
int32_t cache_a_qs[WMITER * TM * BK / 4];
|
|
226
|
-
|
|
227
|
-
int32_t cache_b_qs[TN * BK / 4];
|
|
228
|
-
|
|
229
201
|
ACC_TYPE sums[WMITER * TM * WNITER * TN];
|
|
230
202
|
|
|
231
203
|
[[unroll]] for (uint i = 0; i < WMITER*TM*WNITER*TN; i++) {
|
|
232
204
|
sums[i] = ACC_TYPE(0.0f);
|
|
233
205
|
}
|
|
234
|
-
#endif
|
|
235
206
|
|
|
236
|
-
|
|
237
|
-
FLOAT_TYPE cache_a_dm[WMITER * TM];
|
|
238
|
-
#else
|
|
239
|
-
FLOAT_TYPE_VEC2 cache_a_dm[WMITER * TM];
|
|
240
|
-
#endif
|
|
241
|
-
|
|
242
|
-
FLOAT_TYPE_VEC2 cache_b_ds[TN];
|
|
243
|
-
|
|
244
|
-
for (uint block = start_k; block < end_k; block += BK) {
|
|
207
|
+
for (uint block = start_k; block < end_k; block += BK * BK_STEP) {
|
|
245
208
|
[[unroll]] for (uint l = 0; loadc_a + l < BM; l += loadstride_a) {
|
|
246
|
-
const uint ib = pos_a_ib + (loadc_a + l) * p.stride_a / BK;
|
|
247
|
-
const uint iqs = loadr_a;
|
|
248
209
|
const uint buf_ib = loadc_a + l;
|
|
210
|
+
const uint ib = pos_a_ib + buf_ib * p.stride_a / BK;
|
|
211
|
+
const uint iqs = loadr_a;
|
|
249
212
|
|
|
250
|
-
|
|
251
|
-
|
|
252
|
-
|
|
253
|
-
|
|
254
|
-
buf_a_dm[buf_ib] = get_dm(ib);
|
|
255
|
-
#endif
|
|
213
|
+
[[unroll]] for (uint k_step = 0; k_step < BK_STEP; k_step++) {
|
|
214
|
+
if (block + k_step * BK < end_k) {
|
|
215
|
+
block_a_to_shmem(k_step * BM + buf_ib, ib + k_step, iqs);
|
|
216
|
+
}
|
|
256
217
|
}
|
|
257
|
-
#if QUANT_R == 1
|
|
258
|
-
buf_a_qs[buf_ib * SHMEM_STRIDE + iqs] = repack(ib, iqs);
|
|
259
|
-
#else
|
|
260
|
-
const i32vec2 vals = repack(ib, iqs);
|
|
261
|
-
buf_a_qs[buf_ib * SHMEM_STRIDE + iqs ] = vals.x;
|
|
262
|
-
buf_a_qs[buf_ib * SHMEM_STRIDE + iqs + 4] = vals.y;
|
|
263
|
-
#endif
|
|
264
218
|
}
|
|
265
219
|
[[unroll]] for (uint l = 0; loadc_b + l < BN; l += loadstride_b) {
|
|
220
|
+
const uint buf_ib = loadc_b + l;
|
|
221
|
+
|
|
266
222
|
#ifdef MUL_MAT_ID
|
|
267
|
-
const u16vec2 row_idx = row_ids[
|
|
268
|
-
const uint
|
|
269
|
-
const uint ib = idx / 8;
|
|
270
|
-
const uint iqs = idx & 0x7;
|
|
223
|
+
const u16vec2 row_idx = row_ids[buf_ib];
|
|
224
|
+
const uint ib = pos_b_ib + row_idx.y * p.batch_stride_b / BK + (row_idx.x % p.ne11) * p.stride_b / BK;
|
|
271
225
|
#else
|
|
272
|
-
const uint ib = pos_b_ib +
|
|
273
|
-
const uint iqs = loadr_b;
|
|
226
|
+
const uint ib = pos_b_ib + buf_ib * p.stride_b / BK;
|
|
274
227
|
#endif
|
|
228
|
+
const uint iqs = loadr_b;
|
|
275
229
|
|
|
276
|
-
|
|
277
|
-
|
|
278
|
-
if (iqs == 0) {
|
|
279
|
-
buf_b_ds[buf_ib] = FLOAT_TYPE_VEC2(data_b[ib].ds);
|
|
230
|
+
[[unroll]] for (uint k_step = 0; k_step < BK_STEP; k_step++) {
|
|
231
|
+
block_b_to_shmem(k_step * BN + buf_ib, ib + k_step, iqs, block + k_step * BK < end_k);
|
|
280
232
|
}
|
|
281
|
-
buf_b_qs[buf_ib * SHMEM_STRIDE + iqs] = data_b[ib].qs[iqs];
|
|
282
233
|
}
|
|
283
234
|
|
|
284
235
|
barrier();
|
|
285
236
|
|
|
286
|
-
pos_a_ib +=
|
|
287
|
-
pos_b_ib +=
|
|
237
|
+
pos_a_ib += BK_STEP;
|
|
238
|
+
pos_b_ib += BK_STEP;
|
|
288
239
|
|
|
289
|
-
|
|
290
|
-
[[unroll]] for (uint cm_row = 0; cm_row < cms_per_row; cm_row++) {
|
|
291
|
-
const uint ib_a = warp_r * WM + cm_row * TM;
|
|
240
|
+
for (uint k_step = 0; k_step < BK_STEP; k_step++) {
|
|
292
241
|
// Load from shared into cache
|
|
293
|
-
|
|
294
|
-
|
|
295
|
-
|
|
296
|
-
|
|
297
|
-
cache_a_dm[t_idx] = buf_a_dm[ib_a + t_idx];
|
|
298
|
-
}
|
|
299
|
-
|
|
300
|
-
[[unroll]] for (uint cm_col = 0; cm_col < cms_per_col; cm_col++) {
|
|
301
|
-
const uint ib_b = warp_c * WN + cm_col * TN;
|
|
302
|
-
coopMatLoad(cache_b, buf_b_qs, ib_b * SHMEM_STRIDE, SHMEM_STRIDE, gl_CooperativeMatrixLayoutColumnMajor);
|
|
303
|
-
|
|
304
|
-
// TODO: only cache values that are actually needed
|
|
305
|
-
[[unroll]] for (uint t_idx = 0; t_idx < TN; t_idx++) {
|
|
306
|
-
cache_b_dm[t_idx] = buf_b_d[ib_b + t_idx];
|
|
307
|
-
}
|
|
308
|
-
|
|
309
|
-
cm_result = coopmat<int32_t, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator>(0);
|
|
310
|
-
cm_result = coopMatMulAdd(cache_a, cache_b, cm_result);
|
|
311
|
-
|
|
312
|
-
[[unroll]] for (uint col = 0; col < TN; col += storestride) {
|
|
313
|
-
coopmat_stage[warp_i * TM * TN + (store_c + col) * TM + store_r] = ACC_TYPE(float(cache_a_d[store_r]) * float(cache_b_d[store_c + col]));
|
|
314
|
-
}
|
|
315
|
-
|
|
316
|
-
coopMatLoad(factors, coopmat_stage, warp_i * TM * TN, TM, gl_CooperativeMatrixLayoutColumnMajor);
|
|
317
|
-
sums[cm_col * cms_per_row + cm_row] += factors * coopmat<ACC_TYPE, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator>(cm_result);
|
|
318
|
-
}
|
|
319
|
-
}
|
|
320
|
-
#else
|
|
321
|
-
// Load from shared into cache
|
|
322
|
-
[[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
|
|
323
|
-
[[unroll]] for (uint cr = 0; cr < TM; cr++) {
|
|
324
|
-
const uint ib = warp_r * WM + wsir * WSUBM + tiwr * TM + cr;
|
|
325
|
-
cache_a_dm[wsir * TM + cr] = buf_a_dm[ib];
|
|
326
|
-
[[unroll]] for (uint idx_k = 0; idx_k < BK / 4; idx_k++) {
|
|
327
|
-
cache_a_qs[(wsir * TM + cr) * (BK / 4) + idx_k] = buf_a_qs[ib * SHMEM_STRIDE + idx_k];
|
|
328
|
-
}
|
|
329
|
-
}
|
|
330
|
-
}
|
|
242
|
+
[[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
|
|
243
|
+
[[unroll]] for (uint cr = 0; cr < TM; cr++) {
|
|
244
|
+
const uint reg_ib = wsir * TM + cr;
|
|
245
|
+
const uint buf_ib = warp_r * WM + wsir * WSUBM + tiwr * TM + cr;
|
|
331
246
|
|
|
332
|
-
|
|
333
|
-
[[unroll]] for (uint cc = 0; cc < TN; cc++) {
|
|
334
|
-
const uint ib = warp_c * WN + wsic * WSUBN + tiwc * TN + cc;
|
|
335
|
-
cache_b_ds[cc] = buf_b_ds[ib];
|
|
336
|
-
[[unroll]] for (uint idx_k = 0; idx_k < BK / 4; idx_k++) {
|
|
337
|
-
cache_b_qs[cc * (BK / 4) + idx_k] = buf_b_qs[ib * SHMEM_STRIDE + idx_k];
|
|
247
|
+
block_a_to_registers(reg_ib, k_step * BM + buf_ib);
|
|
338
248
|
}
|
|
339
249
|
}
|
|
340
250
|
|
|
341
|
-
[[unroll]] for (uint
|
|
251
|
+
[[unroll]] for (uint wsic = 0; wsic < WNITER; wsic++) {
|
|
342
252
|
[[unroll]] for (uint cc = 0; cc < TN; cc++) {
|
|
343
|
-
|
|
344
|
-
|
|
345
|
-
const uint sums_idx = (wsic * TN + cc) * (WMITER * TM) + wsir * TM + cr;
|
|
346
|
-
int32_t q_sum = 0;
|
|
347
|
-
[[unroll]] for (uint idx_k = 0; idx_k < BK / 4; idx_k++) {
|
|
348
|
-
q_sum += dotPacked4x8EXT(cache_a_qs[cache_a_idx * (BK / 4) + idx_k],
|
|
349
|
-
cache_b_qs[cc * (BK / 4) + idx_k]);
|
|
350
|
-
}
|
|
253
|
+
const uint ib = k_step * BN + warp_c * WN + wsic * WSUBN + tiwc * TN + cc;
|
|
254
|
+
block_b_to_registers(ib);
|
|
351
255
|
|
|
352
|
-
|
|
256
|
+
[[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
|
|
257
|
+
[[unroll]] for (uint cr = 0; cr < TM; cr++) {
|
|
258
|
+
const uint cache_a_idx = wsir * TM + cr;
|
|
259
|
+
const uint sums_idx = (wsic * TN + cc) * (WMITER * TM) + wsir * TM + cr;
|
|
260
|
+
|
|
261
|
+
sums[sums_idx] += mmq_dot_product(cache_a_idx);
|
|
262
|
+
}
|
|
353
263
|
}
|
|
354
264
|
}
|
|
355
265
|
}
|
|
356
266
|
}
|
|
357
|
-
#endif
|
|
358
267
|
|
|
359
268
|
barrier();
|
|
360
269
|
}
|
|
@@ -366,54 +275,6 @@ void main() {
|
|
|
366
275
|
const uint offsets = batch_idx * p.batch_stride_d + ik * p.batch_stride_d * gl_NumWorkGroups.z;
|
|
367
276
|
#endif
|
|
368
277
|
|
|
369
|
-
#ifdef COOPMAT
|
|
370
|
-
#ifdef MUL_MAT_ID
|
|
371
|
-
[[unroll]] for (uint cm_row = 0; cm_row < cms_per_row; cm_row++) {
|
|
372
|
-
[[unroll]] for (uint cm_col = 0; cm_col < cms_per_col; cm_col++) {
|
|
373
|
-
coopMatStore(sums[cm_col * cms_per_row + cm_row], coopmat_stage, warp_i * TM * TN, TM, gl_CooperativeMatrixLayoutColumnMajor);
|
|
374
|
-
|
|
375
|
-
[[unroll]] for (uint col = 0; col < BN; col += storestride) {
|
|
376
|
-
const uint row_i = dc + cm_col * TN + col + store_c;
|
|
377
|
-
if (row_i >= _ne1) break;
|
|
378
|
-
|
|
379
|
-
const u16vec2 row_idx = row_ids[row_i];
|
|
380
|
-
|
|
381
|
-
data_d[row_idx.y * p.batch_stride_d + row_idx.x * p.stride_d + dr + cm_row * TM + store_r] = D_TYPE(coopmat_stage[warp_i * TM * TN + (col + store_c) * TM + store_r]);
|
|
382
|
-
}
|
|
383
|
-
}
|
|
384
|
-
}
|
|
385
|
-
#else
|
|
386
|
-
const bool is_aligned = p.stride_d % 4 == 0; // Assumption: D_TYPE == float
|
|
387
|
-
|
|
388
|
-
[[unroll]] for (uint cm_row = 0; cm_row < cms_per_row; cm_row++) {
|
|
389
|
-
[[unroll]] for (uint cm_col = 0; cm_col < cms_per_col; cm_col++) {
|
|
390
|
-
const bool is_in_bounds = dr + (cm_row + 1) * TM <= p.M && dc + (cm_col + 1) * TN <= p.N;
|
|
391
|
-
|
|
392
|
-
if (is_aligned && is_in_bounds) {
|
|
393
|
-
// Full coopMat is within bounds and stride_d is aligned with 16B
|
|
394
|
-
coopmat<D_TYPE, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator> cm_dtype = coopmat<D_TYPE, gl_ScopeSubgroup, TM, TN, gl_MatrixUseAccumulator>(sums[cm_col * cms_per_row + cm_row]);
|
|
395
|
-
coopMatStore(cm_dtype, data_d, offsets + (dc + cm_col * TN) * p.stride_d + dr + cm_row * TM, p.stride_d, gl_CooperativeMatrixLayoutColumnMajor);
|
|
396
|
-
} else if (is_in_bounds) {
|
|
397
|
-
// Full coopMat is within bounds, but stride_d is not aligned
|
|
398
|
-
coopMatStore(sums[cm_col * cms_per_row + cm_row], coopmat_stage, warp_i * TM * TN, TM, gl_CooperativeMatrixLayoutColumnMajor);
|
|
399
|
-
|
|
400
|
-
[[unroll]] for (uint col = 0; col < TN; col += storestride) {
|
|
401
|
-
data_d[offsets + (dc + cm_col * TN + col + store_c) * p.stride_d + dr + cm_row * TM + store_r] = D_TYPE(coopmat_stage[warp_i * TM * TN + (col + store_c) * TM + store_r]);
|
|
402
|
-
}
|
|
403
|
-
} else if (dr + cm_row * TM < p.M && dc + cm_col * TN < p.N) {
|
|
404
|
-
// Partial coopMat is within bounds
|
|
405
|
-
coopMatStore(sums[cm_col * cms_per_row + cm_row], coopmat_stage, warp_i * TM * TN, TM, gl_CooperativeMatrixLayoutColumnMajor);
|
|
406
|
-
|
|
407
|
-
[[unroll]] for (uint col = 0; col < TN; col += storestride) {
|
|
408
|
-
if (dr + cm_row * TM + store_r < p.M && dc + cm_col * TN + col + store_c < p.N) {
|
|
409
|
-
data_d[offsets + (dc + cm_col * TN + col + store_c) * p.stride_d + dr + cm_row * TM + store_r] = D_TYPE(coopmat_stage[warp_i * TM * TN + (col + store_c) * TM + store_r]);
|
|
410
|
-
}
|
|
411
|
-
}
|
|
412
|
-
}
|
|
413
|
-
}
|
|
414
|
-
}
|
|
415
|
-
#endif // MUL_MAT_ID
|
|
416
|
-
#else
|
|
417
278
|
[[unroll]] for (uint wsic = 0; wsic < WNITER; wsic++) {
|
|
418
279
|
[[unroll]] for (uint wsir = 0; wsir < WMITER; wsir++) {
|
|
419
280
|
|
|
@@ -424,19 +285,21 @@ void main() {
|
|
|
424
285
|
const uint row_i = dc_warp + cc;
|
|
425
286
|
if (row_i >= _ne1) break;
|
|
426
287
|
|
|
427
|
-
const u16vec2 row_idx = row_ids[row_i];
|
|
288
|
+
const u16vec2 row_idx = row_ids[row_i - ic * BN];
|
|
428
289
|
#endif // MUL_MAT_ID
|
|
429
290
|
[[unroll]] for (uint cr = 0; cr < TM; cr++) {
|
|
291
|
+
const uint sums_idx = (wsic * TN + cc) * WMITER * TM + wsir * TM + cr;
|
|
430
292
|
#ifdef MUL_MAT_ID
|
|
431
|
-
|
|
293
|
+
if (dr_warp + cr < p.M) {
|
|
294
|
+
data_d[row_idx.y * p.batch_stride_d + row_idx.x * p.stride_d + dr_warp + cr] = D_TYPE(sums[sums_idx].x);
|
|
295
|
+
}
|
|
432
296
|
#else
|
|
433
297
|
if (dr_warp + cr < p.M && dc_warp + cc < p.N) {
|
|
434
|
-
data_d[offsets + (dc_warp + cc) * p.stride_d + dr_warp + cr] = D_TYPE(sums[
|
|
298
|
+
data_d[offsets + (dc_warp + cc) * p.stride_d + dr_warp + cr] = D_TYPE(sums[sums_idx].x);
|
|
435
299
|
}
|
|
436
300
|
#endif // MUL_MAT_ID
|
|
437
301
|
}
|
|
438
302
|
}
|
|
439
303
|
}
|
|
440
304
|
}
|
|
441
|
-
#endif // COOPMAT
|
|
442
305
|
}
|