whispercpp 1.3.6 → 1.3.7
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.
- checksums.yaml +4 -4
- data/.document +3 -0
- data/.rdoc_options +2 -0
- data/README.md +38 -5
- data/Rakefile +18 -3
- data/ext/dependencies.rb +10 -4
- data/ext/dependencies_for_windows.rb +17 -0
- data/ext/extconf.rb +20 -8
- data/ext/options.rb +54 -14
- data/ext/options_for_windows.rb +51 -0
- data/ext/ruby_whisper.c +36 -42
- data/ext/ruby_whisper.h +135 -0
- data/ext/ruby_whisper_context.c +107 -28
- data/ext/ruby_whisper_log_queue.c +180 -0
- data/ext/ruby_whisper_log_settable.h +47 -0
- data/ext/ruby_whisper_parakeet.c +49 -0
- data/ext/ruby_whisper_parakeet_context.c +304 -0
- data/ext/ruby_whisper_parakeet_context_params.c +117 -0
- data/ext/ruby_whisper_parakeet_model.c +84 -0
- data/ext/ruby_whisper_parakeet_params.c +548 -0
- data/ext/ruby_whisper_parakeet_segment.c +157 -0
- data/ext/ruby_whisper_parakeet_token.c +188 -0
- data/ext/ruby_whisper_parakeet_transcribe.cpp +58 -0
- data/ext/ruby_whisper_params.c +256 -65
- data/ext/ruby_whisper_segment.c +6 -6
- data/ext/ruby_whisper_transcribe.cpp +42 -15
- data/ext/sources/CMakeLists.txt +41 -3
- data/ext/sources/CMakePresets.json +95 -0
- data/ext/sources/cmake/parakeet-config.cmake.in +30 -0
- data/ext/sources/cmake/parakeet.pc.in +10 -0
- data/ext/sources/cmake/whisper.pc.in +1 -1
- data/ext/sources/examples/CMakeLists.txt +4 -2
- data/ext/sources/examples/bench/bench.cpp +1 -1
- data/ext/sources/examples/cli/cli.cpp +43 -9
- data/ext/sources/examples/common-ggml.cpp +2 -0
- data/ext/sources/examples/common-whisper.cpp +139 -67
- data/ext/sources/examples/common-whisper.h +11 -0
- data/ext/sources/examples/ffmpeg-transcode.cpp +211 -341
- data/ext/sources/examples/parakeet-cli/CMakeLists.txt +8 -0
- data/ext/sources/examples/parakeet-cli/parakeet-cli.cpp +243 -0
- data/ext/sources/examples/parakeet-quantize/CMakeLists.txt +7 -0
- data/ext/sources/examples/parakeet-quantize/parakeet-quantize.cpp +230 -0
- data/ext/sources/examples/server/server.cpp +199 -163
- data/ext/sources/ggml/CMakeLists.txt +21 -13
- data/ext/sources/ggml/cmake/FindNCCL.cmake +36 -0
- data/ext/sources/ggml/cmake/ggml-config.cmake.in +12 -2
- data/ext/sources/ggml/include/ggml-alloc.h +1 -0
- data/ext/sources/ggml/include/ggml-backend.h +72 -10
- data/ext/sources/ggml/include/ggml-cuda.h +3 -0
- data/ext/sources/ggml/include/ggml-rpc.h +3 -3
- data/ext/sources/ggml/include/ggml.h +101 -9
- data/ext/sources/ggml/include/gguf.h +10 -2
- data/ext/sources/ggml/src/CMakeLists.txt +22 -5
- data/ext/sources/ggml/src/ggml-alloc.c +5 -1
- data/ext/sources/ggml/src/ggml-backend-impl.h +22 -2
- data/ext/sources/ggml/src/ggml-backend-meta.cpp +2263 -0
- data/ext/sources/ggml/src/ggml-backend-reg.cpp +12 -0
- data/ext/sources/ggml/src/ggml-backend.cpp +110 -9
- data/ext/sources/ggml/src/ggml-blas/ggml-blas.cpp +4 -0
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.cpp +672 -257
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +71 -0
- data/ext/sources/ggml/src/ggml-cann/common.h +20 -10
- data/ext/sources/ggml/src/ggml-cann/ggml-cann.cpp +211 -30
- data/ext/sources/ggml/src/ggml-common.h +11 -0
- data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +58 -29
- data/ext/sources/ggml/src/ggml-cpu/amx/amx.cpp +2 -0
- data/ext/sources/ggml/src/ggml-cpu/amx/mmq.cpp +16 -16
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/quants.c +116 -7
- data/ext/sources/ggml/src/ggml-cpu/arch/arm/repack.cpp +65 -0
- data/ext/sources/ggml/src/ggml-cpu/arch/loongarch/quants.c +151 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/powerpc/quants.c +0 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/quants.c +4279 -1292
- data/ext/sources/ggml/src/ggml-cpu/arch/riscv/repack.cpp +5 -35
- data/ext/sources/ggml/src/ggml-cpu/arch/s390/quants.c +0 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/wasm/quants.c +72 -1
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/quants.c +177 -27
- data/ext/sources/ggml/src/ggml-cpu/arch/x86/repack.cpp +1 -1
- data/ext/sources/ggml/src/ggml-cpu/arch-fallback.h +5 -0
- data/ext/sources/ggml/src/ggml-cpu/cmake/FindSMTIME.cmake +32 -0
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-impl.h +10 -0
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +95 -5
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.cpp +2 -0
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +146 -134
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +88 -70
- data/ext/sources/ggml/src/ggml-cpu/ops.cpp +372 -73
- data/ext/sources/ggml/src/ggml-cpu/ops.h +3 -0
- data/ext/sources/ggml/src/ggml-cpu/quants.c +55 -0
- data/ext/sources/ggml/src/ggml-cpu/quants.h +3 -0
- data/ext/sources/ggml/src/ggml-cpu/repack.cpp +3 -0
- data/ext/sources/ggml/src/ggml-cpu/simd-gemm.h +90 -0
- data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +3 -16
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.cpp +1402 -687
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime.h +8 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +597 -2766
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime2_kernels.cpp +5768 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_env.cpp +320 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_env.h +55 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/ime_kernels.h +182 -19
- data/ext/sources/ggml/src/ggml-cpu/spacemit/repack.cpp +1795 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/repack.h +14 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/rvv_kernels.cpp +3178 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/rvv_kernels.h +95 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_barrier.h +34 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_mem_pool.cpp +760 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_mem_pool.h +32 -0
- data/ext/sources/ggml/src/ggml-cpu/spacemit/spine_tcm.h +409 -0
- data/ext/sources/ggml/src/ggml-cpu/vec.cpp +37 -53
- data/ext/sources/ggml/src/ggml-cpu/vec.h +225 -240
- data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +17 -7
- data/ext/sources/ggml/src/ggml-cuda/allreduce.cu +971 -0
- data/ext/sources/ggml/src/ggml-cuda/allreduce.cuh +29 -0
- data/ext/sources/ggml/src/ggml-cuda/argsort.cu +62 -26
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +44 -18
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +1 -0
- data/ext/sources/ggml/src/ggml-cuda/common.cuh +242 -28
- data/ext/sources/ggml/src/ggml-cuda/concat.cu +120 -114
- data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cu +45 -21
- data/ext/sources/ggml/src/ggml-cuda/conv2d-transpose.cuh +1 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cu +53 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cuh +10 -0
- data/ext/sources/ggml/src/ggml-cuda/cpy.cu +14 -6
- data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +22 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +278 -44
- data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +331 -130
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cu +12 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile.cuh +126 -27
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec.cuh +40 -15
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +18 -9
- data/ext/sources/ggml/src/ggml-cuda/fattn.cu +152 -49
- data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +2 -0
- data/ext/sources/ggml/src/ggml-cuda/fwht.cu +101 -0
- data/ext/sources/ggml/src/ggml-cuda/fwht.cuh +4 -0
- data/ext/sources/ggml/src/ggml-cuda/gated_delta_net.cu +84 -35
- data/ext/sources/ggml/src/ggml-cuda/getrows.cu +34 -12
- data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +1069 -609
- data/ext/sources/ggml/src/ggml-cuda/im2col.cu +32 -29
- data/ext/sources/ggml/src/ggml-cuda/mean.cu +4 -2
- data/ext/sources/ggml/src/ggml-cuda/mma.cuh +242 -195
- data/ext/sources/ggml/src/ggml-cuda/mmf.cuh +3 -3
- data/ext/sources/ggml/src/ggml-cuda/mmq.cu +18 -12
- data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +502 -423
- data/ext/sources/ggml/src/ggml-cuda/mmvf.cu +19 -12
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +485 -57
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +6 -1
- data/ext/sources/ggml/src/ggml-cuda/norm.cu +36 -10
- data/ext/sources/ggml/src/ggml-cuda/out-prod.cu +23 -7
- data/ext/sources/ggml/src/ggml-cuda/quantize.cu +133 -26
- data/ext/sources/ggml/src/ggml-cuda/quantize.cuh +1 -1
- data/ext/sources/ggml/src/ggml-cuda/reduce_rows.cuh +5 -1
- data/ext/sources/ggml/src/ggml-cuda/rope.cu +11 -4
- data/ext/sources/ggml/src/ggml-cuda/scale.cu +4 -1
- data/ext/sources/ggml/src/ggml-cuda/set-rows.cu +14 -6
- data/ext/sources/ggml/src/ggml-cuda/snake.cu +72 -0
- data/ext/sources/ggml/src/ggml-cuda/snake.cuh +8 -0
- data/ext/sources/ggml/src/ggml-cuda/softcap.cu +4 -1
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +45 -13
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cuh +1 -1
- data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +40 -18
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +8 -4
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_16.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_32.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_16.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_32.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_16.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +1 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_8.cu +2 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq192-dv128.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq320-dv256.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq512-dv512.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-f16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q4_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q5_1.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-bf16-q8_0.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-bf16.cu +7 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-nvfp4.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q1_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/top-k.cu +5 -4
- data/ext/sources/ggml/src/ggml-cuda/topk-moe.cu +26 -23
- data/ext/sources/ggml/src/ggml-cuda/unary.cu +31 -2
- data/ext/sources/ggml/src/ggml-cuda/unary.cuh +2 -0
- data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +80 -0
- data/ext/sources/ggml/src/ggml-cuda/vendors/cuda.h +7 -2
- data/ext/sources/ggml/src/ggml-cuda/vendors/hip.h +22 -4
- data/ext/sources/ggml/src/ggml-cuda/vendors/musa.h +3 -0
- data/ext/sources/ggml/src/ggml-hexagon/CMakeLists.txt +2 -1
- data/ext/sources/ggml/src/ggml-hexagon/ggml-hexagon.cpp +1428 -743
- data/ext/sources/ggml/src/ggml-hexagon/htp/CMakeLists.txt +45 -7
- data/ext/sources/ggml/src/ggml-hexagon/htp/act-ops.c +53 -84
- data/ext/sources/ggml/src/ggml-hexagon/htp/argsort-ops.c +25 -12
- data/ext/sources/ggml/src/ggml-hexagon/htp/binary-ops.c +165 -184
- data/ext/sources/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +5 -5
- data/ext/sources/ggml/src/ggml-hexagon/htp/concat-ops.c +277 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/cpy-ops.c +170 -127
- data/ext/sources/ggml/src/ggml-hexagon/htp/cumsum-ops.c +270 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/diag-ops.c +216 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/fill-ops.c +123 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/flash-attn-ops.c +125 -97
- data/ext/sources/ggml/src/ggml-hexagon/htp/gated-delta-net-ops.c +1148 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/get-rows-ops.c +148 -42
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dma.c +2 -2
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dma.h +252 -62
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-dump.h +9 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hex-utils.h +87 -1
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-flash-attn-ops.c +1878 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-matmul-ops.c +2066 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-ops.c +6 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-ops.h +88 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-profile.h +34 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-queue.c +158 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-queue.h +134 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hmx-utils.h +200 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ctx.h +96 -13
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-ops.h +182 -57
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp_iface.idl +9 -3
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-base.h +71 -3
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-copy.h +27 -10
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-div.h +63 -23
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-exp.h +9 -8
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-flash-attn.h +47 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-log.h +65 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-pow.h +42 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-repl.h +74 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sigmoid.h +1 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-sin-cos.h +90 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/hvx-utils.h +5 -8
- data/ext/sources/ggml/src/ggml-hexagon/htp/main.c +529 -815
- data/ext/sources/ggml/src/ggml-hexagon/htp/matmul-ops.c +2522 -234
- data/ext/sources/ggml/src/ggml-hexagon/htp/pad-ops.c +547 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/repeat-ops.c +148 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/rope-ops.c +291 -95
- data/ext/sources/ggml/src/ggml-hexagon/htp/set-rows-ops.c +59 -37
- data/ext/sources/ggml/src/ggml-hexagon/htp/softmax-ops.c +121 -133
- data/ext/sources/ggml/src/ggml-hexagon/htp/solve-tri-ops.c +267 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp/ssm-conv.c +244 -151
- data/ext/sources/ggml/src/ggml-hexagon/htp/sum-rows-ops.c +6 -6
- data/ext/sources/ggml/src/ggml-hexagon/htp/unary-ops.c +719 -45
- data/ext/sources/ggml/src/ggml-hexagon/htp/vtcm-utils.h +16 -0
- data/ext/sources/ggml/src/ggml-hexagon/htp-opnode.h +272 -0
- data/ext/sources/ggml/src/ggml-hexagon/libggml-htp.inf +3 -1
- data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +22 -9
- data/ext/sources/ggml/src/ggml-impl.h +6 -1
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.cpp +138 -13
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.h +32 -1
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-device.m +164 -28
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +80 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.cpp +190 -19
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-ops.h +2 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.cpp +39 -26
- data/ext/sources/ggml/src/ggml-metal/ggml-metal.metal +823 -322
- data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +5 -6
- data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +54 -5
- data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +12248 -5907
- data/ext/sources/ggml/src/ggml-opencl/kernels/concat.cl +67 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cpy.cl +59 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +1819 -112
- data/ext/sources/ggml/src/ggml-opencl/kernels/gated_delta_net.cl +249 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32_ns.cl +306 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_0_f32_ns.cl +256 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_1_f32_ns.cl +258 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q4_k_f32_ns.cl +283 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_0_f32_ns.cl +260 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_1_f32_ns.cl +262 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q5_k_f32_ns.cl +288 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_moe_q6_k_f32_ns.cl +267 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_iq4_nl_f32.cl +150 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mat_Ab_Bi_8x4.cl → gemm_noshuffle_q4_0_f32.cl} +1 -1
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q4_k_f32.cl +172 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_0_f32.cl +131 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_1_f32.cl +134 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q5_k_f32.cl +176 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_noshuffle_q6_k_f32.cl +140 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/{mul_mm_q8_0_f32_8x4.cl → gemm_noshuffle_q8_0_f32.cl} +1 -1
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemm_xmem_f16_f32_os8.cl +233 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32_ns.cl +165 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_0_f32_ns.cl +120 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_1_f32_ns.cl +123 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q4_k_f32_ns.cl +155 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_0_f32_ns.cl +123 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_1_f32_ns.cl +125 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q5_k_f32_ns.cl +160 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_moe_q6_k_f32_ns.cl +141 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_iq4_nl_f32.cl +302 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle_general.cl → gemv_noshuffle_q4_0_f32.cl} +5 -5
- data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle.cl → gemv_noshuffle_q4_0_f32_spec.cl} +5 -5
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q4_k_f32.cl +318 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_0_f32.cl +291 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_1_f32.cl +294 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q5_k_f32.cl +326 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_q6_k_f32.cl +293 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/get_rows.cl +15 -9
- data/ext/sources/ggml/src/ggml-opencl/kernels/moe_reorder_b.cl +30 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/moe_sort_by_expert.cl +82 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_iq4_nl_f32_l4_lm.cl +171 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q4_k_f32_l4_lm.cl +179 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_0_f32_l4_lm.cl +173 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_1_f32_l4_lm.cl +175 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mm_q5_k_f32_l4_lm.cl +192 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32.cl +164 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_iq4_nl_f32_flat.cl +202 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_k_f32_flat.cl +196 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_0_f32.cl +241 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_0_f32_flat.cl +243 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_1_f32.cl +243 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_1_f32_flat.cl +247 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32.cl +187 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q5_k_f32_flat.cl +203 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q6_k_f32_flat.cl +48 -64
- data/ext/sources/ggml/src/ggml-openvino/ggml-decoder.cpp +15 -5
- data/ext/sources/ggml/src/ggml-openvino/ggml-openvino-extra.cpp +18 -11
- data/ext/sources/ggml/src/ggml-openvino/ggml-openvino.cpp +35 -13
- data/ext/sources/ggml/src/ggml-openvino/ggml-quants.cpp +264 -192
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/rope.cpp +33 -7
- data/ext/sources/ggml/src/ggml-openvino/openvino/op/unary_gelu.cpp +25 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.cpp +1 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/op_table.h +1 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/rt_info/weightless_caching_attributes.hpp +41 -0
- data/ext/sources/ggml/src/ggml-openvino/openvino/translate_session.cpp +27 -3
- data/ext/sources/ggml/src/ggml-openvino/openvino/utils.cpp +67 -36
- data/ext/sources/ggml/src/ggml-openvino/openvino/utils.h +1 -0
- data/ext/sources/ggml/src/ggml-openvino/utils.cpp +101 -44
- data/ext/sources/ggml/src/ggml-openvino/utils.h +23 -3
- data/ext/sources/ggml/src/ggml-opt.cpp +1 -0
- data/ext/sources/ggml/src/ggml-quants.c +289 -114
- data/ext/sources/ggml/src/ggml-quants.h +3 -0
- data/ext/sources/ggml/src/ggml-rpc/CMakeLists.txt +24 -0
- data/ext/sources/ggml/src/ggml-rpc/ggml-rpc.cpp +167 -311
- data/ext/sources/ggml/src/ggml-rpc/transport.cpp +683 -0
- data/ext/sources/ggml/src/ggml-rpc/transport.h +34 -0
- data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +50 -4
- data/ext/sources/ggml/src/ggml-sycl/add-id.cpp +1 -1
- data/ext/sources/ggml/src/ggml-sycl/backend.hpp +3 -1
- data/ext/sources/ggml/src/ggml-sycl/common.cpp +74 -2
- data/ext/sources/ggml/src/ggml-sycl/common.hpp +41 -1
- data/ext/sources/ggml/src/ggml-sycl/convert.cpp +115 -13
- data/ext/sources/ggml/src/ggml-sycl/convert.hpp +9 -0
- data/ext/sources/ggml/src/ggml-sycl/cumsum.cpp +148 -0
- data/ext/sources/ggml/src/ggml-sycl/cumsum.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +663 -0
- data/ext/sources/ggml/src/ggml-sycl/diag.cpp +67 -0
- data/ext/sources/ggml/src/ggml-sycl/diag.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/dmmv.cpp +586 -6
- data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +1 -90
- data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +0 -2
- data/ext/sources/ggml/src/ggml-sycl/fattn-buffers.cpp +56 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-buffers.hpp +63 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-common.hpp +7 -5
- data/ext/sources/ggml/src/ggml-sycl/fattn-tile.cpp +4 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn-tile.hpp +76 -168
- data/ext/sources/ggml/src/ggml-sycl/fattn-vec.hpp +7 -0
- data/ext/sources/ggml/src/ggml-sycl/fattn.cpp +3 -1
- data/ext/sources/ggml/src/ggml-sycl/fill.cpp +55 -0
- data/ext/sources/ggml/src/ggml-sycl/fill.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.cpp +69 -31
- data/ext/sources/ggml/src/ggml-sycl/gated_delta_net.hpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +3 -0
- data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +79 -3
- data/ext/sources/ggml/src/ggml-sycl/ggml-sycl.cpp +823 -190
- data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +353 -89
- data/ext/sources/ggml/src/ggml-sycl/im2col.hpp +5 -3
- data/ext/sources/ggml/src/ggml-sycl/mmvq.cpp +1344 -26
- data/ext/sources/ggml/src/ggml-sycl/mmvq.hpp +16 -0
- data/ext/sources/ggml/src/ggml-sycl/pad.cpp +27 -27
- data/ext/sources/ggml/src/ggml-sycl/quants.hpp +71 -0
- data/ext/sources/ggml/src/ggml-sycl/set_rows.cpp +7 -1
- data/ext/sources/ggml/src/ggml-sycl/solve_tri.cpp +172 -0
- data/ext/sources/ggml/src/ggml-sycl/solve_tri.hpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/ssm_conv.cpp +6 -1
- data/ext/sources/ggml/src/ggml-sycl/ssm_scan.cpp +156 -0
- data/ext/sources/ggml/src/ggml-sycl/ssm_scan.hpp +5 -0
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.cpp +62 -10
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.hpp +18 -6
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-tile-instance-dkq512-dv512.cpp +6 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-f16.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q4_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q5_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-f16-q8_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-f16.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q4_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q5_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_0-q8_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-f16.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q4_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q5_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q4_1-q8_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-f16.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q4_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q5_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_0-q8_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-f16.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q4_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q5_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q5_1-q8_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-f16.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q4_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q5_1.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/template-instances/fattn-vec-instance-q8_0-q8_0.cpp +1 -0
- data/ext/sources/ggml/src/ggml-sycl/type.hpp +112 -0
- data/ext/sources/ggml/src/ggml-sycl/upscale.cpp +410 -0
- data/ext/sources/ggml/src/ggml-sycl/upscale.hpp +9 -0
- data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +215 -53
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-buffer.cpp +4 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend-device.cpp +2 -0
- data/ext/sources/ggml/src/ggml-virtgpu/ggml-backend.cpp +2 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu-shm.cpp +1 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.cpp +1 -0
- data/ext/sources/ggml/src/ggml-virtgpu/virtgpu.h +0 -2
- data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +11 -0
- data/ext/sources/ggml/src/ggml-vulkan/ggml-vulkan.cpp +2060 -535
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +4 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +6 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +146 -13
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +3 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +25 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.glsl +88 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.glsl +643 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_nvfp4.comp +32 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q1_0.comp +29 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dot_product_funcs.glsl +27 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/feature-tests/coopmat2_decode_vector.comp +7 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +197 -48
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.glsl +60 -59
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +115 -113
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +122 -31
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_dequant.glsl +131 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_mmq_funcs.glsl +203 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/fwht.comp +115 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gated_delta_net.comp +125 -64
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.glsl +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.glsl +10 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.glsl +16 -6
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +76 -54
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +122 -27
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +6 -6
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +1 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +1 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq_funcs.glsl +88 -55
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +11 -17
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +43 -10
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +159 -125
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +8 -8
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +24 -9
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +5 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +3 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/snake.comp +49 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +11 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tri.comp +0 -1
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.glsl +79 -2
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +171 -147
- data/ext/sources/ggml/src/ggml-webgpu/CMakeLists.txt +5 -2
- data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp +2202 -283
- data/ext/sources/ggml/src/ggml-webgpu/ggml-webgpu.cpp +2610 -1403
- data/ext/sources/ggml/src/ggml-webgpu/pre_wgsl.hpp +37 -7
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/add_id.wgsl +64 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/binary.wgsl +8 -7
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +76 -95
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/concat.wgsl +19 -1
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/conv2d.wgsl +165 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{cpy.tmpl.wgsl → cpy.wgsl} +25 -50
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn.wgsl +107 -184
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_quant_staging.tmpl +124 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_tile.wgsl +397 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_blk.wgsl +101 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_reduce.wgsl +84 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/flash_attn_vec_split.wgsl +619 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/gated_delta_net.wgsl +149 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.wgsl +183 -78
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.wgsl +155 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/im2col.wgsl +101 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +655 -495
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id.wgsl +195 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_gather.wgsl +52 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_id_vec.wgsl +154 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.wgsl +8 -6
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.wgsl +5 -1
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl +80 -409
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_acc.tmpl +1432 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec_q_acc.tmpl +303 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/quant_inner_loops.tmpl +21 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/quantize_q8.wgsl +173 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm_mul.wgsl +152 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{rope.tmpl.wgsl → rope.wgsl} +71 -142
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/row_norm.wgsl +153 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/scale.wgsl +6 -4
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set.wgsl +109 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows.wgsl +2 -3
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/set_rows_quant.wgsl +224 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/{soft_max.tmpl.wgsl → soft_max.wgsl} +106 -206
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/solve_tri.wgsl +121 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/ssm_conv.wgsl +65 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/ssm_scan.wgsl +193 -0
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/unary.wgsl +68 -48
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/upscale.wgsl +240 -0
- data/ext/sources/ggml/src/ggml-zdnn/ggml-zdnn.cpp +18 -14
- data/ext/sources/ggml/src/ggml-zendnn/CMakeLists.txt +1 -1
- data/ext/sources/ggml/src/ggml-zendnn/ggml-zendnn.cpp +244 -10
- data/ext/sources/ggml/src/ggml.c +110 -28
- data/ext/sources/ggml/src/gguf.cpp +173 -28
- data/ext/sources/include/parakeet.h +342 -0
- data/ext/sources/include/whisper.h +10 -0
- data/ext/sources/media/matmul.png +0 -0
- data/ext/sources/src/CMakeLists.txt +23 -0
- data/ext/sources/src/parakeet-arch.h +188 -0
- data/ext/sources/src/parakeet.cpp +3838 -0
- data/ext/sources/src/whisper.cpp +56 -12
- data/extsources.rb +26 -10
- data/lib/whisper/log_settable.rb +36 -0
- data/lib/whisper/model/uri.rb +13 -1
- data/lib/whisper/output.rb +74 -0
- data/sig/whisper.rbs +411 -62
- data/test/helper.rb +2 -0
- data/test/jfk_reader/jfk_reader.c +50 -7
- data/test/test_callback.rb +1 -0
- data/test/test_package.rb +6 -5
- data/test/test_parakeet.rb +28 -0
- data/test/test_parakeet_callback.rb +107 -0
- data/test/test_parakeet_context.rb +116 -0
- data/test/test_parakeet_context_params.rb +24 -0
- data/test/test_parakeet_model.rb +21 -0
- data/test/test_parakeet_params.rb +78 -0
- data/test/test_parakeet_segment.rb +42 -0
- data/test/test_parakeet_token.rb +73 -0
- data/test/test_params.rb +2 -0
- data/test/test_vad_segment.rb +1 -1
- data/test/test_whisper.rb +24 -6
- data/whispercpp.gemspec +2 -2
- metadata +215 -281
- data/ext/sources/bindings/javascript/CMakeLists.txt +0 -41
- data/ext/sources/bindings/javascript/emscripten.cpp +0 -93
- data/ext/sources/bindings/javascript/libwhisper.worker.js +0 -1
- data/ext/sources/bindings/javascript/package.json +0 -26
- data/ext/sources/bindings/javascript/whisper.js +0 -19
- data/ext/sources/examples/addon.node/CMakeLists.txt +0 -31
- data/ext/sources/examples/addon.node/__test__/whisper.spec.js +0 -133
- data/ext/sources/examples/addon.node/addon.cpp +0 -557
- data/ext/sources/examples/addon.node/index.js +0 -59
- data/ext/sources/examples/addon.node/package.json +0 -16
- data/ext/sources/examples/addon.node/vad-example.js +0 -132
- data/ext/sources/examples/bench.wasm/CMakeLists.txt +0 -49
- data/ext/sources/examples/bench.wasm/emscripten.cpp +0 -87
- data/ext/sources/examples/bench.wasm/index-tmpl.html +0 -285
- data/ext/sources/examples/coi-serviceworker.js +0 -146
- data/ext/sources/examples/command/CMakeLists.txt +0 -10
- data/ext/sources/examples/command/command.cpp +0 -802
- data/ext/sources/examples/command/commands.txt +0 -9
- data/ext/sources/examples/command.wasm/CMakeLists.txt +0 -50
- data/ext/sources/examples/command.wasm/emscripten.cpp +0 -327
- data/ext/sources/examples/command.wasm/index-tmpl.html +0 -415
- data/ext/sources/examples/generate-karaoke.sh +0 -57
- data/ext/sources/examples/helpers.js +0 -191
- data/ext/sources/examples/livestream.sh +0 -112
- data/ext/sources/examples/lsp/CMakeLists.txt +0 -10
- data/ext/sources/examples/lsp/lsp.cpp +0 -471
- data/ext/sources/examples/lsp/whisper.vim +0 -362
- data/ext/sources/examples/python/test_whisper_processor.py +0 -7
- data/ext/sources/examples/python/whisper_processor.py +0 -54
- data/ext/sources/examples/server/bench.js +0 -29
- data/ext/sources/examples/server.py +0 -120
- data/ext/sources/examples/stream/CMakeLists.txt +0 -10
- data/ext/sources/examples/stream/stream.cpp +0 -437
- data/ext/sources/examples/stream.wasm/CMakeLists.txt +0 -49
- data/ext/sources/examples/stream.wasm/emscripten.cpp +0 -216
- data/ext/sources/examples/stream.wasm/index-tmpl.html +0 -491
- data/ext/sources/examples/sycl/CMakeLists.txt +0 -9
- data/ext/sources/examples/sycl/build.sh +0 -22
- data/ext/sources/examples/sycl/ls-sycl-device.cpp +0 -11
- data/ext/sources/examples/sycl/run-whisper.sh +0 -17
- data/ext/sources/examples/talk-llama/CMakeLists.txt +0 -48
- data/ext/sources/examples/talk-llama/eleven-labs.py +0 -80
- data/ext/sources/examples/talk-llama/llama-adapter.cpp +0 -488
- data/ext/sources/examples/talk-llama/llama-adapter.h +0 -89
- data/ext/sources/examples/talk-llama/llama-arch.cpp +0 -2877
- data/ext/sources/examples/talk-llama/llama-arch.h +0 -628
- data/ext/sources/examples/talk-llama/llama-batch.cpp +0 -919
- data/ext/sources/examples/talk-llama/llama-batch.h +0 -173
- data/ext/sources/examples/talk-llama/llama-chat.cpp +0 -896
- data/ext/sources/examples/talk-llama/llama-chat.h +0 -71
- data/ext/sources/examples/talk-llama/llama-context.cpp +0 -3633
- data/ext/sources/examples/talk-llama/llama-context.h +0 -359
- data/ext/sources/examples/talk-llama/llama-cparams.cpp +0 -5
- data/ext/sources/examples/talk-llama/llama-cparams.h +0 -47
- data/ext/sources/examples/talk-llama/llama-ext.h +0 -12
- data/ext/sources/examples/talk-llama/llama-grammar.cpp +0 -1464
- data/ext/sources/examples/talk-llama/llama-grammar.h +0 -194
- data/ext/sources/examples/talk-llama/llama-graph.cpp +0 -2735
- data/ext/sources/examples/talk-llama/llama-graph.h +0 -1031
- data/ext/sources/examples/talk-llama/llama-hparams.cpp +0 -258
- data/ext/sources/examples/talk-llama/llama-hparams.h +0 -353
- data/ext/sources/examples/talk-llama/llama-impl.cpp +0 -171
- data/ext/sources/examples/talk-llama/llama-impl.h +0 -75
- data/ext/sources/examples/talk-llama/llama-io.cpp +0 -15
- data/ext/sources/examples/talk-llama/llama-io.h +0 -35
- data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.cpp +0 -330
- data/ext/sources/examples/talk-llama/llama-kv-cache-iswa.h +0 -137
- data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +0 -2285
- data/ext/sources/examples/talk-llama/llama-kv-cache.h +0 -389
- data/ext/sources/examples/talk-llama/llama-kv-cells.h +0 -533
- data/ext/sources/examples/talk-llama/llama-memory-hybrid-iswa.cpp +0 -275
- data/ext/sources/examples/talk-llama/llama-memory-hybrid-iswa.h +0 -140
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.cpp +0 -268
- data/ext/sources/examples/talk-llama/llama-memory-hybrid.h +0 -139
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.cpp +0 -1165
- data/ext/sources/examples/talk-llama/llama-memory-recurrent.h +0 -182
- data/ext/sources/examples/talk-llama/llama-memory.cpp +0 -59
- data/ext/sources/examples/talk-llama/llama-memory.h +0 -122
- data/ext/sources/examples/talk-llama/llama-mmap.cpp +0 -752
- data/ext/sources/examples/talk-llama/llama-mmap.h +0 -73
- data/ext/sources/examples/talk-llama/llama-model-loader.cpp +0 -1655
- data/ext/sources/examples/talk-llama/llama-model-loader.h +0 -206
- data/ext/sources/examples/talk-llama/llama-model-saver.cpp +0 -299
- data/ext/sources/examples/talk-llama/llama-model-saver.h +0 -40
- data/ext/sources/examples/talk-llama/llama-model.cpp +0 -9056
- data/ext/sources/examples/talk-llama/llama-model.h +0 -597
- data/ext/sources/examples/talk-llama/llama-quant.cpp +0 -1304
- data/ext/sources/examples/talk-llama/llama-quant.h +0 -1
- data/ext/sources/examples/talk-llama/llama-sampler.cpp +0 -3885
- data/ext/sources/examples/talk-llama/llama-sampler.h +0 -42
- data/ext/sources/examples/talk-llama/llama-vocab.cpp +0 -3970
- data/ext/sources/examples/talk-llama/llama-vocab.h +0 -187
- data/ext/sources/examples/talk-llama/llama.cpp +0 -1194
- data/ext/sources/examples/talk-llama/llama.h +0 -1573
- data/ext/sources/examples/talk-llama/models/afmoe.cpp +0 -190
- data/ext/sources/examples/talk-llama/models/apertus.cpp +0 -125
- data/ext/sources/examples/talk-llama/models/arcee.cpp +0 -135
- data/ext/sources/examples/talk-llama/models/arctic.cpp +0 -137
- data/ext/sources/examples/talk-llama/models/arwkv7.cpp +0 -86
- data/ext/sources/examples/talk-llama/models/baichuan.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/bailingmoe.cpp +0 -143
- data/ext/sources/examples/talk-llama/models/bailingmoe2.cpp +0 -133
- data/ext/sources/examples/talk-llama/models/bert.cpp +0 -184
- data/ext/sources/examples/talk-llama/models/bitnet.cpp +0 -145
- data/ext/sources/examples/talk-llama/models/bloom.cpp +0 -101
- data/ext/sources/examples/talk-llama/models/chameleon.cpp +0 -178
- data/ext/sources/examples/talk-llama/models/chatglm.cpp +0 -132
- data/ext/sources/examples/talk-llama/models/codeshell.cpp +0 -111
- data/ext/sources/examples/talk-llama/models/cogvlm.cpp +0 -102
- data/ext/sources/examples/talk-llama/models/cohere2-iswa.cpp +0 -134
- data/ext/sources/examples/talk-llama/models/command-r.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/dbrx.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/deci.cpp +0 -135
- data/ext/sources/examples/talk-llama/models/deepseek.cpp +0 -142
- data/ext/sources/examples/talk-llama/models/deepseek2.cpp +0 -262
- data/ext/sources/examples/talk-llama/models/delta-net-base.cpp +0 -445
- data/ext/sources/examples/talk-llama/models/dots1.cpp +0 -132
- data/ext/sources/examples/talk-llama/models/dream.cpp +0 -105
- data/ext/sources/examples/talk-llama/models/ernie4-5-moe.cpp +0 -148
- data/ext/sources/examples/talk-llama/models/ernie4-5.cpp +0 -110
- data/ext/sources/examples/talk-llama/models/eurobert.cpp +0 -97
- data/ext/sources/examples/talk-llama/models/exaone-moe.cpp +0 -145
- data/ext/sources/examples/talk-llama/models/exaone.cpp +0 -114
- data/ext/sources/examples/talk-llama/models/exaone4.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/falcon-h1.cpp +0 -111
- data/ext/sources/examples/talk-llama/models/falcon.cpp +0 -120
- data/ext/sources/examples/talk-llama/models/gemma-embedding.cpp +0 -116
- data/ext/sources/examples/talk-llama/models/gemma.cpp +0 -112
- data/ext/sources/examples/talk-llama/models/gemma2-iswa.cpp +0 -128
- data/ext/sources/examples/talk-llama/models/gemma3.cpp +0 -155
- data/ext/sources/examples/talk-llama/models/gemma3n-iswa.cpp +0 -384
- data/ext/sources/examples/talk-llama/models/glm4-moe.cpp +0 -170
- data/ext/sources/examples/talk-llama/models/glm4.cpp +0 -157
- data/ext/sources/examples/talk-llama/models/gpt2.cpp +0 -105
- data/ext/sources/examples/talk-llama/models/gptneox.cpp +0 -144
- data/ext/sources/examples/talk-llama/models/granite-hybrid.cpp +0 -195
- data/ext/sources/examples/talk-llama/models/granite.cpp +0 -210
- data/ext/sources/examples/talk-llama/models/grok.cpp +0 -159
- data/ext/sources/examples/talk-llama/models/grovemoe.cpp +0 -139
- data/ext/sources/examples/talk-llama/models/hunyuan-dense.cpp +0 -132
- data/ext/sources/examples/talk-llama/models/hunyuan-moe.cpp +0 -153
- data/ext/sources/examples/talk-llama/models/internlm2.cpp +0 -120
- data/ext/sources/examples/talk-llama/models/jais.cpp +0 -86
- data/ext/sources/examples/talk-llama/models/jais2.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/jamba.cpp +0 -106
- data/ext/sources/examples/talk-llama/models/kimi-linear.cpp +0 -381
- data/ext/sources/examples/talk-llama/models/lfm2.cpp +0 -196
- data/ext/sources/examples/talk-llama/models/llada-moe.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/llada.cpp +0 -99
- data/ext/sources/examples/talk-llama/models/llama-iswa.cpp +0 -178
- data/ext/sources/examples/talk-llama/models/llama.cpp +0 -175
- data/ext/sources/examples/talk-llama/models/maincoder.cpp +0 -117
- data/ext/sources/examples/talk-llama/models/mamba-base.cpp +0 -289
- data/ext/sources/examples/talk-llama/models/mamba.cpp +0 -54
- data/ext/sources/examples/talk-llama/models/mimo2-iswa.cpp +0 -129
- data/ext/sources/examples/talk-llama/models/minicpm3.cpp +0 -200
- data/ext/sources/examples/talk-llama/models/minimax-m2.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/mistral3.cpp +0 -160
- data/ext/sources/examples/talk-llama/models/models.h +0 -704
- data/ext/sources/examples/talk-llama/models/modern-bert.cpp +0 -109
- data/ext/sources/examples/talk-llama/models/mpt.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/nemotron-h.cpp +0 -162
- data/ext/sources/examples/talk-llama/models/nemotron.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/neo-bert.cpp +0 -104
- data/ext/sources/examples/talk-llama/models/olmo.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/olmo2.cpp +0 -150
- data/ext/sources/examples/talk-llama/models/olmoe.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/openai-moe-iswa.cpp +0 -127
- data/ext/sources/examples/talk-llama/models/openelm.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/orion.cpp +0 -123
- data/ext/sources/examples/talk-llama/models/paddleocr.cpp +0 -122
- data/ext/sources/examples/talk-llama/models/pangu-embedded.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/phi2.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/phi3.cpp +0 -152
- data/ext/sources/examples/talk-llama/models/plamo.cpp +0 -110
- data/ext/sources/examples/talk-llama/models/plamo2.cpp +0 -320
- data/ext/sources/examples/talk-llama/models/plamo3.cpp +0 -128
- data/ext/sources/examples/talk-llama/models/plm.cpp +0 -169
- data/ext/sources/examples/talk-llama/models/qwen.cpp +0 -108
- data/ext/sources/examples/talk-llama/models/qwen2.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/qwen2moe.cpp +0 -151
- data/ext/sources/examples/talk-llama/models/qwen2vl.cpp +0 -117
- data/ext/sources/examples/talk-llama/models/qwen3.cpp +0 -120
- data/ext/sources/examples/talk-llama/models/qwen35.cpp +0 -381
- data/ext/sources/examples/talk-llama/models/qwen35moe.cpp +0 -422
- data/ext/sources/examples/talk-llama/models/qwen3moe.cpp +0 -131
- data/ext/sources/examples/talk-llama/models/qwen3next.cpp +0 -525
- data/ext/sources/examples/talk-llama/models/qwen3vl-moe.cpp +0 -140
- data/ext/sources/examples/talk-llama/models/qwen3vl.cpp +0 -132
- data/ext/sources/examples/talk-llama/models/refact.cpp +0 -94
- data/ext/sources/examples/talk-llama/models/rnd1.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/rwkv6-base.cpp +0 -164
- data/ext/sources/examples/talk-llama/models/rwkv6.cpp +0 -94
- data/ext/sources/examples/talk-llama/models/rwkv6qwen2.cpp +0 -86
- data/ext/sources/examples/talk-llama/models/rwkv7-base.cpp +0 -137
- data/ext/sources/examples/talk-llama/models/rwkv7.cpp +0 -90
- data/ext/sources/examples/talk-llama/models/seed-oss.cpp +0 -124
- data/ext/sources/examples/talk-llama/models/smallthinker.cpp +0 -126
- data/ext/sources/examples/talk-llama/models/smollm3.cpp +0 -128
- data/ext/sources/examples/talk-llama/models/stablelm.cpp +0 -146
- data/ext/sources/examples/talk-llama/models/starcoder.cpp +0 -100
- data/ext/sources/examples/talk-llama/models/starcoder2.cpp +0 -121
- data/ext/sources/examples/talk-llama/models/step35-iswa.cpp +0 -165
- data/ext/sources/examples/talk-llama/models/t5-dec.cpp +0 -166
- data/ext/sources/examples/talk-llama/models/t5-enc.cpp +0 -96
- data/ext/sources/examples/talk-llama/models/wavtokenizer-dec.cpp +0 -149
- data/ext/sources/examples/talk-llama/models/xverse.cpp +0 -108
- data/ext/sources/examples/talk-llama/prompts/talk-alpaca.txt +0 -23
- data/ext/sources/examples/talk-llama/speak +0 -40
- data/ext/sources/examples/talk-llama/speak.bat +0 -1
- data/ext/sources/examples/talk-llama/speak.ps1 +0 -14
- data/ext/sources/examples/talk-llama/talk-llama.cpp +0 -813
- data/ext/sources/examples/talk-llama/unicode-data.cpp +0 -7034
- data/ext/sources/examples/talk-llama/unicode-data.h +0 -20
- data/ext/sources/examples/talk-llama/unicode.cpp +0 -1103
- data/ext/sources/examples/talk-llama/unicode.h +0 -111
- data/ext/sources/examples/wchess/CMakeLists.txt +0 -10
- data/ext/sources/examples/wchess/libwchess/CMakeLists.txt +0 -19
- data/ext/sources/examples/wchess/libwchess/Chessboard.cpp +0 -803
- data/ext/sources/examples/wchess/libwchess/Chessboard.h +0 -33
- data/ext/sources/examples/wchess/libwchess/WChess.cpp +0 -193
- data/ext/sources/examples/wchess/libwchess/WChess.h +0 -63
- data/ext/sources/examples/wchess/libwchess/test-chessboard.cpp +0 -117
- data/ext/sources/examples/wchess/wchess.cmd/CMakeLists.txt +0 -8
- data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +0 -253
- data/ext/sources/examples/whisper.wasm/CMakeLists.txt +0 -50
- data/ext/sources/examples/whisper.wasm/emscripten.cpp +0 -118
- data/ext/sources/examples/whisper.wasm/index-tmpl.html +0 -659
- data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +0 -99
- data/ext/sources/ggml/src/ggml-hexagon/htp/htp-msg.h +0 -155
- data/ext/sources/ggml/src/ggml-hexagon/op-desc.h +0 -153
- data/ext/sources/ggml/src/ggml-opencl/kernels/embed_kernel.py +0 -26
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.cpp +0 -123
- data/ext/sources/ggml/src/ggml-openvino/openvino/pass/eliminate_zp.h +0 -17
- data/ext/sources/ggml/src/ggml-virtgpu/regenerate_remoting.py +0 -333
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rte.glsl +0 -5
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +0 -182
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +0 -323
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.wgsl +0 -718
- data/ext/sources/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +0 -123
- data/ext/sources/tests/CMakeLists.txt +0 -112
- data/ext/sources/tests/earnings21/eval.mk +0 -58
- data/ext/sources/tests/earnings21/eval.py +0 -68
- data/ext/sources/tests/earnings21/normalizers/__init__.py +0 -2
- data/ext/sources/tests/earnings21/normalizers/basic.py +0 -80
- data/ext/sources/tests/earnings21/normalizers/english.json +0 -1741
- data/ext/sources/tests/earnings21/normalizers/english.py +0 -550
- data/ext/sources/tests/earnings21/requirements.txt +0 -6
- data/ext/sources/tests/en-0-ref.txt +0 -1
- data/ext/sources/tests/en-1-ref.txt +0 -1
- data/ext/sources/tests/en-2-ref.txt +0 -1
- data/ext/sources/tests/es-0-ref.txt +0 -1
- data/ext/sources/tests/librispeech/eval.mk +0 -39
- data/ext/sources/tests/librispeech/eval.py +0 -47
- data/ext/sources/tests/librispeech/normalizers/__init__.py +0 -2
- data/ext/sources/tests/librispeech/normalizers/basic.py +0 -80
- data/ext/sources/tests/librispeech/normalizers/english.json +0 -1741
- data/ext/sources/tests/librispeech/normalizers/english.py +0 -550
- data/ext/sources/tests/librispeech/requirements.txt +0 -6
- data/ext/sources/tests/run-tests.sh +0 -130
- data/ext/sources/tests/test-c.c +0 -3
- data/ext/sources/tests/test-vad-full.cpp +0 -56
- data/ext/sources/tests/test-vad.cpp +0 -83
- data/ext/sources/tests/test-whisper.js +0 -58
- data/lib/whisper/context.rb +0 -15
- data/lib/whisper/segment.rb +0 -58
- /data/ext/sources/ggml/src/ggml-opencl/kernels/{gemv_noshuffle_general_q8_0_f32.cl → gemv_noshuffle_q8_0_f32.cl} +0 -0
|
@@ -1,2877 +0,0 @@
|
|
|
1
|
-
#include "llama-arch.h"
|
|
2
|
-
|
|
3
|
-
#include "llama-impl.h"
|
|
4
|
-
|
|
5
|
-
#include <map>
|
|
6
|
-
#include <set>
|
|
7
|
-
#include <vector>
|
|
8
|
-
|
|
9
|
-
static const std::map<llm_arch, const char *> LLM_ARCH_NAMES = {
|
|
10
|
-
{ LLM_ARCH_CLIP, "clip" }, // dummy, only used by llama-quantize
|
|
11
|
-
{ LLM_ARCH_LLAMA, "llama" },
|
|
12
|
-
{ LLM_ARCH_LLAMA4, "llama4" },
|
|
13
|
-
{ LLM_ARCH_DECI, "deci" },
|
|
14
|
-
{ LLM_ARCH_FALCON, "falcon" },
|
|
15
|
-
{ LLM_ARCH_GROK, "grok" },
|
|
16
|
-
{ LLM_ARCH_GPT2, "gpt2" },
|
|
17
|
-
{ LLM_ARCH_GPTJ, "gptj" },
|
|
18
|
-
{ LLM_ARCH_GPTNEOX, "gptneox" },
|
|
19
|
-
{ LLM_ARCH_MPT, "mpt" },
|
|
20
|
-
{ LLM_ARCH_BAICHUAN, "baichuan" },
|
|
21
|
-
{ LLM_ARCH_STARCODER, "starcoder" },
|
|
22
|
-
{ LLM_ARCH_REFACT, "refact" },
|
|
23
|
-
{ LLM_ARCH_BERT, "bert" },
|
|
24
|
-
{ LLM_ARCH_MODERN_BERT, "modern-bert" },
|
|
25
|
-
{ LLM_ARCH_NOMIC_BERT, "nomic-bert" },
|
|
26
|
-
{ LLM_ARCH_NOMIC_BERT_MOE, "nomic-bert-moe" },
|
|
27
|
-
{ LLM_ARCH_NEO_BERT, "neo-bert" },
|
|
28
|
-
{ LLM_ARCH_JINA_BERT_V2, "jina-bert-v2" },
|
|
29
|
-
{ LLM_ARCH_JINA_BERT_V3, "jina-bert-v3" },
|
|
30
|
-
{ LLM_ARCH_EUROBERT, "eurobert" },
|
|
31
|
-
{ LLM_ARCH_BLOOM, "bloom" },
|
|
32
|
-
{ LLM_ARCH_STABLELM, "stablelm" },
|
|
33
|
-
{ LLM_ARCH_QWEN, "qwen" },
|
|
34
|
-
{ LLM_ARCH_QWEN2, "qwen2" },
|
|
35
|
-
{ LLM_ARCH_QWEN2MOE, "qwen2moe" },
|
|
36
|
-
{ LLM_ARCH_QWEN2VL, "qwen2vl" },
|
|
37
|
-
{ LLM_ARCH_QWEN3, "qwen3" },
|
|
38
|
-
{ LLM_ARCH_QWEN3MOE, "qwen3moe" },
|
|
39
|
-
{ LLM_ARCH_QWEN3NEXT, "qwen3next" },
|
|
40
|
-
{ LLM_ARCH_QWEN3VL, "qwen3vl" },
|
|
41
|
-
{ LLM_ARCH_QWEN3VLMOE, "qwen3vlmoe" },
|
|
42
|
-
{ LLM_ARCH_QWEN35, "qwen35" },
|
|
43
|
-
{ LLM_ARCH_QWEN35MOE, "qwen35moe" },
|
|
44
|
-
{ LLM_ARCH_PHI2, "phi2" },
|
|
45
|
-
{ LLM_ARCH_PHI3, "phi3" },
|
|
46
|
-
{ LLM_ARCH_PHIMOE, "phimoe" },
|
|
47
|
-
{ LLM_ARCH_PLAMO, "plamo" },
|
|
48
|
-
{ LLM_ARCH_PLAMO2, "plamo2" },
|
|
49
|
-
{ LLM_ARCH_PLAMO3, "plamo3" },
|
|
50
|
-
{ LLM_ARCH_CODESHELL, "codeshell" },
|
|
51
|
-
{ LLM_ARCH_ORION, "orion" },
|
|
52
|
-
{ LLM_ARCH_INTERNLM2, "internlm2" },
|
|
53
|
-
{ LLM_ARCH_MINICPM, "minicpm" },
|
|
54
|
-
{ LLM_ARCH_MINICPM3, "minicpm3" },
|
|
55
|
-
{ LLM_ARCH_GEMMA, "gemma" },
|
|
56
|
-
{ LLM_ARCH_GEMMA2, "gemma2" },
|
|
57
|
-
{ LLM_ARCH_GEMMA3, "gemma3" },
|
|
58
|
-
{ LLM_ARCH_GEMMA3N, "gemma3n" },
|
|
59
|
-
{ LLM_ARCH_GEMMA_EMBEDDING, "gemma-embedding" },
|
|
60
|
-
{ LLM_ARCH_STARCODER2, "starcoder2" },
|
|
61
|
-
{ LLM_ARCH_MAMBA, "mamba" },
|
|
62
|
-
{ LLM_ARCH_MAMBA2, "mamba2" },
|
|
63
|
-
{ LLM_ARCH_JAMBA, "jamba" },
|
|
64
|
-
{ LLM_ARCH_FALCON_H1, "falcon-h1" },
|
|
65
|
-
{ LLM_ARCH_XVERSE, "xverse" },
|
|
66
|
-
{ LLM_ARCH_COMMAND_R, "command-r" },
|
|
67
|
-
{ LLM_ARCH_COHERE2, "cohere2" },
|
|
68
|
-
{ LLM_ARCH_DBRX, "dbrx" },
|
|
69
|
-
{ LLM_ARCH_OLMO, "olmo" },
|
|
70
|
-
{ LLM_ARCH_OLMO2, "olmo2" },
|
|
71
|
-
{ LLM_ARCH_OLMOE, "olmoe" },
|
|
72
|
-
{ LLM_ARCH_OPENELM, "openelm" },
|
|
73
|
-
{ LLM_ARCH_ARCTIC, "arctic" },
|
|
74
|
-
{ LLM_ARCH_DEEPSEEK, "deepseek" },
|
|
75
|
-
{ LLM_ARCH_DEEPSEEK2, "deepseek2" },
|
|
76
|
-
{ LLM_ARCH_CHATGLM, "chatglm" },
|
|
77
|
-
{ LLM_ARCH_GLM4, "glm4" },
|
|
78
|
-
{ LLM_ARCH_GLM4_MOE, "glm4moe" },
|
|
79
|
-
{ LLM_ARCH_GLM_DSA, "glm-dsa" },
|
|
80
|
-
{ LLM_ARCH_BITNET, "bitnet" },
|
|
81
|
-
{ LLM_ARCH_T5, "t5" },
|
|
82
|
-
{ LLM_ARCH_T5ENCODER, "t5encoder" },
|
|
83
|
-
{ LLM_ARCH_JAIS, "jais" },
|
|
84
|
-
{ LLM_ARCH_JAIS2, "jais2" },
|
|
85
|
-
{ LLM_ARCH_NEMOTRON, "nemotron" },
|
|
86
|
-
{ LLM_ARCH_NEMOTRON_H, "nemotron_h" },
|
|
87
|
-
{ LLM_ARCH_NEMOTRON_H_MOE, "nemotron_h_moe" },
|
|
88
|
-
{ LLM_ARCH_EXAONE, "exaone" },
|
|
89
|
-
{ LLM_ARCH_EXAONE4, "exaone4" },
|
|
90
|
-
{ LLM_ARCH_EXAONE_MOE, "exaone-moe" },
|
|
91
|
-
{ LLM_ARCH_RWKV6, "rwkv6" },
|
|
92
|
-
{ LLM_ARCH_RWKV6QWEN2, "rwkv6qwen2" },
|
|
93
|
-
{ LLM_ARCH_RWKV7, "rwkv7" },
|
|
94
|
-
{ LLM_ARCH_ARWKV7, "arwkv7" },
|
|
95
|
-
{ LLM_ARCH_GRANITE, "granite" },
|
|
96
|
-
{ LLM_ARCH_GRANITE_MOE, "granitemoe" },
|
|
97
|
-
{ LLM_ARCH_GRANITE_HYBRID, "granitehybrid" },
|
|
98
|
-
{ LLM_ARCH_CHAMELEON, "chameleon" },
|
|
99
|
-
{ LLM_ARCH_WAVTOKENIZER_DEC, "wavtokenizer-dec" },
|
|
100
|
-
{ LLM_ARCH_PLM, "plm" },
|
|
101
|
-
{ LLM_ARCH_BAILINGMOE, "bailingmoe" },
|
|
102
|
-
{ LLM_ARCH_BAILINGMOE2, "bailingmoe2" },
|
|
103
|
-
{ LLM_ARCH_DOTS1, "dots1" },
|
|
104
|
-
{ LLM_ARCH_ARCEE, "arcee" },
|
|
105
|
-
{ LLM_ARCH_AFMOE, "afmoe" },
|
|
106
|
-
{ LLM_ARCH_ERNIE4_5, "ernie4_5" },
|
|
107
|
-
{ LLM_ARCH_ERNIE4_5_MOE, "ernie4_5-moe" },
|
|
108
|
-
{ LLM_ARCH_HUNYUAN_MOE, "hunyuan-moe" },
|
|
109
|
-
{ LLM_ARCH_HUNYUAN_DENSE, "hunyuan-dense" },
|
|
110
|
-
{ LLM_ARCH_SMOLLM3, "smollm3" },
|
|
111
|
-
{ LLM_ARCH_OPENAI_MOE, "gpt-oss" },
|
|
112
|
-
{ LLM_ARCH_LFM2, "lfm2" },
|
|
113
|
-
{ LLM_ARCH_LFM2MOE, "lfm2moe" },
|
|
114
|
-
{ LLM_ARCH_DREAM, "dream" },
|
|
115
|
-
{ LLM_ARCH_SMALLTHINKER, "smallthinker" },
|
|
116
|
-
{ LLM_ARCH_LLADA, "llada" },
|
|
117
|
-
{ LLM_ARCH_LLADA_MOE, "llada-moe" },
|
|
118
|
-
{ LLM_ARCH_SEED_OSS, "seed_oss" },
|
|
119
|
-
{ LLM_ARCH_GROVEMOE, "grovemoe" },
|
|
120
|
-
{ LLM_ARCH_APERTUS, "apertus" },
|
|
121
|
-
{ LLM_ARCH_MINIMAX_M2, "minimax-m2" },
|
|
122
|
-
{ LLM_ARCH_COGVLM, "cogvlm" },
|
|
123
|
-
{ LLM_ARCH_RND1, "rnd1" },
|
|
124
|
-
{ LLM_ARCH_PANGU_EMBED, "pangu-embedded" },
|
|
125
|
-
{ LLM_ARCH_MISTRAL3, "mistral3" },
|
|
126
|
-
{ LLM_ARCH_PADDLEOCR, "paddleocr" },
|
|
127
|
-
{ LLM_ARCH_MIMO2, "mimo2" },
|
|
128
|
-
{ LLM_ARCH_STEP35, "step35" },
|
|
129
|
-
{ LLM_ARCH_LLAMA_EMBED, "llama-embed" },
|
|
130
|
-
{ LLM_ARCH_MAINCODER, "maincoder" },
|
|
131
|
-
{ LLM_ARCH_KIMI_LINEAR, "kimi-linear" },
|
|
132
|
-
{ LLM_ARCH_UNKNOWN, "(unknown)" },
|
|
133
|
-
};
|
|
134
|
-
|
|
135
|
-
static const std::map<llm_kv, const char *> LLM_KV_NAMES = {
|
|
136
|
-
{ LLM_KV_GENERAL_TYPE, "general.type" },
|
|
137
|
-
{ LLM_KV_GENERAL_ARCHITECTURE, "general.architecture" },
|
|
138
|
-
{ LLM_KV_GENERAL_QUANTIZATION_VERSION, "general.quantization_version" },
|
|
139
|
-
{ LLM_KV_GENERAL_ALIGNMENT, "general.alignment" },
|
|
140
|
-
{ LLM_KV_GENERAL_FILE_TYPE, "general.file_type" },
|
|
141
|
-
{ LLM_KV_GENERAL_SAMPLING_SEQUENCE, "general.sampling.sequence" },
|
|
142
|
-
{ LLM_KV_GENERAL_SAMPLING_TOP_K, "general.sampling.top_k" },
|
|
143
|
-
{ LLM_KV_GENERAL_SAMPLING_TOP_P, "general.sampling.top_p" },
|
|
144
|
-
{ LLM_KV_GENERAL_SAMPLING_MIN_P, "general.sampling.min_p" },
|
|
145
|
-
{ LLM_KV_GENERAL_SAMPLING_XTC_PROBABILITY, "general.sampling.xtc_probability" },
|
|
146
|
-
{ LLM_KV_GENERAL_SAMPLING_XTC_THRESHOLD, "general.sampling.xtc_threshold" },
|
|
147
|
-
{ LLM_KV_GENERAL_SAMPLING_TEMP, "general.sampling.temp" },
|
|
148
|
-
{ LLM_KV_GENERAL_SAMPLING_PENALTY_LAST_N, "general.sampling.penalty_last_n" },
|
|
149
|
-
{ LLM_KV_GENERAL_SAMPLING_PENALTY_REPEAT, "general.sampling.penalty_repeat" },
|
|
150
|
-
{ LLM_KV_GENERAL_SAMPLING_MIROSTAT, "general.sampling.mirostat" },
|
|
151
|
-
{ LLM_KV_GENERAL_SAMPLING_MIROSTAT_TAU, "general.sampling.mirostat_tau" },
|
|
152
|
-
{ LLM_KV_GENERAL_SAMPLING_MIROSTAT_ETA, "general.sampling.mirostat_eta" },
|
|
153
|
-
{ LLM_KV_GENERAL_NAME, "general.name" },
|
|
154
|
-
{ LLM_KV_GENERAL_AUTHOR, "general.author" },
|
|
155
|
-
{ LLM_KV_GENERAL_VERSION, "general.version" },
|
|
156
|
-
{ LLM_KV_GENERAL_URL, "general.url" },
|
|
157
|
-
{ LLM_KV_GENERAL_DESCRIPTION, "general.description" },
|
|
158
|
-
{ LLM_KV_GENERAL_LICENSE, "general.license" },
|
|
159
|
-
{ LLM_KV_GENERAL_SOURCE_URL, "general.source.url" },
|
|
160
|
-
{ LLM_KV_GENERAL_SOURCE_HF_REPO, "general.source.huggingface.repository" },
|
|
161
|
-
|
|
162
|
-
{ LLM_KV_VOCAB_SIZE, "%s.vocab_size" },
|
|
163
|
-
{ LLM_KV_CONTEXT_LENGTH, "%s.context_length" },
|
|
164
|
-
{ LLM_KV_EMBEDDING_LENGTH, "%s.embedding_length" },
|
|
165
|
-
{ LLM_KV_EMBEDDING_LENGTH_OUT, "%s.embedding_length_out" },
|
|
166
|
-
{ LLM_KV_FEATURES_LENGTH, "%s.features_length" },
|
|
167
|
-
{ LLM_KV_BLOCK_COUNT, "%s.block_count" },
|
|
168
|
-
{ LLM_KV_LEADING_DENSE_BLOCK_COUNT, "%s.leading_dense_block_count" },
|
|
169
|
-
{ LLM_KV_FEED_FORWARD_LENGTH, "%s.feed_forward_length" },
|
|
170
|
-
{ LLM_KV_EXPERT_FEED_FORWARD_LENGTH, "%s.expert_feed_forward_length" },
|
|
171
|
-
{ LLM_KV_EXPERT_SHARED_FEED_FORWARD_LENGTH, "%s.expert_shared_feed_forward_length" },
|
|
172
|
-
{ LLM_KV_EXPERT_CHUNK_FEED_FORWARD_LENGTH, "%s.expert_chunk_feed_forward_length" },
|
|
173
|
-
{ LLM_KV_SWIGLU_CLAMP_EXP, "%s.swiglu_clamp_exp" },
|
|
174
|
-
{ LLM_KV_SWIGLU_CLAMP_SHEXP, "%s.swiglu_clamp_shexp" },
|
|
175
|
-
{ LLM_KV_USE_PARALLEL_RESIDUAL, "%s.use_parallel_residual" },
|
|
176
|
-
{ LLM_KV_TENSOR_DATA_LAYOUT, "%s.tensor_data_layout" },
|
|
177
|
-
{ LLM_KV_EXPERT_COUNT, "%s.expert_count" },
|
|
178
|
-
{ LLM_KV_EXPERT_USED_COUNT, "%s.expert_used_count" },
|
|
179
|
-
{ LLM_KV_EXPERT_SHARED_COUNT, "%s.expert_shared_count" },
|
|
180
|
-
{ LLM_KV_EXPERT_GROUP_COUNT, "%s.expert_group_count" },
|
|
181
|
-
{ LLM_KV_EXPERT_GROUP_USED_COUNT, "%s.expert_group_used_count" },
|
|
182
|
-
{ LLM_KV_EXPERT_WEIGHTS_SCALE, "%s.expert_weights_scale" },
|
|
183
|
-
{ LLM_KV_EXPERT_WEIGHTS_NORM, "%s.expert_weights_norm" },
|
|
184
|
-
{ LLM_KV_EXPERT_GATING_FUNC, "%s.expert_gating_func" },
|
|
185
|
-
{ LLM_KV_EXPERT_GROUP_SCALE, "%s.expert_group_scale" },
|
|
186
|
-
{ LLM_KV_EXPERTS_PER_GROUP, "%s.experts_per_group" },
|
|
187
|
-
{ LLM_KV_MOE_EVERY_N_LAYERS, "%s.moe_every_n_layers" },
|
|
188
|
-
{ LLM_KV_MOE_LATENT_SIZE, "%s.moe_latent_size" },
|
|
189
|
-
{ LLM_KV_NEXTN_PREDICT_LAYERS, "%s.nextn_predict_layers" },
|
|
190
|
-
{ LLM_KV_NUM_DEEPSTACK_LAYERS, "%s.n_deepstack_layers" },
|
|
191
|
-
{ LLM_KV_POOLING_TYPE, "%s.pooling_type" },
|
|
192
|
-
{ LLM_KV_LOGIT_SCALE, "%s.logit_scale" },
|
|
193
|
-
{ LLM_KV_DECODER_START_TOKEN_ID, "%s.decoder_start_token_id" },
|
|
194
|
-
{ LLM_KV_DECODER_BLOCK_COUNT, "%s.decoder_block_count" },
|
|
195
|
-
{ LLM_KV_ATTN_LOGIT_SOFTCAPPING, "%s.attn_logit_softcapping" },
|
|
196
|
-
{ LLM_KV_ROUTER_LOGIT_SOFTCAPPING, "%s.router_logit_softcapping" },
|
|
197
|
-
{ LLM_KV_FINAL_LOGIT_SOFTCAPPING, "%s.final_logit_softcapping" },
|
|
198
|
-
{ LLM_KV_SWIN_NORM, "%s.swin_norm" },
|
|
199
|
-
{ LLM_KV_RESCALE_EVERY_N_LAYERS, "%s.rescale_every_n_layers" },
|
|
200
|
-
{ LLM_KV_TIME_MIX_EXTRA_DIM, "%s.time_mix_extra_dim" },
|
|
201
|
-
{ LLM_KV_TIME_DECAY_EXTRA_DIM, "%s.time_decay_extra_dim" },
|
|
202
|
-
{ LLM_KV_RESIDUAL_SCALE, "%s.residual_scale" },
|
|
203
|
-
{ LLM_KV_EMBEDDING_SCALE, "%s.embedding_scale" },
|
|
204
|
-
{ LLM_KV_TOKEN_SHIFT_COUNT, "%s.token_shift_count" },
|
|
205
|
-
{ LLM_KV_INTERLEAVE_MOE_LAYER_STEP, "%s.interleave_moe_layer_step" },
|
|
206
|
-
{ LLM_KV_FULL_ATTENTION_INTERVAL, "%s.full_attention_interval" },
|
|
207
|
-
|
|
208
|
-
{ LLM_KV_ATTENTION_HEAD_COUNT, "%s.attention.head_count" },
|
|
209
|
-
{ LLM_KV_ATTENTION_HEAD_COUNT_KV, "%s.attention.head_count_kv" },
|
|
210
|
-
{ LLM_KV_ATTENTION_MAX_ALIBI_BIAS, "%s.attention.max_alibi_bias" },
|
|
211
|
-
{ LLM_KV_ATTENTION_CLAMP_KQV, "%s.attention.clamp_kqv" },
|
|
212
|
-
{ LLM_KV_ATTENTION_KEY_LENGTH, "%s.attention.key_length" },
|
|
213
|
-
{ LLM_KV_ATTENTION_VALUE_LENGTH, "%s.attention.value_length" },
|
|
214
|
-
{ LLM_KV_ATTENTION_LAYERNORM_EPS, "%s.attention.layer_norm_epsilon" },
|
|
215
|
-
{ LLM_KV_ATTENTION_LAYERNORM_RMS_EPS, "%s.attention.layer_norm_rms_epsilon" },
|
|
216
|
-
{ LLM_KV_ATTENTION_GROUPNORM_EPS, "%s.attention.group_norm_epsilon" },
|
|
217
|
-
{ LLM_KV_ATTENTION_GROUPNORM_GROUPS, "%s.attention.group_norm_groups" },
|
|
218
|
-
{ LLM_KV_ATTENTION_CAUSAL, "%s.attention.causal" },
|
|
219
|
-
{ LLM_KV_ATTENTION_Q_LORA_RANK, "%s.attention.q_lora_rank" },
|
|
220
|
-
{ LLM_KV_ATTENTION_KV_LORA_RANK, "%s.attention.kv_lora_rank" },
|
|
221
|
-
{ LLM_KV_ATTENTION_DECAY_LORA_RANK, "%s.attention.decay_lora_rank" },
|
|
222
|
-
{ LLM_KV_ATTENTION_ICLR_LORA_RANK, "%s.attention.iclr_lora_rank" },
|
|
223
|
-
{ LLM_KV_ATTENTION_VALUE_RESIDUAL_MIX_LORA_RANK, "%s.attention.value_residual_mix_lora_rank" },
|
|
224
|
-
{ LLM_KV_ATTENTION_GATE_LORA_RANK, "%s.attention.gate_lora_rank" },
|
|
225
|
-
{ LLM_KV_ATTENTION_RELATIVE_BUCKETS_COUNT, "%s.attention.relative_buckets_count" },
|
|
226
|
-
{ LLM_KV_ATTENTION_SLIDING_WINDOW, "%s.attention.sliding_window" },
|
|
227
|
-
{ LLM_KV_ATTENTION_SLIDING_WINDOW_PATTERN, "%s.attention.sliding_window_pattern" },
|
|
228
|
-
{ LLM_KV_ATTENTION_SCALE, "%s.attention.scale" },
|
|
229
|
-
{ LLM_KV_ATTENTION_OUTPUT_SCALE, "%s.attention.output_scale" },
|
|
230
|
-
{ LLM_KV_ATTENTION_TEMPERATURE_LENGTH, "%s.attention.temperature_length" },
|
|
231
|
-
{ LLM_KV_ATTENTION_TEMPERATURE_SCALE, "%s.attention.temperature_scale" },
|
|
232
|
-
{ LLM_KV_ATTENTION_KEY_LENGTH_MLA, "%s.attention.key_length_mla" },
|
|
233
|
-
{ LLM_KV_ATTENTION_VALUE_LENGTH_MLA, "%s.attention.value_length_mla" },
|
|
234
|
-
{ LLM_KV_ATTENTION_KEY_LENGTH_SWA, "%s.attention.key_length_swa" },
|
|
235
|
-
{ LLM_KV_ATTENTION_VALUE_LENGTH_SWA, "%s.attention.value_length_swa" },
|
|
236
|
-
{ LLM_KV_ATTENTION_INDEXER_HEAD_COUNT, "%s.attention.indexer.head_count" },
|
|
237
|
-
{ LLM_KV_ATTENTION_INDEXER_KEY_LENGTH, "%s.attention.indexer.key_length" },
|
|
238
|
-
{ LLM_KV_ATTENTION_INDEXER_TOP_K, "%s.attention.indexer.top_k" },
|
|
239
|
-
|
|
240
|
-
{ LLM_KV_ROPE_DIMENSION_COUNT, "%s.rope.dimension_count" },
|
|
241
|
-
{ LLM_KV_ROPE_DIMENSION_COUNT_SWA, "%s.rope.dimension_count_swa" },
|
|
242
|
-
{ LLM_KV_ROPE_DIMENSION_SECTIONS, "%s.rope.dimension_sections" },
|
|
243
|
-
{ LLM_KV_ROPE_FREQ_BASE, "%s.rope.freq_base" },
|
|
244
|
-
{ LLM_KV_ROPE_FREQ_BASE_SWA, "%s.rope.freq_base_swa" },
|
|
245
|
-
{ LLM_KV_ROPE_SCALE_LINEAR, "%s.rope.scale_linear" },
|
|
246
|
-
{ LLM_KV_ROPE_SCALING_TYPE, "%s.rope.scaling.type" },
|
|
247
|
-
{ LLM_KV_ROPE_SCALING_FACTOR, "%s.rope.scaling.factor" },
|
|
248
|
-
{ LLM_KV_ROPE_SCALING_ATTN_FACTOR, "%s.rope.scaling.attn_factor" },
|
|
249
|
-
{ LLM_KV_ROPE_SCALING_ORIG_CTX_LEN, "%s.rope.scaling.original_context_length" },
|
|
250
|
-
{ LLM_KV_ROPE_SCALING_FINETUNED, "%s.rope.scaling.finetuned" },
|
|
251
|
-
{ LLM_KV_ROPE_SCALING_YARN_LOG_MUL, "%s.rope.scaling.yarn_log_multiplier" },
|
|
252
|
-
{ LLM_KV_ROPE_SCALING_YARN_EXT_FACTOR, "%s.rope.scaling.yarn_ext_factor" },
|
|
253
|
-
{ LLM_KV_ROPE_SCALING_YARN_ATTN_FACTOR, "%s.rope.scaling.yarn_attn_factor" },
|
|
254
|
-
{ LLM_KV_ROPE_SCALING_YARN_BETA_FAST, "%s.rope.scaling.yarn_beta_fast" },
|
|
255
|
-
{ LLM_KV_ROPE_SCALING_YARN_BETA_SLOW, "%s.rope.scaling.yarn_beta_slow" },
|
|
256
|
-
|
|
257
|
-
{ LLM_KV_SPLIT_NO, "split.no" },
|
|
258
|
-
{ LLM_KV_SPLIT_COUNT, "split.count" },
|
|
259
|
-
{ LLM_KV_SPLIT_TENSORS_COUNT, "split.tensors.count" },
|
|
260
|
-
|
|
261
|
-
{ LLM_KV_SSM_CONV_KERNEL, "%s.ssm.conv_kernel" },
|
|
262
|
-
{ LLM_KV_SSM_INNER_SIZE, "%s.ssm.inner_size" },
|
|
263
|
-
{ LLM_KV_SSM_STATE_SIZE, "%s.ssm.state_size" },
|
|
264
|
-
{ LLM_KV_SSM_TIME_STEP_RANK, "%s.ssm.time_step_rank" },
|
|
265
|
-
{ LLM_KV_SSM_GROUP_COUNT, "%s.ssm.group_count" },
|
|
266
|
-
{ LLM_KV_SSM_DT_B_C_RMS, "%s.ssm.dt_b_c_rms" },
|
|
267
|
-
|
|
268
|
-
{ LLM_KV_KDA_HEAD_DIM, "%s.kda.head_dim" },
|
|
269
|
-
|
|
270
|
-
{ LLM_KV_WKV_HEAD_SIZE, "%s.wkv.head_size" },
|
|
271
|
-
|
|
272
|
-
{ LLM_KV_POSNET_EMBEDDING_LENGTH, "%s.posnet.embedding_length" },
|
|
273
|
-
{ LLM_KV_POSNET_BLOCK_COUNT, "%s.posnet.block_count" },
|
|
274
|
-
|
|
275
|
-
{ LLM_KV_CONVNEXT_EMBEDDING_LENGTH, "%s.convnext.embedding_length" },
|
|
276
|
-
{ LLM_KV_CONVNEXT_BLOCK_COUNT, "%s.convnext.block_count" },
|
|
277
|
-
|
|
278
|
-
{ LLM_KV_CLASSIFIER_OUTPUT_LABELS, "%s.classifier.output_labels" },
|
|
279
|
-
|
|
280
|
-
{ LLM_KV_SHORTCONV_L_CACHE, "%s.shortconv.l_cache" },
|
|
281
|
-
// sentence-transformers dense modules feature dims
|
|
282
|
-
{ LLM_KV_DENSE_2_FEAT_IN, "%s.dense_2_feat_in" },
|
|
283
|
-
{ LLM_KV_DENSE_2_FEAT_OUT, "%s.dense_2_feat_out" },
|
|
284
|
-
{ LLM_KV_DENSE_3_FEAT_IN, "%s.dense_3_feat_in" },
|
|
285
|
-
{ LLM_KV_DENSE_3_FEAT_OUT, "%s.dense_3_feat_out" },
|
|
286
|
-
|
|
287
|
-
{ LLM_KV_TOKENIZER_MODEL, "tokenizer.ggml.model" },
|
|
288
|
-
{ LLM_KV_TOKENIZER_PRE, "tokenizer.ggml.pre" },
|
|
289
|
-
{ LLM_KV_TOKENIZER_LIST, "tokenizer.ggml.tokens" },
|
|
290
|
-
{ LLM_KV_TOKENIZER_TOKEN_TYPE, "tokenizer.ggml.token_type" },
|
|
291
|
-
{ LLM_KV_TOKENIZER_TOKEN_TYPE_COUNT, "tokenizer.ggml.token_type_count" },
|
|
292
|
-
{ LLM_KV_TOKENIZER_SCORES, "tokenizer.ggml.scores" },
|
|
293
|
-
{ LLM_KV_TOKENIZER_MERGES, "tokenizer.ggml.merges" },
|
|
294
|
-
{ LLM_KV_TOKENIZER_BOS_ID, "tokenizer.ggml.bos_token_id" },
|
|
295
|
-
{ LLM_KV_TOKENIZER_EOS_ID, "tokenizer.ggml.eos_token_id" },
|
|
296
|
-
{ LLM_KV_TOKENIZER_EOT_ID, "tokenizer.ggml.eot_token_id" },
|
|
297
|
-
{ LLM_KV_TOKENIZER_EOM_ID, "tokenizer.ggml.eom_token_id" },
|
|
298
|
-
{ LLM_KV_TOKENIZER_UNK_ID, "tokenizer.ggml.unknown_token_id" },
|
|
299
|
-
{ LLM_KV_TOKENIZER_SEP_ID, "tokenizer.ggml.seperator_token_id" },
|
|
300
|
-
{ LLM_KV_TOKENIZER_PAD_ID, "tokenizer.ggml.padding_token_id" },
|
|
301
|
-
{ LLM_KV_TOKENIZER_CLS_ID, "tokenizer.ggml.cls_token_id" },
|
|
302
|
-
{ LLM_KV_TOKENIZER_MASK_ID, "tokenizer.ggml.mask_token_id" },
|
|
303
|
-
{ LLM_KV_TOKENIZER_ADD_BOS, "tokenizer.ggml.add_bos_token" },
|
|
304
|
-
{ LLM_KV_TOKENIZER_ADD_EOS, "tokenizer.ggml.add_eos_token" },
|
|
305
|
-
{ LLM_KV_TOKENIZER_ADD_SEP, "tokenizer.ggml.add_sep_token" },
|
|
306
|
-
{ LLM_KV_TOKENIZER_ADD_PREFIX, "tokenizer.ggml.add_space_prefix" },
|
|
307
|
-
{ LLM_KV_TOKENIZER_REMOVE_EXTRA_WS, "tokenizer.ggml.remove_extra_whitespaces" },
|
|
308
|
-
{ LLM_KV_TOKENIZER_PRECOMPILED_CHARSMAP, "tokenizer.ggml.precompiled_charsmap" },
|
|
309
|
-
{ LLM_KV_TOKENIZER_HF_JSON, "tokenizer.huggingface.json" },
|
|
310
|
-
{ LLM_KV_TOKENIZER_RWKV, "tokenizer.rwkv.world" },
|
|
311
|
-
{ LLM_KV_TOKENIZER_CHAT_TEMPLATE, "tokenizer.chat_template" },
|
|
312
|
-
{ LLM_KV_TOKENIZER_FIM_PRE_ID, "tokenizer.ggml.fim_pre_token_id" },
|
|
313
|
-
{ LLM_KV_TOKENIZER_FIM_SUF_ID, "tokenizer.ggml.fim_suf_token_id" },
|
|
314
|
-
{ LLM_KV_TOKENIZER_FIM_MID_ID, "tokenizer.ggml.fim_mid_token_id" },
|
|
315
|
-
{ LLM_KV_TOKENIZER_FIM_PAD_ID, "tokenizer.ggml.fim_pad_token_id" },
|
|
316
|
-
{ LLM_KV_TOKENIZER_FIM_REP_ID, "tokenizer.ggml.fim_rep_token_id" },
|
|
317
|
-
{ LLM_KV_TOKENIZER_FIM_SEP_ID, "tokenizer.ggml.fim_sep_token_id" },
|
|
318
|
-
|
|
319
|
-
{ LLM_KV_ADAPTER_TYPE, "adapter.type" },
|
|
320
|
-
{ LLM_KV_ADAPTER_LORA_ALPHA, "adapter.lora.alpha" },
|
|
321
|
-
{ LLM_KV_ADAPTER_LORA_TASK_NAME, "adapter.lora.task_name" },
|
|
322
|
-
{ LLM_KV_ADAPTER_LORA_PROMPT_PREFIX, "adapter.lora.prompt_prefix" },
|
|
323
|
-
{ LLM_KV_ADAPTER_ALORA_INVOCATION_TOKENS, "adapter.alora.invocation_tokens" },
|
|
324
|
-
|
|
325
|
-
{ LLM_KV_XIELU_ALPHA_N, "xielu.alpha_n" },
|
|
326
|
-
{ LLM_KV_XIELU_ALPHA_P, "xielu.alpha_p" },
|
|
327
|
-
{ LLM_KV_XIELU_BETA, "xielu.beta" },
|
|
328
|
-
{ LLM_KV_XIELU_EPS, "xielu.eps" },
|
|
329
|
-
|
|
330
|
-
// deprecated
|
|
331
|
-
{ LLM_KV_TOKENIZER_PREFIX_ID, "tokenizer.ggml.prefix_token_id" },
|
|
332
|
-
{ LLM_KV_TOKENIZER_SUFFIX_ID, "tokenizer.ggml.suffix_token_id" },
|
|
333
|
-
{ LLM_KV_TOKENIZER_MIDDLE_ID, "tokenizer.ggml.middle_token_id" },
|
|
334
|
-
};
|
|
335
|
-
|
|
336
|
-
static const std::map<llm_tensor, const char *> LLM_TENSOR_NAMES = {
|
|
337
|
-
{ LLM_TENSOR_TOKEN_EMBD, "token_embd" },
|
|
338
|
-
{ LLM_TENSOR_OUTPUT_NORM, "output_norm" },
|
|
339
|
-
{ LLM_TENSOR_OUTPUT_NORM_LFM2, "token_embd_norm" }, // fix for wrong tensor name
|
|
340
|
-
{ LLM_TENSOR_OUTPUT, "output" },
|
|
341
|
-
{ LLM_TENSOR_ROPE_FREQS, "rope_freqs" },
|
|
342
|
-
{ LLM_TENSOR_ATTN_NORM, "blk.%d.attn_norm" },
|
|
343
|
-
{ LLM_TENSOR_ATTN_Q, "blk.%d.attn_q" },
|
|
344
|
-
{ LLM_TENSOR_ATTN_K, "blk.%d.attn_k" },
|
|
345
|
-
{ LLM_TENSOR_ATTN_V, "blk.%d.attn_v" },
|
|
346
|
-
{ LLM_TENSOR_ATTN_OUT, "blk.%d.attn_output" },
|
|
347
|
-
{ LLM_TENSOR_ATTN_ROT_EMBD, "blk.%d.attn_rot_embd" },
|
|
348
|
-
{ LLM_TENSOR_FFN_GATE_INP, "blk.%d.ffn_gate_inp" },
|
|
349
|
-
{ LLM_TENSOR_FFN_NORM, "blk.%d.ffn_norm" },
|
|
350
|
-
{ LLM_TENSOR_FFN_GATE, "blk.%d.ffn_gate" },
|
|
351
|
-
{ LLM_TENSOR_FFN_DOWN, "blk.%d.ffn_down" },
|
|
352
|
-
{ LLM_TENSOR_FFN_UP, "blk.%d.ffn_up" },
|
|
353
|
-
{ LLM_TENSOR_FFN_GATE_EXP, "blk.%d.ffn_gate.%d" },
|
|
354
|
-
{ LLM_TENSOR_FFN_DOWN_EXP, "blk.%d.ffn_down.%d" },
|
|
355
|
-
{ LLM_TENSOR_FFN_UP_EXP, "blk.%d.ffn_up.%d" },
|
|
356
|
-
{ LLM_TENSOR_FFN_GATE_EXPS, "blk.%d.ffn_gate_exps" },
|
|
357
|
-
{ LLM_TENSOR_FFN_GATE_UP_EXPS, "blk.%d.ffn_gate_up_exps" },
|
|
358
|
-
{ LLM_TENSOR_FFN_DOWN_EXPS, "blk.%d.ffn_down_exps" },
|
|
359
|
-
{ LLM_TENSOR_FFN_UP_EXPS, "blk.%d.ffn_up_exps" },
|
|
360
|
-
{ LLM_TENSOR_ATTN_POST_NORM, "blk.%d.post_attention_norm" },
|
|
361
|
-
{ LLM_TENSOR_ATTN_Q_NORM, "blk.%d.attn_q_norm" },
|
|
362
|
-
{ LLM_TENSOR_ATTN_K_NORM, "blk.%d.attn_k_norm" },
|
|
363
|
-
{ LLM_TENSOR_ATTN_GATE, "blk.%d.attn_gate" },
|
|
364
|
-
{ LLM_TENSOR_FFN_POST_NORM, "blk.%d.post_ffw_norm" },
|
|
365
|
-
{ LLM_TENSOR_FFN_GATE_SHEXP, "blk.%d.ffn_gate_shexp" },
|
|
366
|
-
{ LLM_TENSOR_FFN_UP_SHEXP, "blk.%d.ffn_up_shexp" },
|
|
367
|
-
{ LLM_TENSOR_FFN_DOWN_SHEXP, "blk.%d.ffn_down_shexp" },
|
|
368
|
-
{ LLM_TENSOR_FFN_EXP_PROBS_B, "blk.%d.exp_probs_b" },
|
|
369
|
-
{ LLM_TENSOR_FFN_LATENT_DOWN, "blk.%d.ffn_latent_down" },
|
|
370
|
-
{ LLM_TENSOR_FFN_LATENT_UP, "blk.%d.ffn_latent_up" },
|
|
371
|
-
{ LLM_TENSOR_ATTN_NORM_2, "blk.%d.attn_norm_2" },
|
|
372
|
-
{ LLM_TENSOR_ATTN_QKV, "blk.%d.attn_qkv" },
|
|
373
|
-
{ LLM_TENSOR_LAYER_OUT_NORM, "blk.%d.layer_output_norm" },
|
|
374
|
-
{ LLM_TENSOR_ATTN_OUT_NORM, "blk.%d.attn_output_norm" },
|
|
375
|
-
{ LLM_TENSOR_POS_EMBD, "position_embd" },
|
|
376
|
-
{ LLM_TENSOR_FFN_ACT, "blk.%d.ffn.act" },
|
|
377
|
-
{ LLM_TENSOR_TOKEN_EMBD_NORM, "token_embd_norm" },
|
|
378
|
-
{ LLM_TENSOR_TOKEN_TYPES, "token_types" },
|
|
379
|
-
{ LLM_TENSOR_CLS, "cls" },
|
|
380
|
-
{ LLM_TENSOR_CLS_OUT, "cls.output" },
|
|
381
|
-
{ LLM_TENSOR_CLS_NORM, "cls.norm" },
|
|
382
|
-
{ LLM_TENSOR_ENC_OUTPUT_NORM, "enc.output_norm" },
|
|
383
|
-
{ LLM_TENSOR_FFN_GATE_INP_SHEXP, "blk.%d.ffn_gate_inp_shexp" },
|
|
384
|
-
{ LLM_TENSOR_SSM_A_NOSCAN, "blk.%d.ssm_a" },
|
|
385
|
-
{ LLM_TENSOR_SSM_CONV1D, "blk.%d.ssm_conv1d" },
|
|
386
|
-
{ LLM_TENSOR_SSM_DT, "blk.%d.ssm_dt" },
|
|
387
|
-
{ LLM_TENSOR_SSM_BETA_ALPHA, "blk.%d.ssm_ba" },
|
|
388
|
-
{ LLM_TENSOR_SSM_ALPHA, "blk.%d.ssm_alpha" },
|
|
389
|
-
{ LLM_TENSOR_SSM_IN, "blk.%d.ssm_in" },
|
|
390
|
-
{ LLM_TENSOR_SSM_NORM, "blk.%d.ssm_norm" },
|
|
391
|
-
{ LLM_TENSOR_SSM_OUT, "blk.%d.ssm_out" },
|
|
392
|
-
{ LLM_TENSOR_ROPE_FACTORS_LONG, "rope_factors_long" },
|
|
393
|
-
{ LLM_TENSOR_ROPE_FACTORS_SHORT, "rope_factors_short" },
|
|
394
|
-
{ LLM_TENSOR_SSM_X, "blk.%d.ssm_x" },
|
|
395
|
-
{ LLM_TENSOR_SSM_A, "blk.%d.ssm_a" },
|
|
396
|
-
{ LLM_TENSOR_SSM_D, "blk.%d.ssm_d" },
|
|
397
|
-
{ LLM_TENSOR_SSM_DT_NORM, "blk.%d.ssm_dt_norm" },
|
|
398
|
-
{ LLM_TENSOR_SSM_B_NORM, "blk.%d.ssm_b_norm" },
|
|
399
|
-
{ LLM_TENSOR_SSM_C_NORM, "blk.%d.ssm_c_norm" },
|
|
400
|
-
{ LLM_TENSOR_SSM_CONV1D_Q, "blk.%d.ssm_conv1d_q" },
|
|
401
|
-
{ LLM_TENSOR_SSM_CONV1D_K, "blk.%d.ssm_conv1d_k" },
|
|
402
|
-
{ LLM_TENSOR_SSM_CONV1D_V, "blk.%d.ssm_conv1d_v" },
|
|
403
|
-
{ LLM_TENSOR_SSM_F_A, "blk.%d.ssm_f_a" },
|
|
404
|
-
{ LLM_TENSOR_SSM_F_B, "blk.%d.ssm_f_b" },
|
|
405
|
-
{ LLM_TENSOR_SSM_BETA, "blk.%d.ssm_beta" },
|
|
406
|
-
{ LLM_TENSOR_SSM_G_A, "blk.%d.ssm_g_a" },
|
|
407
|
-
{ LLM_TENSOR_SSM_G_B, "blk.%d.ssm_g_b" },
|
|
408
|
-
{ LLM_TENSOR_SSM_NORM, "blk.%d.ssm_norm" },
|
|
409
|
-
{ LLM_TENSOR_ATTN_Q_A_NORM, "blk.%d.attn_q_a_norm" },
|
|
410
|
-
{ LLM_TENSOR_ATTN_KV_A_NORM, "blk.%d.attn_kv_a_norm" },
|
|
411
|
-
{ LLM_TENSOR_ATTN_Q_A, "blk.%d.attn_q_a" },
|
|
412
|
-
{ LLM_TENSOR_ATTN_Q_B, "blk.%d.attn_q_b" },
|
|
413
|
-
{ LLM_TENSOR_ATTN_KV_A_MQA, "blk.%d.attn_kv_a_mqa" },
|
|
414
|
-
{ LLM_TENSOR_ATTN_KV_B, "blk.%d.attn_kv_b" },
|
|
415
|
-
{ LLM_TENSOR_PER_LAYER_TOKEN_EMBD, "per_layer_token_embd" },
|
|
416
|
-
{ LLM_TENSOR_PER_LAYER_MODEL_PROJ, "per_layer_model_proj" },
|
|
417
|
-
{ LLM_TENSOR_PER_LAYER_PROJ_NORM, "per_layer_proj_norm" },
|
|
418
|
-
{ LLM_TENSOR_ALTUP_UNEMBD_PROJ, "altup_unembd_proj" },
|
|
419
|
-
{ LLM_TENSOR_ALTUP_PROJ, "altup_proj" },
|
|
420
|
-
{ LLM_TENSOR_PER_LAYER_INP_GATE, "blk.%d.inp_gate" },
|
|
421
|
-
{ LLM_TENSOR_PER_LAYER_PROJ, "blk.%d.proj" },
|
|
422
|
-
{ LLM_TENSOR_PER_LAYER_POST_NORM, "blk.%d.post_norm" },
|
|
423
|
-
{ LLM_TENSOR_ALTUP_CORRECT_COEF, "blk.%d.altup_correct_coef" },
|
|
424
|
-
{ LLM_TENSOR_ALTUP_CORRECT_SCALE, "blk.%d.altup_correct_scale" },
|
|
425
|
-
{ LLM_TENSOR_ALTUP_PREDICT_COEF, "blk.%d.altup_predict_coef" },
|
|
426
|
-
{ LLM_TENSOR_ALTUP_ROUTER, "blk.%d.altup_router" },
|
|
427
|
-
{ LLM_TENSOR_ALTUP_ROUTER_NORM, "blk.%d.altup_router_norm" },
|
|
428
|
-
{ LLM_TENSOR_LAUREL_L, "blk.%d.laurel_l" },
|
|
429
|
-
{ LLM_TENSOR_LAUREL_R, "blk.%d.laurel_r" },
|
|
430
|
-
{ LLM_TENSOR_LAUREL_POST_NORM, "blk.%d.laurel_post_norm" },
|
|
431
|
-
{ LLM_TENSOR_DENSE_2_OUT, "dense_2" },
|
|
432
|
-
{ LLM_TENSOR_DENSE_3_OUT, "dense_3" },
|
|
433
|
-
{ LLM_TENSOR_FFN_NORM_EXPS, "blk.%d.ffn_norm_exps" },
|
|
434
|
-
{ LLM_TENSOR_ATTN_K_B, "blk.%d.attn_k_b" },
|
|
435
|
-
{ LLM_TENSOR_ATTN_V_B, "blk.%d.attn_v_b" },
|
|
436
|
-
{ LLM_TENSOR_NEXTN_EH_PROJ, "blk.%d.nextn.eh_proj" },
|
|
437
|
-
{ LLM_TENSOR_NEXTN_EMBED_TOKENS, "blk.%d.nextn.embed_tokens" },
|
|
438
|
-
{ LLM_TENSOR_NEXTN_ENORM, "blk.%d.nextn.enorm" },
|
|
439
|
-
{ LLM_TENSOR_NEXTN_HNORM, "blk.%d.nextn.hnorm" },
|
|
440
|
-
{ LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, "blk.%d.nextn.shared_head_head" },
|
|
441
|
-
{ LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, "blk.%d.nextn.shared_head_norm" },
|
|
442
|
-
{ LLM_TENSOR_ATTN_SUB_NORM, "blk.%d.attn_sub_norm" },
|
|
443
|
-
{ LLM_TENSOR_FFN_SUB_NORM, "blk.%d.ffn_sub_norm" },
|
|
444
|
-
{ LLM_TENSOR_DEC_OUTPUT_NORM, "dec.output_norm" },
|
|
445
|
-
{ LLM_TENSOR_DEC_ATTN_NORM, "dec.blk.%d.attn_norm" },
|
|
446
|
-
{ LLM_TENSOR_DEC_ATTN_Q, "dec.blk.%d.attn_q" },
|
|
447
|
-
{ LLM_TENSOR_DEC_ATTN_K, "dec.blk.%d.attn_k" },
|
|
448
|
-
{ LLM_TENSOR_DEC_ATTN_V, "dec.blk.%d.attn_v" },
|
|
449
|
-
{ LLM_TENSOR_DEC_ATTN_OUT, "dec.blk.%d.attn_o" },
|
|
450
|
-
{ LLM_TENSOR_DEC_ATTN_REL_B, "dec.blk.%d.attn_rel_b" },
|
|
451
|
-
{ LLM_TENSOR_DEC_CROSS_ATTN_NORM, "dec.blk.%d.cross_attn_norm" },
|
|
452
|
-
{ LLM_TENSOR_DEC_CROSS_ATTN_Q, "dec.blk.%d.cross_attn_q" },
|
|
453
|
-
{ LLM_TENSOR_DEC_CROSS_ATTN_K, "dec.blk.%d.cross_attn_k" },
|
|
454
|
-
{ LLM_TENSOR_DEC_CROSS_ATTN_V, "dec.blk.%d.cross_attn_v" },
|
|
455
|
-
{ LLM_TENSOR_DEC_CROSS_ATTN_OUT, "dec.blk.%d.cross_attn_o" },
|
|
456
|
-
{ LLM_TENSOR_DEC_CROSS_ATTN_REL_B, "dec.blk.%d.cross_attn_rel_b" },
|
|
457
|
-
{ LLM_TENSOR_DEC_FFN_NORM, "dec.blk.%d.ffn_norm" },
|
|
458
|
-
{ LLM_TENSOR_DEC_FFN_GATE, "dec.blk.%d.ffn_gate" },
|
|
459
|
-
{ LLM_TENSOR_DEC_FFN_DOWN, "dec.blk.%d.ffn_down" },
|
|
460
|
-
{ LLM_TENSOR_DEC_FFN_UP, "dec.blk.%d.ffn_up" },
|
|
461
|
-
{ LLM_TENSOR_ENC_ATTN_NORM, "enc.blk.%d.attn_norm" },
|
|
462
|
-
{ LLM_TENSOR_ENC_ATTN_Q, "enc.blk.%d.attn_q" },
|
|
463
|
-
{ LLM_TENSOR_ENC_ATTN_K, "enc.blk.%d.attn_k" },
|
|
464
|
-
{ LLM_TENSOR_ENC_ATTN_V, "enc.blk.%d.attn_v" },
|
|
465
|
-
{ LLM_TENSOR_ENC_ATTN_OUT, "enc.blk.%d.attn_o" },
|
|
466
|
-
{ LLM_TENSOR_ENC_ATTN_REL_B, "enc.blk.%d.attn_rel_b" },
|
|
467
|
-
{ LLM_TENSOR_ENC_FFN_NORM, "enc.blk.%d.ffn_norm" },
|
|
468
|
-
{ LLM_TENSOR_ENC_FFN_GATE, "enc.blk.%d.ffn_gate" },
|
|
469
|
-
{ LLM_TENSOR_ENC_FFN_DOWN, "enc.blk.%d.ffn_down" },
|
|
470
|
-
{ LLM_TENSOR_ENC_FFN_UP, "enc.blk.%d.ffn_up" },
|
|
471
|
-
{ LLM_TENSOR_TIME_MIX_W1, "blk.%d.time_mix_w1" },
|
|
472
|
-
{ LLM_TENSOR_TIME_MIX_W2, "blk.%d.time_mix_w2" },
|
|
473
|
-
{ LLM_TENSOR_TIME_MIX_LERP_X, "blk.%d.time_mix_lerp_x" },
|
|
474
|
-
{ LLM_TENSOR_TIME_MIX_LERP_W, "blk.%d.time_mix_lerp_w" },
|
|
475
|
-
{ LLM_TENSOR_TIME_MIX_LERP_K, "blk.%d.time_mix_lerp_k" },
|
|
476
|
-
{ LLM_TENSOR_TIME_MIX_LERP_V, "blk.%d.time_mix_lerp_v" },
|
|
477
|
-
{ LLM_TENSOR_TIME_MIX_LERP_R, "blk.%d.time_mix_lerp_r" },
|
|
478
|
-
{ LLM_TENSOR_TIME_MIX_LERP_G, "blk.%d.time_mix_lerp_g" },
|
|
479
|
-
{ LLM_TENSOR_TIME_MIX_LERP_FUSED, "blk.%d.time_mix_lerp_fused" },
|
|
480
|
-
{ LLM_TENSOR_TIME_MIX_FIRST, "blk.%d.time_mix_first" },
|
|
481
|
-
{ LLM_TENSOR_TIME_MIX_DECAY, "blk.%d.time_mix_decay" },
|
|
482
|
-
{ LLM_TENSOR_TIME_MIX_DECAY_W1, "blk.%d.time_mix_decay_w1" },
|
|
483
|
-
{ LLM_TENSOR_TIME_MIX_DECAY_W2, "blk.%d.time_mix_decay_w2" },
|
|
484
|
-
{ LLM_TENSOR_TIME_MIX_KEY, "blk.%d.time_mix_key" },
|
|
485
|
-
{ LLM_TENSOR_TIME_MIX_VALUE, "blk.%d.time_mix_value" },
|
|
486
|
-
{ LLM_TENSOR_TIME_MIX_RECEPTANCE, "blk.%d.time_mix_receptance" },
|
|
487
|
-
{ LLM_TENSOR_TIME_MIX_GATE, "blk.%d.time_mix_gate" },
|
|
488
|
-
{ LLM_TENSOR_TIME_MIX_LN, "blk.%d.time_mix_ln" },
|
|
489
|
-
{ LLM_TENSOR_TIME_MIX_OUTPUT, "blk.%d.time_mix_output" },
|
|
490
|
-
{ LLM_TENSOR_CHANNEL_MIX_LERP_K, "blk.%d.channel_mix_lerp_k" },
|
|
491
|
-
{ LLM_TENSOR_CHANNEL_MIX_LERP_R, "blk.%d.channel_mix_lerp_r" },
|
|
492
|
-
{ LLM_TENSOR_CHANNEL_MIX_KEY, "blk.%d.channel_mix_key" },
|
|
493
|
-
{ LLM_TENSOR_CHANNEL_MIX_VALUE, "blk.%d.channel_mix_value" },
|
|
494
|
-
{ LLM_TENSOR_CHANNEL_MIX_RECEPTANCE, "blk.%d.channel_mix_receptance" },
|
|
495
|
-
{ LLM_TENSOR_TIME_MIX_W0, "blk.%d.time_mix_w0" },
|
|
496
|
-
{ LLM_TENSOR_TIME_MIX_A0, "blk.%d.time_mix_a0" },
|
|
497
|
-
{ LLM_TENSOR_TIME_MIX_A1, "blk.%d.time_mix_a1" },
|
|
498
|
-
{ LLM_TENSOR_TIME_MIX_A2, "blk.%d.time_mix_a2" },
|
|
499
|
-
{ LLM_TENSOR_TIME_MIX_V0, "blk.%d.time_mix_v0" },
|
|
500
|
-
{ LLM_TENSOR_TIME_MIX_V1, "blk.%d.time_mix_v1" },
|
|
501
|
-
{ LLM_TENSOR_TIME_MIX_V2, "blk.%d.time_mix_v2" },
|
|
502
|
-
{ LLM_TENSOR_TIME_MIX_G1, "blk.%d.time_mix_g1" },
|
|
503
|
-
{ LLM_TENSOR_TIME_MIX_G2, "blk.%d.time_mix_g2" },
|
|
504
|
-
{ LLM_TENSOR_TIME_MIX_K_K, "blk.%d.time_mix_k_k" },
|
|
505
|
-
{ LLM_TENSOR_TIME_MIX_K_A, "blk.%d.time_mix_k_a" },
|
|
506
|
-
{ LLM_TENSOR_TIME_MIX_R_K, "blk.%d.time_mix_r_k" },
|
|
507
|
-
{ LLM_TENSOR_CONV1D, "conv1d" },
|
|
508
|
-
{ LLM_TENSOR_CONVNEXT_DW, "convnext.%d.dw" },
|
|
509
|
-
{ LLM_TENSOR_CONVNEXT_NORM, "convnext.%d.norm" },
|
|
510
|
-
{ LLM_TENSOR_CONVNEXT_PW1, "convnext.%d.pw1" },
|
|
511
|
-
{ LLM_TENSOR_CONVNEXT_PW2, "convnext.%d.pw2" },
|
|
512
|
-
{ LLM_TENSOR_CONVNEXT_GAMMA, "convnext.%d.gamma" },
|
|
513
|
-
{ LLM_TENSOR_POS_NET_CONV1, "posnet.%d.conv1" },
|
|
514
|
-
{ LLM_TENSOR_POS_NET_CONV2, "posnet.%d.conv2" },
|
|
515
|
-
{ LLM_TENSOR_POS_NET_NORM, "posnet.%d.norm" },
|
|
516
|
-
{ LLM_TENSOR_POS_NET_NORM1, "posnet.%d.norm1" },
|
|
517
|
-
{ LLM_TENSOR_POS_NET_NORM2, "posnet.%d.norm2" },
|
|
518
|
-
{ LLM_TENSOR_POS_NET_ATTN_NORM, "posnet.%d.attn_norm" },
|
|
519
|
-
{ LLM_TENSOR_POS_NET_ATTN_Q, "posnet.%d.attn_q" },
|
|
520
|
-
{ LLM_TENSOR_POS_NET_ATTN_K, "posnet.%d.attn_k" },
|
|
521
|
-
{ LLM_TENSOR_POS_NET_ATTN_V, "posnet.%d.attn_v" },
|
|
522
|
-
{ LLM_TENSOR_POS_NET_ATTN_OUT, "posnet.%d.attn_output" },
|
|
523
|
-
{ LLM_TENSOR_ATTN_SINKS, "blk.%d.attn_sinks" },
|
|
524
|
-
{ LLM_TENSOR_SHORTCONV_CONV, "blk.%d.shortconv.conv" },
|
|
525
|
-
{ LLM_TENSOR_SHORTCONV_INPROJ, "blk.%d.shortconv.in_proj" },
|
|
526
|
-
{ LLM_TENSOR_SHORTCONV_OUTPROJ, "blk.%d.shortconv.out_proj" },
|
|
527
|
-
{ LLM_TENSOR_FFN_GATE_CHEXPS, "blk.%d.ffn_gate_chexps" },
|
|
528
|
-
{ LLM_TENSOR_FFN_DOWN_CHEXPS, "blk.%d.ffn_down_chexps" },
|
|
529
|
-
{ LLM_TENSOR_FFN_UP_CHEXPS, "blk.%d.ffn_up_chexps" },
|
|
530
|
-
{ LLM_TENSOR_VISEXP_ATTN_QKV, "blk.%d.vis_attn_qkv" },
|
|
531
|
-
{ LLM_TENSOR_VISEXP_ATTN_OUT, "blk.%d.vis_attn_output" },
|
|
532
|
-
{ LLM_TENSOR_VISEXP_FFN_GATE, "blk.%d.vis_gate" },
|
|
533
|
-
{ LLM_TENSOR_VISEXP_FFN_DOWN, "blk.%d.vis_down" },
|
|
534
|
-
{ LLM_TENSOR_VISEXP_FFN_UP, "blk.%d.vis_up" },
|
|
535
|
-
{ LLM_TENSOR_INDEXER_K_NORM, "blk.%d.indexer.k_norm" },
|
|
536
|
-
{ LLM_TENSOR_INDEXER_PROJ, "blk.%d.indexer.proj" },
|
|
537
|
-
{ LLM_TENSOR_INDEXER_ATTN_K, "blk.%d.indexer.attn_k" },
|
|
538
|
-
{ LLM_TENSOR_INDEXER_ATTN_Q_B, "blk.%d.indexer.attn_q_b" },
|
|
539
|
-
};
|
|
540
|
-
|
|
541
|
-
static std::set<llm_tensor> llm_get_tensor_names(llm_arch arch) {
|
|
542
|
-
switch (arch) {
|
|
543
|
-
case LLM_ARCH_CLIP:
|
|
544
|
-
return {};
|
|
545
|
-
case LLM_ARCH_LLAMA:
|
|
546
|
-
case LLM_ARCH_DECI:
|
|
547
|
-
case LLM_ARCH_MISTRAL3:
|
|
548
|
-
case LLM_ARCH_LLAMA_EMBED:
|
|
549
|
-
return {
|
|
550
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
551
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
552
|
-
LLM_TENSOR_OUTPUT,
|
|
553
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
554
|
-
LLM_TENSOR_ATTN_NORM,
|
|
555
|
-
LLM_TENSOR_ATTN_Q,
|
|
556
|
-
LLM_TENSOR_ATTN_K,
|
|
557
|
-
LLM_TENSOR_ATTN_V,
|
|
558
|
-
LLM_TENSOR_ATTN_OUT,
|
|
559
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
560
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
561
|
-
LLM_TENSOR_FFN_NORM,
|
|
562
|
-
LLM_TENSOR_FFN_GATE,
|
|
563
|
-
LLM_TENSOR_FFN_DOWN,
|
|
564
|
-
LLM_TENSOR_FFN_UP,
|
|
565
|
-
LLM_TENSOR_FFN_GATE_EXP,
|
|
566
|
-
LLM_TENSOR_FFN_DOWN_EXP,
|
|
567
|
-
LLM_TENSOR_FFN_UP_EXP,
|
|
568
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
569
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
570
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
571
|
-
};
|
|
572
|
-
case LLM_ARCH_ARCEE:
|
|
573
|
-
case LLM_ARCH_STARCODER2:
|
|
574
|
-
case LLM_ARCH_NEMOTRON:
|
|
575
|
-
return {
|
|
576
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
577
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
578
|
-
LLM_TENSOR_OUTPUT,
|
|
579
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
580
|
-
LLM_TENSOR_ATTN_NORM,
|
|
581
|
-
LLM_TENSOR_ATTN_Q,
|
|
582
|
-
LLM_TENSOR_ATTN_K,
|
|
583
|
-
LLM_TENSOR_ATTN_V,
|
|
584
|
-
LLM_TENSOR_ATTN_OUT,
|
|
585
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
586
|
-
LLM_TENSOR_FFN_NORM,
|
|
587
|
-
LLM_TENSOR_FFN_DOWN,
|
|
588
|
-
LLM_TENSOR_FFN_UP,
|
|
589
|
-
};
|
|
590
|
-
case LLM_ARCH_AFMOE:
|
|
591
|
-
return {
|
|
592
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
593
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
594
|
-
LLM_TENSOR_OUTPUT,
|
|
595
|
-
LLM_TENSOR_ATTN_NORM,
|
|
596
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
597
|
-
LLM_TENSOR_ATTN_Q,
|
|
598
|
-
LLM_TENSOR_ATTN_K,
|
|
599
|
-
LLM_TENSOR_ATTN_V,
|
|
600
|
-
LLM_TENSOR_ATTN_OUT,
|
|
601
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
602
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
603
|
-
LLM_TENSOR_ATTN_GATE,
|
|
604
|
-
LLM_TENSOR_FFN_NORM,
|
|
605
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
606
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
607
|
-
LLM_TENSOR_FFN_GATE,
|
|
608
|
-
LLM_TENSOR_FFN_DOWN,
|
|
609
|
-
LLM_TENSOR_FFN_UP,
|
|
610
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
611
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
612
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
613
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
614
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
615
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
616
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
617
|
-
};
|
|
618
|
-
case LLM_ARCH_LLAMA4:
|
|
619
|
-
return {
|
|
620
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
621
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
622
|
-
LLM_TENSOR_OUTPUT,
|
|
623
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
624
|
-
LLM_TENSOR_ATTN_NORM,
|
|
625
|
-
LLM_TENSOR_ATTN_Q,
|
|
626
|
-
LLM_TENSOR_ATTN_K,
|
|
627
|
-
LLM_TENSOR_ATTN_V,
|
|
628
|
-
LLM_TENSOR_ATTN_OUT,
|
|
629
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
630
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
631
|
-
LLM_TENSOR_FFN_NORM,
|
|
632
|
-
LLM_TENSOR_FFN_GATE,
|
|
633
|
-
LLM_TENSOR_FFN_DOWN,
|
|
634
|
-
LLM_TENSOR_FFN_UP,
|
|
635
|
-
LLM_TENSOR_FFN_GATE_EXP,
|
|
636
|
-
LLM_TENSOR_FFN_DOWN_EXP,
|
|
637
|
-
LLM_TENSOR_FFN_UP_EXP,
|
|
638
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
639
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
640
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
641
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
642
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
643
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
644
|
-
};
|
|
645
|
-
case LLM_ARCH_BAICHUAN:
|
|
646
|
-
case LLM_ARCH_ORION:
|
|
647
|
-
case LLM_ARCH_XVERSE:
|
|
648
|
-
case LLM_ARCH_EXAONE:
|
|
649
|
-
return {
|
|
650
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
651
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
652
|
-
LLM_TENSOR_OUTPUT,
|
|
653
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
654
|
-
LLM_TENSOR_ATTN_NORM,
|
|
655
|
-
LLM_TENSOR_ATTN_Q,
|
|
656
|
-
LLM_TENSOR_ATTN_K,
|
|
657
|
-
LLM_TENSOR_ATTN_V,
|
|
658
|
-
LLM_TENSOR_ATTN_OUT,
|
|
659
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
660
|
-
LLM_TENSOR_FFN_NORM,
|
|
661
|
-
LLM_TENSOR_FFN_GATE,
|
|
662
|
-
LLM_TENSOR_FFN_DOWN,
|
|
663
|
-
LLM_TENSOR_FFN_UP,
|
|
664
|
-
};
|
|
665
|
-
case LLM_ARCH_FALCON:
|
|
666
|
-
return {
|
|
667
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
668
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
669
|
-
LLM_TENSOR_OUTPUT,
|
|
670
|
-
LLM_TENSOR_ATTN_NORM,
|
|
671
|
-
LLM_TENSOR_ATTN_NORM_2,
|
|
672
|
-
LLM_TENSOR_ATTN_QKV,
|
|
673
|
-
LLM_TENSOR_ATTN_OUT,
|
|
674
|
-
LLM_TENSOR_FFN_DOWN,
|
|
675
|
-
LLM_TENSOR_FFN_UP,
|
|
676
|
-
};
|
|
677
|
-
case LLM_ARCH_GROK:
|
|
678
|
-
return {
|
|
679
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
680
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
681
|
-
LLM_TENSOR_OUTPUT,
|
|
682
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
683
|
-
LLM_TENSOR_ATTN_NORM,
|
|
684
|
-
LLM_TENSOR_ATTN_Q,
|
|
685
|
-
LLM_TENSOR_ATTN_K,
|
|
686
|
-
LLM_TENSOR_ATTN_V,
|
|
687
|
-
LLM_TENSOR_ATTN_OUT,
|
|
688
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
689
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
690
|
-
LLM_TENSOR_FFN_NORM,
|
|
691
|
-
LLM_TENSOR_FFN_GATE,
|
|
692
|
-
LLM_TENSOR_FFN_DOWN,
|
|
693
|
-
LLM_TENSOR_FFN_UP,
|
|
694
|
-
LLM_TENSOR_FFN_GATE_EXP,
|
|
695
|
-
LLM_TENSOR_FFN_DOWN_EXP,
|
|
696
|
-
LLM_TENSOR_FFN_UP_EXP,
|
|
697
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
698
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
699
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
700
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
701
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
702
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
703
|
-
};
|
|
704
|
-
case LLM_ARCH_GPT2:
|
|
705
|
-
case LLM_ARCH_STARCODER:
|
|
706
|
-
return {
|
|
707
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
708
|
-
LLM_TENSOR_POS_EMBD,
|
|
709
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
710
|
-
LLM_TENSOR_OUTPUT,
|
|
711
|
-
LLM_TENSOR_ATTN_NORM,
|
|
712
|
-
LLM_TENSOR_ATTN_QKV,
|
|
713
|
-
LLM_TENSOR_ATTN_OUT,
|
|
714
|
-
LLM_TENSOR_FFN_NORM,
|
|
715
|
-
LLM_TENSOR_FFN_UP,
|
|
716
|
-
LLM_TENSOR_FFN_DOWN,
|
|
717
|
-
};
|
|
718
|
-
case LLM_ARCH_GPTNEOX:
|
|
719
|
-
return {
|
|
720
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
721
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
722
|
-
LLM_TENSOR_OUTPUT,
|
|
723
|
-
LLM_TENSOR_ATTN_NORM,
|
|
724
|
-
LLM_TENSOR_ATTN_QKV,
|
|
725
|
-
LLM_TENSOR_ATTN_OUT,
|
|
726
|
-
LLM_TENSOR_FFN_NORM,
|
|
727
|
-
LLM_TENSOR_FFN_DOWN,
|
|
728
|
-
LLM_TENSOR_FFN_UP,
|
|
729
|
-
};
|
|
730
|
-
case LLM_ARCH_MPT:
|
|
731
|
-
return {
|
|
732
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
733
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
734
|
-
LLM_TENSOR_OUTPUT,
|
|
735
|
-
LLM_TENSOR_ATTN_NORM,
|
|
736
|
-
LLM_TENSOR_FFN_NORM,
|
|
737
|
-
LLM_TENSOR_ATTN_QKV,
|
|
738
|
-
LLM_TENSOR_ATTN_OUT,
|
|
739
|
-
LLM_TENSOR_FFN_DOWN,
|
|
740
|
-
LLM_TENSOR_FFN_UP,
|
|
741
|
-
LLM_TENSOR_FFN_ACT,
|
|
742
|
-
LLM_TENSOR_POS_EMBD,
|
|
743
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
744
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
745
|
-
};
|
|
746
|
-
case LLM_ARCH_REFACT:
|
|
747
|
-
case LLM_ARCH_QWEN2:
|
|
748
|
-
case LLM_ARCH_QWEN2VL:
|
|
749
|
-
case LLM_ARCH_INTERNLM2:
|
|
750
|
-
case LLM_ARCH_GRANITE:
|
|
751
|
-
case LLM_ARCH_ERNIE4_5:
|
|
752
|
-
case LLM_ARCH_PADDLEOCR:
|
|
753
|
-
case LLM_ARCH_SMOLLM3:
|
|
754
|
-
case LLM_ARCH_DREAM:
|
|
755
|
-
case LLM_ARCH_LLADA:
|
|
756
|
-
case LLM_ARCH_PANGU_EMBED:
|
|
757
|
-
return {
|
|
758
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
759
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
760
|
-
LLM_TENSOR_OUTPUT,
|
|
761
|
-
LLM_TENSOR_ATTN_NORM,
|
|
762
|
-
LLM_TENSOR_ATTN_Q,
|
|
763
|
-
LLM_TENSOR_ATTN_K,
|
|
764
|
-
LLM_TENSOR_ATTN_V,
|
|
765
|
-
LLM_TENSOR_ATTN_OUT,
|
|
766
|
-
LLM_TENSOR_FFN_NORM,
|
|
767
|
-
LLM_TENSOR_FFN_GATE,
|
|
768
|
-
LLM_TENSOR_FFN_DOWN,
|
|
769
|
-
LLM_TENSOR_FFN_UP,
|
|
770
|
-
};
|
|
771
|
-
case LLM_ARCH_BERT:
|
|
772
|
-
return {
|
|
773
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
774
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
775
|
-
LLM_TENSOR_TOKEN_TYPES,
|
|
776
|
-
LLM_TENSOR_POS_EMBD,
|
|
777
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
778
|
-
LLM_TENSOR_ATTN_QKV,
|
|
779
|
-
LLM_TENSOR_ATTN_Q,
|
|
780
|
-
LLM_TENSOR_ATTN_K,
|
|
781
|
-
LLM_TENSOR_ATTN_V,
|
|
782
|
-
LLM_TENSOR_ATTN_OUT,
|
|
783
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
784
|
-
LLM_TENSOR_FFN_DOWN,
|
|
785
|
-
LLM_TENSOR_FFN_UP,
|
|
786
|
-
LLM_TENSOR_CLS,
|
|
787
|
-
LLM_TENSOR_CLS_OUT,
|
|
788
|
-
};
|
|
789
|
-
case LLM_ARCH_NOMIC_BERT:
|
|
790
|
-
return {
|
|
791
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
792
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
793
|
-
LLM_TENSOR_TOKEN_TYPES,
|
|
794
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
795
|
-
LLM_TENSOR_ATTN_QKV,
|
|
796
|
-
LLM_TENSOR_ATTN_OUT,
|
|
797
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
798
|
-
LLM_TENSOR_FFN_GATE,
|
|
799
|
-
LLM_TENSOR_FFN_DOWN,
|
|
800
|
-
LLM_TENSOR_FFN_UP,
|
|
801
|
-
};
|
|
802
|
-
case LLM_ARCH_NOMIC_BERT_MOE:
|
|
803
|
-
return {
|
|
804
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
805
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
806
|
-
LLM_TENSOR_TOKEN_TYPES,
|
|
807
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
808
|
-
LLM_TENSOR_ATTN_QKV,
|
|
809
|
-
LLM_TENSOR_ATTN_OUT,
|
|
810
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
811
|
-
LLM_TENSOR_FFN_GATE,
|
|
812
|
-
LLM_TENSOR_FFN_DOWN,
|
|
813
|
-
LLM_TENSOR_FFN_UP,
|
|
814
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
815
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
816
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
817
|
-
};
|
|
818
|
-
case LLM_ARCH_NEO_BERT:
|
|
819
|
-
return {
|
|
820
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
821
|
-
LLM_TENSOR_ATTN_NORM,
|
|
822
|
-
LLM_TENSOR_ATTN_QKV,
|
|
823
|
-
LLM_TENSOR_ATTN_OUT,
|
|
824
|
-
LLM_TENSOR_FFN_NORM,
|
|
825
|
-
LLM_TENSOR_FFN_DOWN,
|
|
826
|
-
LLM_TENSOR_FFN_UP,
|
|
827
|
-
LLM_TENSOR_ENC_OUTPUT_NORM,
|
|
828
|
-
LLM_TENSOR_CLS,
|
|
829
|
-
LLM_TENSOR_CLS_OUT,
|
|
830
|
-
};
|
|
831
|
-
case LLM_ARCH_EUROBERT:
|
|
832
|
-
return {
|
|
833
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
834
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
835
|
-
LLM_TENSOR_ATTN_NORM,
|
|
836
|
-
LLM_TENSOR_ATTN_Q,
|
|
837
|
-
LLM_TENSOR_ATTN_K,
|
|
838
|
-
LLM_TENSOR_ATTN_V,
|
|
839
|
-
LLM_TENSOR_ATTN_OUT,
|
|
840
|
-
LLM_TENSOR_FFN_NORM,
|
|
841
|
-
LLM_TENSOR_FFN_GATE,
|
|
842
|
-
LLM_TENSOR_FFN_UP,
|
|
843
|
-
LLM_TENSOR_FFN_DOWN,
|
|
844
|
-
};
|
|
845
|
-
case LLM_ARCH_MODERN_BERT:
|
|
846
|
-
return {
|
|
847
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
848
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
849
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
850
|
-
LLM_TENSOR_ATTN_NORM,
|
|
851
|
-
LLM_TENSOR_ATTN_OUT,
|
|
852
|
-
LLM_TENSOR_ATTN_QKV,
|
|
853
|
-
LLM_TENSOR_FFN_DOWN,
|
|
854
|
-
LLM_TENSOR_FFN_UP,
|
|
855
|
-
LLM_TENSOR_FFN_NORM,
|
|
856
|
-
LLM_TENSOR_CLS,
|
|
857
|
-
LLM_TENSOR_CLS_OUT,
|
|
858
|
-
LLM_TENSOR_CLS_NORM,
|
|
859
|
-
};
|
|
860
|
-
case LLM_ARCH_JINA_BERT_V2:
|
|
861
|
-
return {
|
|
862
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
863
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
864
|
-
LLM_TENSOR_TOKEN_TYPES,
|
|
865
|
-
LLM_TENSOR_ATTN_NORM_2,
|
|
866
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
867
|
-
LLM_TENSOR_ATTN_Q,
|
|
868
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
869
|
-
LLM_TENSOR_ATTN_K,
|
|
870
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
871
|
-
LLM_TENSOR_ATTN_V,
|
|
872
|
-
LLM_TENSOR_ATTN_OUT,
|
|
873
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
874
|
-
LLM_TENSOR_FFN_DOWN,
|
|
875
|
-
LLM_TENSOR_FFN_GATE,
|
|
876
|
-
LLM_TENSOR_FFN_UP,
|
|
877
|
-
LLM_TENSOR_CLS,
|
|
878
|
-
};
|
|
879
|
-
case LLM_ARCH_JINA_BERT_V3:
|
|
880
|
-
return {
|
|
881
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
882
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
883
|
-
LLM_TENSOR_TOKEN_TYPES,
|
|
884
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
885
|
-
LLM_TENSOR_ATTN_QKV,
|
|
886
|
-
LLM_TENSOR_ATTN_OUT,
|
|
887
|
-
LLM_TENSOR_FFN_DOWN,
|
|
888
|
-
LLM_TENSOR_FFN_UP,
|
|
889
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
890
|
-
};
|
|
891
|
-
case LLM_ARCH_BLOOM:
|
|
892
|
-
return {
|
|
893
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
894
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
895
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
896
|
-
LLM_TENSOR_OUTPUT,
|
|
897
|
-
LLM_TENSOR_ATTN_NORM,
|
|
898
|
-
LLM_TENSOR_ATTN_QKV,
|
|
899
|
-
LLM_TENSOR_ATTN_OUT,
|
|
900
|
-
LLM_TENSOR_FFN_NORM,
|
|
901
|
-
LLM_TENSOR_FFN_UP,
|
|
902
|
-
LLM_TENSOR_FFN_DOWN,
|
|
903
|
-
};
|
|
904
|
-
case LLM_ARCH_STABLELM:
|
|
905
|
-
return {
|
|
906
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
907
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
908
|
-
LLM_TENSOR_OUTPUT,
|
|
909
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
910
|
-
LLM_TENSOR_ATTN_NORM,
|
|
911
|
-
LLM_TENSOR_ATTN_Q,
|
|
912
|
-
LLM_TENSOR_ATTN_K,
|
|
913
|
-
LLM_TENSOR_ATTN_V,
|
|
914
|
-
LLM_TENSOR_ATTN_OUT,
|
|
915
|
-
LLM_TENSOR_FFN_NORM,
|
|
916
|
-
LLM_TENSOR_FFN_GATE,
|
|
917
|
-
LLM_TENSOR_FFN_DOWN,
|
|
918
|
-
LLM_TENSOR_FFN_UP,
|
|
919
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
920
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
921
|
-
};
|
|
922
|
-
case LLM_ARCH_QWEN:
|
|
923
|
-
return {
|
|
924
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
925
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
926
|
-
LLM_TENSOR_OUTPUT,
|
|
927
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
928
|
-
LLM_TENSOR_ATTN_NORM,
|
|
929
|
-
LLM_TENSOR_ATTN_QKV,
|
|
930
|
-
LLM_TENSOR_ATTN_OUT,
|
|
931
|
-
LLM_TENSOR_FFN_NORM,
|
|
932
|
-
LLM_TENSOR_FFN_GATE,
|
|
933
|
-
LLM_TENSOR_FFN_DOWN,
|
|
934
|
-
LLM_TENSOR_FFN_UP,
|
|
935
|
-
};
|
|
936
|
-
case LLM_ARCH_QWEN2MOE:
|
|
937
|
-
return {
|
|
938
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
939
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
940
|
-
LLM_TENSOR_OUTPUT,
|
|
941
|
-
LLM_TENSOR_ATTN_NORM,
|
|
942
|
-
LLM_TENSOR_ATTN_Q,
|
|
943
|
-
LLM_TENSOR_ATTN_K,
|
|
944
|
-
LLM_TENSOR_ATTN_V,
|
|
945
|
-
LLM_TENSOR_ATTN_OUT,
|
|
946
|
-
LLM_TENSOR_FFN_NORM,
|
|
947
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
948
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
949
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
950
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
951
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
952
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
953
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
954
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
955
|
-
};
|
|
956
|
-
case LLM_ARCH_QWEN3:
|
|
957
|
-
return {
|
|
958
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
959
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
960
|
-
LLM_TENSOR_OUTPUT,
|
|
961
|
-
LLM_TENSOR_CLS_OUT,
|
|
962
|
-
LLM_TENSOR_ATTN_NORM,
|
|
963
|
-
LLM_TENSOR_ATTN_Q,
|
|
964
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
965
|
-
LLM_TENSOR_ATTN_K,
|
|
966
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
967
|
-
LLM_TENSOR_ATTN_V,
|
|
968
|
-
LLM_TENSOR_ATTN_OUT,
|
|
969
|
-
LLM_TENSOR_FFN_NORM,
|
|
970
|
-
LLM_TENSOR_FFN_GATE,
|
|
971
|
-
LLM_TENSOR_FFN_DOWN,
|
|
972
|
-
LLM_TENSOR_FFN_UP,
|
|
973
|
-
};
|
|
974
|
-
case LLM_ARCH_QWEN3MOE:
|
|
975
|
-
case LLM_ARCH_QWEN3VLMOE:
|
|
976
|
-
case LLM_ARCH_OLMOE:
|
|
977
|
-
case LLM_ARCH_LLADA_MOE:
|
|
978
|
-
case LLM_ARCH_RND1:
|
|
979
|
-
return {
|
|
980
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
981
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
982
|
-
LLM_TENSOR_OUTPUT,
|
|
983
|
-
LLM_TENSOR_ATTN_NORM,
|
|
984
|
-
LLM_TENSOR_ATTN_Q,
|
|
985
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
986
|
-
LLM_TENSOR_ATTN_K,
|
|
987
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
988
|
-
LLM_TENSOR_ATTN_V,
|
|
989
|
-
LLM_TENSOR_ATTN_OUT,
|
|
990
|
-
LLM_TENSOR_FFN_NORM,
|
|
991
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
992
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
993
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
994
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
995
|
-
};
|
|
996
|
-
case LLM_ARCH_QWEN3NEXT:
|
|
997
|
-
return {
|
|
998
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
999
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1000
|
-
LLM_TENSOR_OUTPUT,
|
|
1001
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1002
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1003
|
-
LLM_TENSOR_ATTN_Q,
|
|
1004
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1005
|
-
LLM_TENSOR_ATTN_K,
|
|
1006
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1007
|
-
LLM_TENSOR_ATTN_V,
|
|
1008
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1009
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1010
|
-
LLM_TENSOR_ATTN_GATE,
|
|
1011
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1012
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1013
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1014
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1015
|
-
LLM_TENSOR_FFN_GATE_UP_EXPS,
|
|
1016
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
1017
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1018
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1019
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1020
|
-
LLM_TENSOR_SSM_A_NOSCAN,
|
|
1021
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1022
|
-
LLM_TENSOR_SSM_DT,
|
|
1023
|
-
LLM_TENSOR_SSM_BETA_ALPHA,
|
|
1024
|
-
LLM_TENSOR_SSM_IN,
|
|
1025
|
-
LLM_TENSOR_SSM_NORM,
|
|
1026
|
-
LLM_TENSOR_SSM_OUT,
|
|
1027
|
-
};
|
|
1028
|
-
case LLM_ARCH_QWEN35:
|
|
1029
|
-
return {
|
|
1030
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1031
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1032
|
-
LLM_TENSOR_OUTPUT,
|
|
1033
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1034
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1035
|
-
LLM_TENSOR_ATTN_Q,
|
|
1036
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1037
|
-
LLM_TENSOR_ATTN_K,
|
|
1038
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1039
|
-
LLM_TENSOR_ATTN_V,
|
|
1040
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1041
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1042
|
-
LLM_TENSOR_ATTN_GATE,
|
|
1043
|
-
LLM_TENSOR_FFN_GATE,
|
|
1044
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1045
|
-
LLM_TENSOR_FFN_UP,
|
|
1046
|
-
LLM_TENSOR_SSM_A_NOSCAN,
|
|
1047
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1048
|
-
LLM_TENSOR_SSM_DT,
|
|
1049
|
-
LLM_TENSOR_SSM_BETA,
|
|
1050
|
-
LLM_TENSOR_SSM_ALPHA,
|
|
1051
|
-
LLM_TENSOR_SSM_NORM,
|
|
1052
|
-
LLM_TENSOR_SSM_OUT,
|
|
1053
|
-
};
|
|
1054
|
-
case LLM_ARCH_QWEN35MOE:
|
|
1055
|
-
return {
|
|
1056
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1057
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1058
|
-
LLM_TENSOR_OUTPUT,
|
|
1059
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1060
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1061
|
-
LLM_TENSOR_ATTN_Q,
|
|
1062
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1063
|
-
LLM_TENSOR_ATTN_K,
|
|
1064
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1065
|
-
LLM_TENSOR_ATTN_V,
|
|
1066
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1067
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1068
|
-
LLM_TENSOR_ATTN_GATE,
|
|
1069
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1070
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1071
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1072
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1073
|
-
LLM_TENSOR_FFN_GATE_UP_EXPS,
|
|
1074
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
1075
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1076
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1077
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1078
|
-
LLM_TENSOR_SSM_A_NOSCAN,
|
|
1079
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1080
|
-
LLM_TENSOR_SSM_DT,
|
|
1081
|
-
LLM_TENSOR_SSM_BETA,
|
|
1082
|
-
LLM_TENSOR_SSM_ALPHA,
|
|
1083
|
-
LLM_TENSOR_SSM_NORM,
|
|
1084
|
-
LLM_TENSOR_SSM_OUT,
|
|
1085
|
-
};
|
|
1086
|
-
case LLM_ARCH_QWEN3VL:
|
|
1087
|
-
case LLM_ARCH_CHAMELEON:
|
|
1088
|
-
case LLM_ARCH_HUNYUAN_DENSE:
|
|
1089
|
-
return {
|
|
1090
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1091
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1092
|
-
LLM_TENSOR_OUTPUT,
|
|
1093
|
-
LLM_TENSOR_CLS_OUT,
|
|
1094
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1095
|
-
LLM_TENSOR_ATTN_Q,
|
|
1096
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1097
|
-
LLM_TENSOR_ATTN_K,
|
|
1098
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1099
|
-
LLM_TENSOR_ATTN_V,
|
|
1100
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1101
|
-
LLM_TENSOR_FFN_NORM,
|
|
1102
|
-
LLM_TENSOR_FFN_GATE,
|
|
1103
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1104
|
-
LLM_TENSOR_FFN_UP,
|
|
1105
|
-
};
|
|
1106
|
-
case LLM_ARCH_PHI2:
|
|
1107
|
-
return {
|
|
1108
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1109
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1110
|
-
LLM_TENSOR_OUTPUT,
|
|
1111
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1112
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1113
|
-
LLM_TENSOR_ATTN_Q,
|
|
1114
|
-
LLM_TENSOR_ATTN_K,
|
|
1115
|
-
LLM_TENSOR_ATTN_V,
|
|
1116
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1117
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1118
|
-
LLM_TENSOR_FFN_UP,
|
|
1119
|
-
};
|
|
1120
|
-
case LLM_ARCH_PHI3:
|
|
1121
|
-
return {
|
|
1122
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1123
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1124
|
-
LLM_TENSOR_OUTPUT,
|
|
1125
|
-
LLM_TENSOR_ROPE_FACTORS_LONG,
|
|
1126
|
-
LLM_TENSOR_ROPE_FACTORS_SHORT,
|
|
1127
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1128
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1129
|
-
LLM_TENSOR_ATTN_Q,
|
|
1130
|
-
LLM_TENSOR_ATTN_K,
|
|
1131
|
-
LLM_TENSOR_ATTN_V,
|
|
1132
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1133
|
-
LLM_TENSOR_FFN_NORM,
|
|
1134
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1135
|
-
LLM_TENSOR_FFN_UP,
|
|
1136
|
-
};
|
|
1137
|
-
case LLM_ARCH_PHIMOE:
|
|
1138
|
-
return {
|
|
1139
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1140
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1141
|
-
LLM_TENSOR_OUTPUT,
|
|
1142
|
-
LLM_TENSOR_ROPE_FACTORS_LONG,
|
|
1143
|
-
LLM_TENSOR_ROPE_FACTORS_SHORT,
|
|
1144
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1145
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1146
|
-
LLM_TENSOR_ATTN_Q,
|
|
1147
|
-
LLM_TENSOR_ATTN_K,
|
|
1148
|
-
LLM_TENSOR_ATTN_V,
|
|
1149
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1150
|
-
LLM_TENSOR_FFN_NORM,
|
|
1151
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1152
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1153
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1154
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1155
|
-
};
|
|
1156
|
-
case LLM_ARCH_PLAMO:
|
|
1157
|
-
return {
|
|
1158
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1159
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1160
|
-
LLM_TENSOR_OUTPUT,
|
|
1161
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1162
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1163
|
-
LLM_TENSOR_ATTN_Q,
|
|
1164
|
-
LLM_TENSOR_ATTN_K,
|
|
1165
|
-
LLM_TENSOR_ATTN_V,
|
|
1166
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1167
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
1168
|
-
LLM_TENSOR_FFN_GATE,
|
|
1169
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1170
|
-
LLM_TENSOR_FFN_UP,
|
|
1171
|
-
};
|
|
1172
|
-
case LLM_ARCH_PLAMO2:
|
|
1173
|
-
return {
|
|
1174
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1175
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1176
|
-
LLM_TENSOR_OUTPUT,
|
|
1177
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1178
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1179
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1180
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1181
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1182
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1183
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
1184
|
-
LLM_TENSOR_FFN_NORM,
|
|
1185
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1186
|
-
LLM_TENSOR_FFN_UP,
|
|
1187
|
-
LLM_TENSOR_SSM_IN,
|
|
1188
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1189
|
-
LLM_TENSOR_SSM_X,
|
|
1190
|
-
LLM_TENSOR_SSM_DT,
|
|
1191
|
-
LLM_TENSOR_SSM_A,
|
|
1192
|
-
LLM_TENSOR_SSM_D,
|
|
1193
|
-
LLM_TENSOR_SSM_OUT,
|
|
1194
|
-
LLM_TENSOR_SSM_DT_NORM,
|
|
1195
|
-
LLM_TENSOR_SSM_B_NORM,
|
|
1196
|
-
LLM_TENSOR_SSM_C_NORM,
|
|
1197
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1198
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1199
|
-
};
|
|
1200
|
-
case LLM_ARCH_PLAMO3:
|
|
1201
|
-
return {
|
|
1202
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1203
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1204
|
-
LLM_TENSOR_OUTPUT,
|
|
1205
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1206
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1207
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1208
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1209
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1210
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1211
|
-
LLM_TENSOR_FFN_NORM,
|
|
1212
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1213
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1214
|
-
LLM_TENSOR_FFN_UP,
|
|
1215
|
-
};
|
|
1216
|
-
case LLM_ARCH_CODESHELL:
|
|
1217
|
-
return {
|
|
1218
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1219
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1220
|
-
LLM_TENSOR_OUTPUT,
|
|
1221
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1222
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1223
|
-
LLM_TENSOR_ATTN_Q,
|
|
1224
|
-
LLM_TENSOR_ATTN_K,
|
|
1225
|
-
LLM_TENSOR_ATTN_V,
|
|
1226
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1227
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1228
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
1229
|
-
LLM_TENSOR_FFN_NORM,
|
|
1230
|
-
LLM_TENSOR_FFN_GATE,
|
|
1231
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1232
|
-
LLM_TENSOR_FFN_UP,
|
|
1233
|
-
};
|
|
1234
|
-
case LLM_ARCH_MINICPM:
|
|
1235
|
-
return {
|
|
1236
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1237
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1238
|
-
LLM_TENSOR_OUTPUT,
|
|
1239
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1240
|
-
LLM_TENSOR_ROPE_FACTORS_LONG,
|
|
1241
|
-
LLM_TENSOR_ROPE_FACTORS_SHORT,
|
|
1242
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1243
|
-
LLM_TENSOR_ATTN_Q,
|
|
1244
|
-
LLM_TENSOR_ATTN_K,
|
|
1245
|
-
LLM_TENSOR_ATTN_V,
|
|
1246
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1247
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
1248
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1249
|
-
LLM_TENSOR_FFN_NORM,
|
|
1250
|
-
LLM_TENSOR_FFN_GATE,
|
|
1251
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1252
|
-
LLM_TENSOR_FFN_UP,
|
|
1253
|
-
LLM_TENSOR_FFN_GATE_EXP,
|
|
1254
|
-
LLM_TENSOR_FFN_DOWN_EXP,
|
|
1255
|
-
LLM_TENSOR_FFN_UP_EXP,
|
|
1256
|
-
};
|
|
1257
|
-
case LLM_ARCH_MINICPM3:
|
|
1258
|
-
return {
|
|
1259
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1260
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1261
|
-
LLM_TENSOR_OUTPUT,
|
|
1262
|
-
LLM_TENSOR_ROPE_FACTORS_LONG,
|
|
1263
|
-
LLM_TENSOR_ROPE_FACTORS_SHORT,
|
|
1264
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1265
|
-
LLM_TENSOR_ATTN_Q_A_NORM,
|
|
1266
|
-
LLM_TENSOR_ATTN_KV_A_NORM,
|
|
1267
|
-
LLM_TENSOR_ATTN_Q,
|
|
1268
|
-
LLM_TENSOR_ATTN_Q_A,
|
|
1269
|
-
LLM_TENSOR_ATTN_Q_B,
|
|
1270
|
-
LLM_TENSOR_ATTN_KV_A_MQA,
|
|
1271
|
-
LLM_TENSOR_ATTN_KV_B,
|
|
1272
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1273
|
-
LLM_TENSOR_FFN_NORM,
|
|
1274
|
-
LLM_TENSOR_FFN_GATE,
|
|
1275
|
-
LLM_TENSOR_FFN_UP,
|
|
1276
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1277
|
-
};
|
|
1278
|
-
case LLM_ARCH_GEMMA:
|
|
1279
|
-
return {
|
|
1280
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1281
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1282
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1283
|
-
LLM_TENSOR_ATTN_Q,
|
|
1284
|
-
LLM_TENSOR_ATTN_K,
|
|
1285
|
-
LLM_TENSOR_ATTN_V,
|
|
1286
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1287
|
-
LLM_TENSOR_FFN_NORM,
|
|
1288
|
-
LLM_TENSOR_FFN_GATE,
|
|
1289
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1290
|
-
LLM_TENSOR_FFN_UP,
|
|
1291
|
-
};
|
|
1292
|
-
case LLM_ARCH_GEMMA2:
|
|
1293
|
-
return {
|
|
1294
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1295
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1296
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1297
|
-
LLM_TENSOR_ATTN_Q,
|
|
1298
|
-
LLM_TENSOR_ATTN_K,
|
|
1299
|
-
LLM_TENSOR_ATTN_V,
|
|
1300
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1301
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1302
|
-
LLM_TENSOR_FFN_NORM,
|
|
1303
|
-
LLM_TENSOR_FFN_GATE,
|
|
1304
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1305
|
-
LLM_TENSOR_FFN_UP,
|
|
1306
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1307
|
-
};
|
|
1308
|
-
case LLM_ARCH_GEMMA3:
|
|
1309
|
-
return {
|
|
1310
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1311
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1312
|
-
LLM_TENSOR_OUTPUT,
|
|
1313
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1314
|
-
LLM_TENSOR_ATTN_Q,
|
|
1315
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1316
|
-
LLM_TENSOR_ATTN_K,
|
|
1317
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1318
|
-
LLM_TENSOR_ATTN_V,
|
|
1319
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1320
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1321
|
-
LLM_TENSOR_FFN_NORM,
|
|
1322
|
-
LLM_TENSOR_FFN_GATE,
|
|
1323
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1324
|
-
LLM_TENSOR_FFN_UP,
|
|
1325
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1326
|
-
};
|
|
1327
|
-
case LLM_ARCH_GEMMA3N:
|
|
1328
|
-
return {
|
|
1329
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1330
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1331
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1332
|
-
LLM_TENSOR_ATTN_Q,
|
|
1333
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1334
|
-
LLM_TENSOR_ATTN_K,
|
|
1335
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1336
|
-
LLM_TENSOR_ATTN_V,
|
|
1337
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1338
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1339
|
-
LLM_TENSOR_FFN_NORM,
|
|
1340
|
-
LLM_TENSOR_FFN_GATE,
|
|
1341
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1342
|
-
LLM_TENSOR_FFN_UP,
|
|
1343
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1344
|
-
LLM_TENSOR_PER_LAYER_TOKEN_EMBD,
|
|
1345
|
-
LLM_TENSOR_PER_LAYER_MODEL_PROJ,
|
|
1346
|
-
LLM_TENSOR_PER_LAYER_PROJ_NORM,
|
|
1347
|
-
LLM_TENSOR_ALTUP_UNEMBD_PROJ,
|
|
1348
|
-
LLM_TENSOR_ALTUP_PROJ,
|
|
1349
|
-
LLM_TENSOR_PER_LAYER_INP_GATE,
|
|
1350
|
-
LLM_TENSOR_PER_LAYER_PROJ,
|
|
1351
|
-
LLM_TENSOR_PER_LAYER_POST_NORM,
|
|
1352
|
-
LLM_TENSOR_ALTUP_CORRECT_COEF,
|
|
1353
|
-
LLM_TENSOR_ALTUP_CORRECT_SCALE,
|
|
1354
|
-
LLM_TENSOR_ALTUP_PREDICT_COEF,
|
|
1355
|
-
LLM_TENSOR_ALTUP_ROUTER,
|
|
1356
|
-
LLM_TENSOR_ALTUP_ROUTER_NORM,
|
|
1357
|
-
LLM_TENSOR_LAUREL_L,
|
|
1358
|
-
LLM_TENSOR_LAUREL_R,
|
|
1359
|
-
LLM_TENSOR_LAUREL_POST_NORM,
|
|
1360
|
-
};
|
|
1361
|
-
case LLM_ARCH_GEMMA_EMBEDDING:
|
|
1362
|
-
return {
|
|
1363
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1364
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1365
|
-
LLM_TENSOR_OUTPUT,
|
|
1366
|
-
LLM_TENSOR_DENSE_2_OUT,
|
|
1367
|
-
LLM_TENSOR_DENSE_3_OUT,
|
|
1368
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1369
|
-
LLM_TENSOR_ATTN_Q,
|
|
1370
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1371
|
-
LLM_TENSOR_ATTN_K,
|
|
1372
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1373
|
-
LLM_TENSOR_ATTN_V,
|
|
1374
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1375
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1376
|
-
LLM_TENSOR_FFN_NORM,
|
|
1377
|
-
LLM_TENSOR_FFN_GATE,
|
|
1378
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1379
|
-
LLM_TENSOR_FFN_UP,
|
|
1380
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1381
|
-
};
|
|
1382
|
-
case LLM_ARCH_MAMBA:
|
|
1383
|
-
return {
|
|
1384
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1385
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1386
|
-
LLM_TENSOR_OUTPUT,
|
|
1387
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1388
|
-
LLM_TENSOR_SSM_IN,
|
|
1389
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1390
|
-
LLM_TENSOR_SSM_X,
|
|
1391
|
-
LLM_TENSOR_SSM_DT,
|
|
1392
|
-
LLM_TENSOR_SSM_A,
|
|
1393
|
-
LLM_TENSOR_SSM_D,
|
|
1394
|
-
LLM_TENSOR_SSM_OUT,
|
|
1395
|
-
};
|
|
1396
|
-
case LLM_ARCH_MAMBA2:
|
|
1397
|
-
return {
|
|
1398
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1399
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1400
|
-
LLM_TENSOR_OUTPUT,
|
|
1401
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1402
|
-
LLM_TENSOR_SSM_IN,
|
|
1403
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1404
|
-
LLM_TENSOR_SSM_DT,
|
|
1405
|
-
LLM_TENSOR_SSM_A,
|
|
1406
|
-
LLM_TENSOR_SSM_D,
|
|
1407
|
-
LLM_TENSOR_SSM_NORM,
|
|
1408
|
-
LLM_TENSOR_SSM_OUT,
|
|
1409
|
-
};
|
|
1410
|
-
case LLM_ARCH_JAMBA:
|
|
1411
|
-
return {
|
|
1412
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1413
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1414
|
-
LLM_TENSOR_OUTPUT,
|
|
1415
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1416
|
-
LLM_TENSOR_SSM_IN,
|
|
1417
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1418
|
-
LLM_TENSOR_SSM_X,
|
|
1419
|
-
LLM_TENSOR_SSM_DT,
|
|
1420
|
-
LLM_TENSOR_SSM_DT_NORM,
|
|
1421
|
-
LLM_TENSOR_SSM_A,
|
|
1422
|
-
LLM_TENSOR_SSM_B_NORM,
|
|
1423
|
-
LLM_TENSOR_SSM_C_NORM,
|
|
1424
|
-
LLM_TENSOR_SSM_D,
|
|
1425
|
-
LLM_TENSOR_SSM_OUT,
|
|
1426
|
-
LLM_TENSOR_ATTN_Q,
|
|
1427
|
-
LLM_TENSOR_ATTN_K,
|
|
1428
|
-
LLM_TENSOR_ATTN_V,
|
|
1429
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1430
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1431
|
-
LLM_TENSOR_FFN_NORM,
|
|
1432
|
-
LLM_TENSOR_FFN_GATE,
|
|
1433
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1434
|
-
LLM_TENSOR_FFN_UP,
|
|
1435
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1436
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1437
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1438
|
-
};
|
|
1439
|
-
case LLM_ARCH_FALCON_H1:
|
|
1440
|
-
return {
|
|
1441
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1442
|
-
LLM_TENSOR_OUTPUT,
|
|
1443
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1444
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1445
|
-
LLM_TENSOR_ATTN_Q,
|
|
1446
|
-
LLM_TENSOR_ATTN_K,
|
|
1447
|
-
LLM_TENSOR_ATTN_V,
|
|
1448
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1449
|
-
LLM_TENSOR_SSM_IN,
|
|
1450
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1451
|
-
LLM_TENSOR_SSM_DT,
|
|
1452
|
-
LLM_TENSOR_SSM_A,
|
|
1453
|
-
LLM_TENSOR_SSM_D,
|
|
1454
|
-
LLM_TENSOR_SSM_NORM,
|
|
1455
|
-
LLM_TENSOR_SSM_OUT,
|
|
1456
|
-
LLM_TENSOR_FFN_NORM,
|
|
1457
|
-
LLM_TENSOR_FFN_GATE,
|
|
1458
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1459
|
-
LLM_TENSOR_FFN_UP,
|
|
1460
|
-
};
|
|
1461
|
-
case LLM_ARCH_COMMAND_R:
|
|
1462
|
-
return {
|
|
1463
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1464
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1465
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1466
|
-
LLM_TENSOR_ATTN_Q,
|
|
1467
|
-
LLM_TENSOR_ATTN_K,
|
|
1468
|
-
LLM_TENSOR_ATTN_V,
|
|
1469
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1470
|
-
LLM_TENSOR_FFN_GATE,
|
|
1471
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1472
|
-
LLM_TENSOR_FFN_UP,
|
|
1473
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1474
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1475
|
-
};
|
|
1476
|
-
case LLM_ARCH_COHERE2:
|
|
1477
|
-
return {
|
|
1478
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1479
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1480
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1481
|
-
LLM_TENSOR_ATTN_Q,
|
|
1482
|
-
LLM_TENSOR_ATTN_K,
|
|
1483
|
-
LLM_TENSOR_ATTN_V,
|
|
1484
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1485
|
-
LLM_TENSOR_FFN_GATE,
|
|
1486
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1487
|
-
LLM_TENSOR_FFN_UP,
|
|
1488
|
-
};
|
|
1489
|
-
case LLM_ARCH_DBRX:
|
|
1490
|
-
return {
|
|
1491
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1492
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1493
|
-
LLM_TENSOR_OUTPUT,
|
|
1494
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1495
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1496
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1497
|
-
LLM_TENSOR_ATTN_OUT_NORM,
|
|
1498
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1499
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1500
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1501
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1502
|
-
};
|
|
1503
|
-
case LLM_ARCH_OLMO:
|
|
1504
|
-
return {
|
|
1505
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1506
|
-
LLM_TENSOR_OUTPUT,
|
|
1507
|
-
LLM_TENSOR_ATTN_Q,
|
|
1508
|
-
LLM_TENSOR_ATTN_K,
|
|
1509
|
-
LLM_TENSOR_ATTN_V,
|
|
1510
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1511
|
-
LLM_TENSOR_FFN_GATE,
|
|
1512
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1513
|
-
LLM_TENSOR_FFN_UP,
|
|
1514
|
-
};
|
|
1515
|
-
case LLM_ARCH_OLMO2:
|
|
1516
|
-
return {
|
|
1517
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1518
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1519
|
-
LLM_TENSOR_OUTPUT,
|
|
1520
|
-
LLM_TENSOR_ATTN_Q,
|
|
1521
|
-
LLM_TENSOR_ATTN_K,
|
|
1522
|
-
LLM_TENSOR_ATTN_V,
|
|
1523
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1524
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1525
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1526
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1527
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1528
|
-
LLM_TENSOR_FFN_GATE,
|
|
1529
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1530
|
-
LLM_TENSOR_FFN_UP,
|
|
1531
|
-
};
|
|
1532
|
-
case LLM_ARCH_OPENELM:
|
|
1533
|
-
return {
|
|
1534
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1535
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1536
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1537
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1538
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1539
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1540
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1541
|
-
LLM_TENSOR_FFN_NORM,
|
|
1542
|
-
LLM_TENSOR_FFN_GATE,
|
|
1543
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1544
|
-
LLM_TENSOR_FFN_UP,
|
|
1545
|
-
};
|
|
1546
|
-
case LLM_ARCH_ARCTIC:
|
|
1547
|
-
return {
|
|
1548
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1549
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1550
|
-
LLM_TENSOR_OUTPUT,
|
|
1551
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1552
|
-
LLM_TENSOR_ATTN_Q,
|
|
1553
|
-
LLM_TENSOR_ATTN_K,
|
|
1554
|
-
LLM_TENSOR_ATTN_V,
|
|
1555
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1556
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1557
|
-
LLM_TENSOR_FFN_NORM,
|
|
1558
|
-
LLM_TENSOR_FFN_GATE,
|
|
1559
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1560
|
-
LLM_TENSOR_FFN_UP,
|
|
1561
|
-
LLM_TENSOR_FFN_NORM_EXPS,
|
|
1562
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1563
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1564
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1565
|
-
};
|
|
1566
|
-
case LLM_ARCH_DEEPSEEK:
|
|
1567
|
-
return {
|
|
1568
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1569
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1570
|
-
LLM_TENSOR_OUTPUT,
|
|
1571
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1572
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1573
|
-
LLM_TENSOR_ATTN_Q,
|
|
1574
|
-
LLM_TENSOR_ATTN_K,
|
|
1575
|
-
LLM_TENSOR_ATTN_V,
|
|
1576
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1577
|
-
LLM_TENSOR_ATTN_ROT_EMBD,
|
|
1578
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1579
|
-
LLM_TENSOR_FFN_NORM,
|
|
1580
|
-
LLM_TENSOR_FFN_GATE,
|
|
1581
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1582
|
-
LLM_TENSOR_FFN_UP,
|
|
1583
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1584
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1585
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1586
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
1587
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1588
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1589
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1590
|
-
};
|
|
1591
|
-
case LLM_ARCH_DEEPSEEK2:
|
|
1592
|
-
return {
|
|
1593
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1594
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1595
|
-
LLM_TENSOR_OUTPUT,
|
|
1596
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1597
|
-
LLM_TENSOR_ATTN_Q_A_NORM,
|
|
1598
|
-
LLM_TENSOR_ATTN_KV_A_NORM,
|
|
1599
|
-
LLM_TENSOR_ATTN_Q,
|
|
1600
|
-
LLM_TENSOR_ATTN_Q_A,
|
|
1601
|
-
LLM_TENSOR_ATTN_Q_B,
|
|
1602
|
-
LLM_TENSOR_ATTN_KV_A_MQA,
|
|
1603
|
-
LLM_TENSOR_ATTN_KV_B,
|
|
1604
|
-
LLM_TENSOR_ATTN_K_B,
|
|
1605
|
-
LLM_TENSOR_ATTN_V_B,
|
|
1606
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1607
|
-
LLM_TENSOR_FFN_NORM,
|
|
1608
|
-
LLM_TENSOR_FFN_GATE,
|
|
1609
|
-
LLM_TENSOR_FFN_UP,
|
|
1610
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1611
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1612
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1613
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1614
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1615
|
-
LLM_TENSOR_FFN_GATE_UP_EXPS,
|
|
1616
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
1617
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1618
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1619
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1620
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
1621
|
-
};
|
|
1622
|
-
case LLM_ARCH_PLM:
|
|
1623
|
-
return {
|
|
1624
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1625
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1626
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1627
|
-
LLM_TENSOR_ATTN_Q,
|
|
1628
|
-
LLM_TENSOR_ATTN_KV_A_MQA,
|
|
1629
|
-
LLM_TENSOR_ATTN_KV_A_NORM,
|
|
1630
|
-
LLM_TENSOR_ATTN_KV_B,
|
|
1631
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1632
|
-
LLM_TENSOR_FFN_NORM,
|
|
1633
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1634
|
-
LLM_TENSOR_FFN_UP,
|
|
1635
|
-
};
|
|
1636
|
-
case LLM_ARCH_CHATGLM:
|
|
1637
|
-
return {
|
|
1638
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1639
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1640
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1641
|
-
LLM_TENSOR_OUTPUT,
|
|
1642
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1643
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1644
|
-
LLM_TENSOR_ATTN_Q,
|
|
1645
|
-
LLM_TENSOR_ATTN_K,
|
|
1646
|
-
LLM_TENSOR_ATTN_V,
|
|
1647
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1648
|
-
LLM_TENSOR_FFN_NORM,
|
|
1649
|
-
LLM_TENSOR_FFN_UP,
|
|
1650
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1651
|
-
};
|
|
1652
|
-
case LLM_ARCH_GLM4:
|
|
1653
|
-
return {
|
|
1654
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1655
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1656
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1657
|
-
LLM_TENSOR_OUTPUT,
|
|
1658
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1659
|
-
LLM_TENSOR_ATTN_Q,
|
|
1660
|
-
LLM_TENSOR_ATTN_K,
|
|
1661
|
-
LLM_TENSOR_ATTN_V,
|
|
1662
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1663
|
-
LLM_TENSOR_FFN_NORM,
|
|
1664
|
-
LLM_TENSOR_FFN_UP,
|
|
1665
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1666
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1667
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1668
|
-
LLM_TENSOR_NEXTN_EH_PROJ,
|
|
1669
|
-
LLM_TENSOR_NEXTN_EMBED_TOKENS,
|
|
1670
|
-
LLM_TENSOR_NEXTN_ENORM,
|
|
1671
|
-
LLM_TENSOR_NEXTN_HNORM,
|
|
1672
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
|
|
1673
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
|
|
1674
|
-
};
|
|
1675
|
-
case LLM_ARCH_GLM4_MOE:
|
|
1676
|
-
return {
|
|
1677
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1678
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1679
|
-
LLM_TENSOR_OUTPUT,
|
|
1680
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1681
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1682
|
-
LLM_TENSOR_ATTN_Q,
|
|
1683
|
-
LLM_TENSOR_ATTN_K,
|
|
1684
|
-
LLM_TENSOR_ATTN_V,
|
|
1685
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1686
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1687
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1688
|
-
LLM_TENSOR_FFN_GATE,
|
|
1689
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1690
|
-
LLM_TENSOR_FFN_UP,
|
|
1691
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1692
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1693
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1694
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1695
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1696
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1697
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1698
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
1699
|
-
LLM_TENSOR_NEXTN_EH_PROJ,
|
|
1700
|
-
LLM_TENSOR_NEXTN_EMBED_TOKENS,
|
|
1701
|
-
LLM_TENSOR_NEXTN_ENORM,
|
|
1702
|
-
LLM_TENSOR_NEXTN_HNORM,
|
|
1703
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
|
|
1704
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
|
|
1705
|
-
};
|
|
1706
|
-
case LLM_ARCH_GLM_DSA:
|
|
1707
|
-
return {
|
|
1708
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1709
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1710
|
-
LLM_TENSOR_OUTPUT,
|
|
1711
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1712
|
-
LLM_TENSOR_ATTN_Q_A_NORM,
|
|
1713
|
-
LLM_TENSOR_ATTN_KV_A_NORM,
|
|
1714
|
-
LLM_TENSOR_ATTN_Q,
|
|
1715
|
-
LLM_TENSOR_ATTN_Q_A,
|
|
1716
|
-
LLM_TENSOR_ATTN_Q_B,
|
|
1717
|
-
LLM_TENSOR_ATTN_KV_A_MQA,
|
|
1718
|
-
LLM_TENSOR_ATTN_KV_B,
|
|
1719
|
-
LLM_TENSOR_ATTN_K_B,
|
|
1720
|
-
LLM_TENSOR_ATTN_V_B,
|
|
1721
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1722
|
-
LLM_TENSOR_FFN_NORM,
|
|
1723
|
-
LLM_TENSOR_FFN_GATE,
|
|
1724
|
-
LLM_TENSOR_FFN_UP,
|
|
1725
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1726
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1727
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1728
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1729
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1730
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
1731
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1732
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1733
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1734
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
1735
|
-
LLM_TENSOR_INDEXER_K_NORM,
|
|
1736
|
-
LLM_TENSOR_INDEXER_PROJ,
|
|
1737
|
-
LLM_TENSOR_INDEXER_ATTN_K,
|
|
1738
|
-
LLM_TENSOR_INDEXER_ATTN_Q_B,
|
|
1739
|
-
LLM_TENSOR_NEXTN_EH_PROJ,
|
|
1740
|
-
LLM_TENSOR_NEXTN_EMBED_TOKENS,
|
|
1741
|
-
LLM_TENSOR_NEXTN_ENORM,
|
|
1742
|
-
LLM_TENSOR_NEXTN_HNORM,
|
|
1743
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
|
|
1744
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
|
|
1745
|
-
};
|
|
1746
|
-
case LLM_ARCH_BITNET:
|
|
1747
|
-
return {
|
|
1748
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1749
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1750
|
-
LLM_TENSOR_ATTN_Q,
|
|
1751
|
-
LLM_TENSOR_ATTN_K,
|
|
1752
|
-
LLM_TENSOR_ATTN_V,
|
|
1753
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1754
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1755
|
-
LLM_TENSOR_ATTN_SUB_NORM,
|
|
1756
|
-
LLM_TENSOR_FFN_GATE,
|
|
1757
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1758
|
-
LLM_TENSOR_FFN_UP,
|
|
1759
|
-
LLM_TENSOR_FFN_NORM,
|
|
1760
|
-
LLM_TENSOR_FFN_SUB_NORM,
|
|
1761
|
-
};
|
|
1762
|
-
case LLM_ARCH_T5:
|
|
1763
|
-
return {
|
|
1764
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1765
|
-
LLM_TENSOR_OUTPUT,
|
|
1766
|
-
LLM_TENSOR_DEC_OUTPUT_NORM,
|
|
1767
|
-
LLM_TENSOR_DEC_ATTN_NORM,
|
|
1768
|
-
LLM_TENSOR_DEC_ATTN_Q,
|
|
1769
|
-
LLM_TENSOR_DEC_ATTN_K,
|
|
1770
|
-
LLM_TENSOR_DEC_ATTN_V,
|
|
1771
|
-
LLM_TENSOR_DEC_ATTN_OUT,
|
|
1772
|
-
LLM_TENSOR_DEC_ATTN_REL_B,
|
|
1773
|
-
LLM_TENSOR_DEC_CROSS_ATTN_NORM,
|
|
1774
|
-
LLM_TENSOR_DEC_CROSS_ATTN_Q,
|
|
1775
|
-
LLM_TENSOR_DEC_CROSS_ATTN_K,
|
|
1776
|
-
LLM_TENSOR_DEC_CROSS_ATTN_V,
|
|
1777
|
-
LLM_TENSOR_DEC_CROSS_ATTN_OUT,
|
|
1778
|
-
LLM_TENSOR_DEC_CROSS_ATTN_REL_B,
|
|
1779
|
-
LLM_TENSOR_DEC_FFN_NORM,
|
|
1780
|
-
LLM_TENSOR_DEC_FFN_GATE,
|
|
1781
|
-
LLM_TENSOR_DEC_FFN_DOWN,
|
|
1782
|
-
LLM_TENSOR_DEC_FFN_UP,
|
|
1783
|
-
LLM_TENSOR_ENC_OUTPUT_NORM,
|
|
1784
|
-
LLM_TENSOR_ENC_ATTN_NORM,
|
|
1785
|
-
LLM_TENSOR_ENC_ATTN_Q,
|
|
1786
|
-
LLM_TENSOR_ENC_ATTN_K,
|
|
1787
|
-
LLM_TENSOR_ENC_ATTN_V,
|
|
1788
|
-
LLM_TENSOR_ENC_ATTN_OUT,
|
|
1789
|
-
LLM_TENSOR_ENC_ATTN_REL_B,
|
|
1790
|
-
LLM_TENSOR_ENC_FFN_NORM,
|
|
1791
|
-
LLM_TENSOR_ENC_FFN_GATE,
|
|
1792
|
-
LLM_TENSOR_ENC_FFN_DOWN,
|
|
1793
|
-
LLM_TENSOR_ENC_FFN_UP,
|
|
1794
|
-
};
|
|
1795
|
-
case LLM_ARCH_T5ENCODER:
|
|
1796
|
-
return {
|
|
1797
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1798
|
-
LLM_TENSOR_OUTPUT,
|
|
1799
|
-
LLM_TENSOR_ENC_OUTPUT_NORM,
|
|
1800
|
-
LLM_TENSOR_ENC_ATTN_NORM,
|
|
1801
|
-
LLM_TENSOR_ENC_ATTN_Q,
|
|
1802
|
-
LLM_TENSOR_ENC_ATTN_K,
|
|
1803
|
-
LLM_TENSOR_ENC_ATTN_V,
|
|
1804
|
-
LLM_TENSOR_ENC_ATTN_OUT,
|
|
1805
|
-
LLM_TENSOR_ENC_ATTN_REL_B,
|
|
1806
|
-
LLM_TENSOR_ENC_FFN_NORM,
|
|
1807
|
-
LLM_TENSOR_ENC_FFN_GATE,
|
|
1808
|
-
LLM_TENSOR_ENC_FFN_DOWN,
|
|
1809
|
-
LLM_TENSOR_ENC_FFN_UP,
|
|
1810
|
-
};
|
|
1811
|
-
case LLM_ARCH_JAIS:
|
|
1812
|
-
return {
|
|
1813
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1814
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1815
|
-
LLM_TENSOR_OUTPUT,
|
|
1816
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1817
|
-
LLM_TENSOR_ATTN_QKV,
|
|
1818
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1819
|
-
LLM_TENSOR_FFN_NORM,
|
|
1820
|
-
LLM_TENSOR_FFN_UP,
|
|
1821
|
-
LLM_TENSOR_FFN_GATE,
|
|
1822
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1823
|
-
};
|
|
1824
|
-
case LLM_ARCH_JAIS2:
|
|
1825
|
-
return {
|
|
1826
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1827
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1828
|
-
LLM_TENSOR_OUTPUT,
|
|
1829
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1830
|
-
LLM_TENSOR_ATTN_Q,
|
|
1831
|
-
LLM_TENSOR_ATTN_K,
|
|
1832
|
-
LLM_TENSOR_ATTN_V,
|
|
1833
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1834
|
-
LLM_TENSOR_FFN_NORM,
|
|
1835
|
-
LLM_TENSOR_FFN_UP,
|
|
1836
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1837
|
-
};
|
|
1838
|
-
case LLM_ARCH_NEMOTRON_H:
|
|
1839
|
-
return {
|
|
1840
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1841
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1842
|
-
LLM_TENSOR_OUTPUT,
|
|
1843
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1844
|
-
LLM_TENSOR_SSM_IN,
|
|
1845
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1846
|
-
LLM_TENSOR_SSM_DT,
|
|
1847
|
-
LLM_TENSOR_SSM_A,
|
|
1848
|
-
LLM_TENSOR_SSM_D,
|
|
1849
|
-
LLM_TENSOR_SSM_NORM,
|
|
1850
|
-
LLM_TENSOR_SSM_OUT,
|
|
1851
|
-
LLM_TENSOR_ATTN_Q,
|
|
1852
|
-
LLM_TENSOR_ATTN_K,
|
|
1853
|
-
LLM_TENSOR_ATTN_V,
|
|
1854
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1855
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1856
|
-
LLM_TENSOR_FFN_UP,
|
|
1857
|
-
};
|
|
1858
|
-
case LLM_ARCH_NEMOTRON_H_MOE:
|
|
1859
|
-
return {
|
|
1860
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1861
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1862
|
-
LLM_TENSOR_OUTPUT,
|
|
1863
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1864
|
-
// mamba(2) ssm layers
|
|
1865
|
-
LLM_TENSOR_SSM_IN,
|
|
1866
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
1867
|
-
LLM_TENSOR_SSM_DT,
|
|
1868
|
-
LLM_TENSOR_SSM_A,
|
|
1869
|
-
LLM_TENSOR_SSM_D,
|
|
1870
|
-
LLM_TENSOR_SSM_NORM,
|
|
1871
|
-
LLM_TENSOR_SSM_OUT,
|
|
1872
|
-
// attention layers
|
|
1873
|
-
LLM_TENSOR_ATTN_Q,
|
|
1874
|
-
LLM_TENSOR_ATTN_K,
|
|
1875
|
-
LLM_TENSOR_ATTN_V,
|
|
1876
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1877
|
-
// dense FFN
|
|
1878
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1879
|
-
LLM_TENSOR_FFN_UP,
|
|
1880
|
-
// MoE FFN (for MoE layers)
|
|
1881
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1882
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1883
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1884
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
1885
|
-
LLM_TENSOR_FFN_LATENT_DOWN,
|
|
1886
|
-
LLM_TENSOR_FFN_LATENT_UP,
|
|
1887
|
-
// MoE shared expert layer
|
|
1888
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1889
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1890
|
-
};
|
|
1891
|
-
case LLM_ARCH_EXAONE4:
|
|
1892
|
-
return {
|
|
1893
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1894
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1895
|
-
LLM_TENSOR_OUTPUT,
|
|
1896
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1897
|
-
LLM_TENSOR_ATTN_Q,
|
|
1898
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1899
|
-
LLM_TENSOR_ATTN_K,
|
|
1900
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1901
|
-
LLM_TENSOR_ATTN_V,
|
|
1902
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1903
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
1904
|
-
LLM_TENSOR_FFN_GATE,
|
|
1905
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1906
|
-
LLM_TENSOR_FFN_UP,
|
|
1907
|
-
LLM_TENSOR_FFN_POST_NORM,
|
|
1908
|
-
};
|
|
1909
|
-
case LLM_ARCH_EXAONE_MOE:
|
|
1910
|
-
return {
|
|
1911
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1912
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1913
|
-
LLM_TENSOR_OUTPUT,
|
|
1914
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
1915
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1916
|
-
LLM_TENSOR_ATTN_Q,
|
|
1917
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
1918
|
-
LLM_TENSOR_ATTN_K,
|
|
1919
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
1920
|
-
LLM_TENSOR_ATTN_V,
|
|
1921
|
-
LLM_TENSOR_ATTN_OUT,
|
|
1922
|
-
LLM_TENSOR_FFN_NORM,
|
|
1923
|
-
LLM_TENSOR_FFN_GATE,
|
|
1924
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1925
|
-
LLM_TENSOR_FFN_UP,
|
|
1926
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
1927
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
1928
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
1929
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
1930
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
1931
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
1932
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
1933
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
1934
|
-
LLM_TENSOR_NEXTN_EH_PROJ,
|
|
1935
|
-
LLM_TENSOR_NEXTN_EMBED_TOKENS,
|
|
1936
|
-
LLM_TENSOR_NEXTN_ENORM,
|
|
1937
|
-
LLM_TENSOR_NEXTN_HNORM,
|
|
1938
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
|
|
1939
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
|
|
1940
|
-
};
|
|
1941
|
-
case LLM_ARCH_RWKV6:
|
|
1942
|
-
return {
|
|
1943
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1944
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
1945
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1946
|
-
LLM_TENSOR_OUTPUT,
|
|
1947
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1948
|
-
LLM_TENSOR_ATTN_NORM_2,
|
|
1949
|
-
LLM_TENSOR_TIME_MIX_W1,
|
|
1950
|
-
LLM_TENSOR_TIME_MIX_W2,
|
|
1951
|
-
LLM_TENSOR_TIME_MIX_LERP_X,
|
|
1952
|
-
LLM_TENSOR_TIME_MIX_LERP_W,
|
|
1953
|
-
LLM_TENSOR_TIME_MIX_LERP_K,
|
|
1954
|
-
LLM_TENSOR_TIME_MIX_LERP_V,
|
|
1955
|
-
LLM_TENSOR_TIME_MIX_LERP_R,
|
|
1956
|
-
LLM_TENSOR_TIME_MIX_LERP_G,
|
|
1957
|
-
LLM_TENSOR_TIME_MIX_LERP_FUSED,
|
|
1958
|
-
LLM_TENSOR_TIME_MIX_FIRST,
|
|
1959
|
-
LLM_TENSOR_TIME_MIX_DECAY,
|
|
1960
|
-
LLM_TENSOR_TIME_MIX_DECAY_W1,
|
|
1961
|
-
LLM_TENSOR_TIME_MIX_DECAY_W2,
|
|
1962
|
-
LLM_TENSOR_TIME_MIX_KEY,
|
|
1963
|
-
LLM_TENSOR_TIME_MIX_VALUE,
|
|
1964
|
-
LLM_TENSOR_TIME_MIX_RECEPTANCE,
|
|
1965
|
-
LLM_TENSOR_TIME_MIX_GATE,
|
|
1966
|
-
LLM_TENSOR_TIME_MIX_LN,
|
|
1967
|
-
LLM_TENSOR_TIME_MIX_OUTPUT,
|
|
1968
|
-
LLM_TENSOR_CHANNEL_MIX_LERP_K,
|
|
1969
|
-
LLM_TENSOR_CHANNEL_MIX_LERP_R,
|
|
1970
|
-
LLM_TENSOR_CHANNEL_MIX_KEY,
|
|
1971
|
-
LLM_TENSOR_CHANNEL_MIX_VALUE,
|
|
1972
|
-
LLM_TENSOR_CHANNEL_MIX_RECEPTANCE,
|
|
1973
|
-
};
|
|
1974
|
-
case LLM_ARCH_RWKV6QWEN2:
|
|
1975
|
-
return {
|
|
1976
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
1977
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
1978
|
-
LLM_TENSOR_OUTPUT,
|
|
1979
|
-
LLM_TENSOR_ATTN_NORM,
|
|
1980
|
-
LLM_TENSOR_TIME_MIX_W1,
|
|
1981
|
-
LLM_TENSOR_TIME_MIX_W2,
|
|
1982
|
-
LLM_TENSOR_TIME_MIX_LERP_X,
|
|
1983
|
-
LLM_TENSOR_TIME_MIX_LERP_FUSED,
|
|
1984
|
-
LLM_TENSOR_TIME_MIX_FIRST,
|
|
1985
|
-
LLM_TENSOR_TIME_MIX_DECAY,
|
|
1986
|
-
LLM_TENSOR_TIME_MIX_DECAY_W1,
|
|
1987
|
-
LLM_TENSOR_TIME_MIX_DECAY_W2,
|
|
1988
|
-
LLM_TENSOR_TIME_MIX_KEY,
|
|
1989
|
-
LLM_TENSOR_TIME_MIX_VALUE,
|
|
1990
|
-
LLM_TENSOR_TIME_MIX_RECEPTANCE,
|
|
1991
|
-
LLM_TENSOR_TIME_MIX_GATE,
|
|
1992
|
-
LLM_TENSOR_TIME_MIX_OUTPUT,
|
|
1993
|
-
LLM_TENSOR_FFN_NORM,
|
|
1994
|
-
LLM_TENSOR_FFN_GATE,
|
|
1995
|
-
LLM_TENSOR_FFN_DOWN,
|
|
1996
|
-
LLM_TENSOR_FFN_UP,
|
|
1997
|
-
};
|
|
1998
|
-
case LLM_ARCH_RWKV7:
|
|
1999
|
-
return {
|
|
2000
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2001
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
2002
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2003
|
-
LLM_TENSOR_OUTPUT,
|
|
2004
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2005
|
-
LLM_TENSOR_ATTN_NORM_2,
|
|
2006
|
-
LLM_TENSOR_TIME_MIX_W0,
|
|
2007
|
-
LLM_TENSOR_TIME_MIX_W1,
|
|
2008
|
-
LLM_TENSOR_TIME_MIX_W2,
|
|
2009
|
-
LLM_TENSOR_TIME_MIX_A0,
|
|
2010
|
-
LLM_TENSOR_TIME_MIX_A1,
|
|
2011
|
-
LLM_TENSOR_TIME_MIX_A2,
|
|
2012
|
-
LLM_TENSOR_TIME_MIX_V0,
|
|
2013
|
-
LLM_TENSOR_TIME_MIX_V1,
|
|
2014
|
-
LLM_TENSOR_TIME_MIX_V2,
|
|
2015
|
-
LLM_TENSOR_TIME_MIX_G1,
|
|
2016
|
-
LLM_TENSOR_TIME_MIX_G2,
|
|
2017
|
-
LLM_TENSOR_TIME_MIX_K_K,
|
|
2018
|
-
LLM_TENSOR_TIME_MIX_K_A,
|
|
2019
|
-
LLM_TENSOR_TIME_MIX_R_K,
|
|
2020
|
-
LLM_TENSOR_TIME_MIX_LERP_FUSED,
|
|
2021
|
-
LLM_TENSOR_TIME_MIX_KEY,
|
|
2022
|
-
LLM_TENSOR_TIME_MIX_VALUE,
|
|
2023
|
-
LLM_TENSOR_TIME_MIX_RECEPTANCE,
|
|
2024
|
-
LLM_TENSOR_TIME_MIX_LN,
|
|
2025
|
-
LLM_TENSOR_TIME_MIX_OUTPUT,
|
|
2026
|
-
LLM_TENSOR_CHANNEL_MIX_LERP_K,
|
|
2027
|
-
LLM_TENSOR_CHANNEL_MIX_KEY,
|
|
2028
|
-
LLM_TENSOR_CHANNEL_MIX_VALUE,
|
|
2029
|
-
};
|
|
2030
|
-
case LLM_ARCH_ARWKV7:
|
|
2031
|
-
return {
|
|
2032
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2033
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
2034
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2035
|
-
LLM_TENSOR_OUTPUT,
|
|
2036
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2037
|
-
LLM_TENSOR_TIME_MIX_W0,
|
|
2038
|
-
LLM_TENSOR_TIME_MIX_W1,
|
|
2039
|
-
LLM_TENSOR_TIME_MIX_W2,
|
|
2040
|
-
LLM_TENSOR_TIME_MIX_A0,
|
|
2041
|
-
LLM_TENSOR_TIME_MIX_A1,
|
|
2042
|
-
LLM_TENSOR_TIME_MIX_A2,
|
|
2043
|
-
LLM_TENSOR_TIME_MIX_V0,
|
|
2044
|
-
LLM_TENSOR_TIME_MIX_V1,
|
|
2045
|
-
LLM_TENSOR_TIME_MIX_V2,
|
|
2046
|
-
LLM_TENSOR_TIME_MIX_G1,
|
|
2047
|
-
LLM_TENSOR_TIME_MIX_G2,
|
|
2048
|
-
LLM_TENSOR_TIME_MIX_K_K,
|
|
2049
|
-
LLM_TENSOR_TIME_MIX_K_A,
|
|
2050
|
-
LLM_TENSOR_TIME_MIX_R_K,
|
|
2051
|
-
LLM_TENSOR_TIME_MIX_LERP_FUSED,
|
|
2052
|
-
LLM_TENSOR_TIME_MIX_KEY,
|
|
2053
|
-
LLM_TENSOR_TIME_MIX_VALUE,
|
|
2054
|
-
LLM_TENSOR_TIME_MIX_RECEPTANCE,
|
|
2055
|
-
LLM_TENSOR_TIME_MIX_LN,
|
|
2056
|
-
LLM_TENSOR_TIME_MIX_OUTPUT,
|
|
2057
|
-
LLM_TENSOR_FFN_NORM,
|
|
2058
|
-
LLM_TENSOR_FFN_GATE,
|
|
2059
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2060
|
-
LLM_TENSOR_FFN_UP,
|
|
2061
|
-
};
|
|
2062
|
-
case LLM_ARCH_GRANITE_MOE:
|
|
2063
|
-
return {
|
|
2064
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2065
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2066
|
-
LLM_TENSOR_OUTPUT,
|
|
2067
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2068
|
-
LLM_TENSOR_ATTN_Q,
|
|
2069
|
-
LLM_TENSOR_ATTN_K,
|
|
2070
|
-
LLM_TENSOR_ATTN_V,
|
|
2071
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2072
|
-
LLM_TENSOR_FFN_NORM,
|
|
2073
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2074
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2075
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2076
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2077
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2078
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2079
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2080
|
-
};
|
|
2081
|
-
case LLM_ARCH_GRANITE_HYBRID:
|
|
2082
|
-
return {
|
|
2083
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2084
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2085
|
-
LLM_TENSOR_OUTPUT,
|
|
2086
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2087
|
-
LLM_TENSOR_SSM_IN,
|
|
2088
|
-
LLM_TENSOR_SSM_CONV1D,
|
|
2089
|
-
LLM_TENSOR_SSM_DT,
|
|
2090
|
-
LLM_TENSOR_SSM_A,
|
|
2091
|
-
LLM_TENSOR_SSM_D,
|
|
2092
|
-
LLM_TENSOR_SSM_NORM,
|
|
2093
|
-
LLM_TENSOR_SSM_OUT,
|
|
2094
|
-
LLM_TENSOR_ATTN_Q,
|
|
2095
|
-
LLM_TENSOR_ATTN_K,
|
|
2096
|
-
LLM_TENSOR_ATTN_V,
|
|
2097
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2098
|
-
LLM_TENSOR_FFN_NORM,
|
|
2099
|
-
LLM_TENSOR_FFN_GATE,
|
|
2100
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2101
|
-
LLM_TENSOR_FFN_UP,
|
|
2102
|
-
LLM_TENSOR_FFN_NORM,
|
|
2103
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2104
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2105
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2106
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2107
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2108
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2109
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2110
|
-
};
|
|
2111
|
-
case LLM_ARCH_WAVTOKENIZER_DEC:
|
|
2112
|
-
return {
|
|
2113
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2114
|
-
LLM_TENSOR_TOKEN_EMBD_NORM,
|
|
2115
|
-
LLM_TENSOR_CONV1D,
|
|
2116
|
-
LLM_TENSOR_CONVNEXT_DW,
|
|
2117
|
-
LLM_TENSOR_CONVNEXT_NORM,
|
|
2118
|
-
LLM_TENSOR_CONVNEXT_PW1,
|
|
2119
|
-
LLM_TENSOR_CONVNEXT_PW2,
|
|
2120
|
-
LLM_TENSOR_CONVNEXT_GAMMA,
|
|
2121
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2122
|
-
LLM_TENSOR_OUTPUT,
|
|
2123
|
-
LLM_TENSOR_POS_NET_CONV1,
|
|
2124
|
-
LLM_TENSOR_POS_NET_CONV2,
|
|
2125
|
-
LLM_TENSOR_POS_NET_NORM,
|
|
2126
|
-
LLM_TENSOR_POS_NET_NORM1,
|
|
2127
|
-
LLM_TENSOR_POS_NET_NORM2,
|
|
2128
|
-
LLM_TENSOR_POS_NET_ATTN_NORM,
|
|
2129
|
-
LLM_TENSOR_POS_NET_ATTN_Q,
|
|
2130
|
-
LLM_TENSOR_POS_NET_ATTN_K,
|
|
2131
|
-
LLM_TENSOR_POS_NET_ATTN_V,
|
|
2132
|
-
LLM_TENSOR_POS_NET_ATTN_OUT,
|
|
2133
|
-
};
|
|
2134
|
-
case LLM_ARCH_BAILINGMOE:
|
|
2135
|
-
return {
|
|
2136
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2137
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2138
|
-
LLM_TENSOR_OUTPUT,
|
|
2139
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
2140
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2141
|
-
LLM_TENSOR_ATTN_Q,
|
|
2142
|
-
LLM_TENSOR_ATTN_K,
|
|
2143
|
-
LLM_TENSOR_ATTN_V,
|
|
2144
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2145
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2146
|
-
LLM_TENSOR_FFN_NORM,
|
|
2147
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2148
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2149
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2150
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
2151
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2152
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2153
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2154
|
-
};
|
|
2155
|
-
case LLM_ARCH_BAILINGMOE2:
|
|
2156
|
-
return {
|
|
2157
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2158
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2159
|
-
LLM_TENSOR_OUTPUT,
|
|
2160
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2161
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2162
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2163
|
-
LLM_TENSOR_ATTN_QKV,
|
|
2164
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2165
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2166
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2167
|
-
LLM_TENSOR_FFN_NORM,
|
|
2168
|
-
LLM_TENSOR_FFN_GATE,
|
|
2169
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2170
|
-
LLM_TENSOR_FFN_UP,
|
|
2171
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2172
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2173
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2174
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2175
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2176
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2177
|
-
LLM_TENSOR_NEXTN_EH_PROJ,
|
|
2178
|
-
LLM_TENSOR_NEXTN_EMBED_TOKENS,
|
|
2179
|
-
LLM_TENSOR_NEXTN_ENORM,
|
|
2180
|
-
LLM_TENSOR_NEXTN_HNORM,
|
|
2181
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD,
|
|
2182
|
-
LLM_TENSOR_NEXTN_SHARED_HEAD_NORM,
|
|
2183
|
-
LLM_TENSOR_LAYER_OUT_NORM,
|
|
2184
|
-
};
|
|
2185
|
-
case LLM_ARCH_DOTS1:
|
|
2186
|
-
return {
|
|
2187
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2188
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2189
|
-
LLM_TENSOR_OUTPUT,
|
|
2190
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2191
|
-
LLM_TENSOR_ATTN_Q,
|
|
2192
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2193
|
-
LLM_TENSOR_ATTN_K,
|
|
2194
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2195
|
-
LLM_TENSOR_ATTN_V,
|
|
2196
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2197
|
-
LLM_TENSOR_FFN_NORM,
|
|
2198
|
-
LLM_TENSOR_FFN_GATE,
|
|
2199
|
-
LLM_TENSOR_FFN_UP,
|
|
2200
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2201
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2202
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2203
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2204
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2205
|
-
LLM_TENSOR_FFN_GATE_INP_SHEXP,
|
|
2206
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2207
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2208
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2209
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2210
|
-
};
|
|
2211
|
-
case LLM_ARCH_ERNIE4_5_MOE:
|
|
2212
|
-
return {
|
|
2213
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2214
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2215
|
-
LLM_TENSOR_OUTPUT,
|
|
2216
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2217
|
-
LLM_TENSOR_ATTN_Q,
|
|
2218
|
-
LLM_TENSOR_ATTN_K,
|
|
2219
|
-
LLM_TENSOR_ATTN_V,
|
|
2220
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2221
|
-
LLM_TENSOR_FFN_NORM,
|
|
2222
|
-
LLM_TENSOR_FFN_GATE,
|
|
2223
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2224
|
-
LLM_TENSOR_FFN_UP,
|
|
2225
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2226
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2227
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2228
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2229
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2230
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2231
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2232
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2233
|
-
};
|
|
2234
|
-
case LLM_ARCH_HUNYUAN_MOE:
|
|
2235
|
-
return {
|
|
2236
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2237
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2238
|
-
LLM_TENSOR_OUTPUT,
|
|
2239
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2240
|
-
LLM_TENSOR_ATTN_Q,
|
|
2241
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2242
|
-
LLM_TENSOR_ATTN_K,
|
|
2243
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2244
|
-
LLM_TENSOR_ATTN_V,
|
|
2245
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2246
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2247
|
-
LLM_TENSOR_FFN_NORM,
|
|
2248
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2249
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2250
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2251
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2252
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2253
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2254
|
-
};
|
|
2255
|
-
case LLM_ARCH_OPENAI_MOE:
|
|
2256
|
-
return {
|
|
2257
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2258
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2259
|
-
LLM_TENSOR_OUTPUT,
|
|
2260
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2261
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
2262
|
-
LLM_TENSOR_ATTN_Q,
|
|
2263
|
-
LLM_TENSOR_ATTN_K,
|
|
2264
|
-
LLM_TENSOR_ATTN_V,
|
|
2265
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2266
|
-
LLM_TENSOR_ATTN_SINKS,
|
|
2267
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2268
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2269
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2270
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2271
|
-
};
|
|
2272
|
-
case LLM_ARCH_LFM2:
|
|
2273
|
-
return {
|
|
2274
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2275
|
-
LLM_TENSOR_ATTN_Q,
|
|
2276
|
-
LLM_TENSOR_ATTN_K,
|
|
2277
|
-
LLM_TENSOR_ATTN_V,
|
|
2278
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2279
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2280
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2281
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2282
|
-
LLM_TENSOR_FFN_GATE,
|
|
2283
|
-
LLM_TENSOR_FFN_NORM,
|
|
2284
|
-
LLM_TENSOR_FFN_UP,
|
|
2285
|
-
LLM_TENSOR_SHORTCONV_CONV,
|
|
2286
|
-
LLM_TENSOR_SHORTCONV_INPROJ,
|
|
2287
|
-
LLM_TENSOR_SHORTCONV_OUTPROJ,
|
|
2288
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2289
|
-
LLM_TENSOR_OUTPUT_NORM_LFM2,
|
|
2290
|
-
LLM_TENSOR_OUTPUT,
|
|
2291
|
-
LLM_TENSOR_DENSE_2_OUT,
|
|
2292
|
-
};
|
|
2293
|
-
case LLM_ARCH_LFM2MOE:
|
|
2294
|
-
return {
|
|
2295
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2296
|
-
LLM_TENSOR_ATTN_Q,
|
|
2297
|
-
LLM_TENSOR_ATTN_K,
|
|
2298
|
-
LLM_TENSOR_ATTN_V,
|
|
2299
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2300
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2301
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2302
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2303
|
-
LLM_TENSOR_FFN_GATE,
|
|
2304
|
-
LLM_TENSOR_FFN_NORM,
|
|
2305
|
-
LLM_TENSOR_FFN_UP,
|
|
2306
|
-
LLM_TENSOR_SHORTCONV_CONV,
|
|
2307
|
-
LLM_TENSOR_SHORTCONV_INPROJ,
|
|
2308
|
-
LLM_TENSOR_SHORTCONV_OUTPROJ,
|
|
2309
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2310
|
-
LLM_TENSOR_OUTPUT_NORM_LFM2,
|
|
2311
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2312
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2313
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2314
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2315
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2316
|
-
};
|
|
2317
|
-
case LLM_ARCH_SMALLTHINKER:
|
|
2318
|
-
return {
|
|
2319
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2320
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2321
|
-
LLM_TENSOR_OUTPUT,
|
|
2322
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2323
|
-
LLM_TENSOR_ATTN_Q,
|
|
2324
|
-
LLM_TENSOR_ATTN_K,
|
|
2325
|
-
LLM_TENSOR_ATTN_V,
|
|
2326
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2327
|
-
LLM_TENSOR_FFN_NORM,
|
|
2328
|
-
LLM_TENSOR_FFN_GATE,
|
|
2329
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2330
|
-
LLM_TENSOR_FFN_UP,
|
|
2331
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2332
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2333
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2334
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2335
|
-
};
|
|
2336
|
-
case LLM_ARCH_APERTUS:
|
|
2337
|
-
return {
|
|
2338
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2339
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2340
|
-
LLM_TENSOR_OUTPUT,
|
|
2341
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
2342
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2343
|
-
LLM_TENSOR_ATTN_Q,
|
|
2344
|
-
LLM_TENSOR_ATTN_K,
|
|
2345
|
-
LLM_TENSOR_ATTN_V,
|
|
2346
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2347
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2348
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2349
|
-
LLM_TENSOR_FFN_NORM,
|
|
2350
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2351
|
-
LLM_TENSOR_FFN_UP,
|
|
2352
|
-
};
|
|
2353
|
-
case LLM_ARCH_SEED_OSS:
|
|
2354
|
-
return {
|
|
2355
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2356
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2357
|
-
LLM_TENSOR_OUTPUT,
|
|
2358
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2359
|
-
LLM_TENSOR_ATTN_Q,
|
|
2360
|
-
LLM_TENSOR_ATTN_K,
|
|
2361
|
-
LLM_TENSOR_ATTN_V,
|
|
2362
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2363
|
-
LLM_TENSOR_ATTN_POST_NORM,
|
|
2364
|
-
LLM_TENSOR_FFN_GATE,
|
|
2365
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2366
|
-
LLM_TENSOR_FFN_UP,
|
|
2367
|
-
};
|
|
2368
|
-
case LLM_ARCH_GROVEMOE:
|
|
2369
|
-
return {
|
|
2370
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2371
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2372
|
-
LLM_TENSOR_OUTPUT,
|
|
2373
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2374
|
-
LLM_TENSOR_ATTN_Q,
|
|
2375
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2376
|
-
LLM_TENSOR_ATTN_K,
|
|
2377
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2378
|
-
LLM_TENSOR_ATTN_V,
|
|
2379
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2380
|
-
LLM_TENSOR_FFN_NORM,
|
|
2381
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2382
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2383
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2384
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2385
|
-
LLM_TENSOR_FFN_GATE_CHEXPS,
|
|
2386
|
-
LLM_TENSOR_FFN_DOWN_CHEXPS,
|
|
2387
|
-
LLM_TENSOR_FFN_UP_CHEXPS,
|
|
2388
|
-
};
|
|
2389
|
-
case LLM_ARCH_MINIMAX_M2:
|
|
2390
|
-
return {
|
|
2391
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2392
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2393
|
-
LLM_TENSOR_OUTPUT,
|
|
2394
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2395
|
-
LLM_TENSOR_ATTN_Q,
|
|
2396
|
-
LLM_TENSOR_ATTN_K,
|
|
2397
|
-
LLM_TENSOR_ATTN_V,
|
|
2398
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2399
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2400
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2401
|
-
LLM_TENSOR_FFN_NORM,
|
|
2402
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2403
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2404
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2405
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2406
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2407
|
-
};
|
|
2408
|
-
case LLM_ARCH_COGVLM:
|
|
2409
|
-
return {
|
|
2410
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2411
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2412
|
-
LLM_TENSOR_OUTPUT,
|
|
2413
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2414
|
-
LLM_TENSOR_ATTN_QKV,
|
|
2415
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2416
|
-
LLM_TENSOR_FFN_NORM,
|
|
2417
|
-
LLM_TENSOR_FFN_GATE,
|
|
2418
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2419
|
-
LLM_TENSOR_FFN_UP,
|
|
2420
|
-
LLM_TENSOR_VISEXP_ATTN_QKV,
|
|
2421
|
-
LLM_TENSOR_VISEXP_ATTN_OUT,
|
|
2422
|
-
LLM_TENSOR_VISEXP_FFN_GATE,
|
|
2423
|
-
LLM_TENSOR_VISEXP_FFN_DOWN,
|
|
2424
|
-
LLM_TENSOR_VISEXP_FFN_UP,
|
|
2425
|
-
};
|
|
2426
|
-
case LLM_ARCH_MIMO2:
|
|
2427
|
-
return {
|
|
2428
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2429
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2430
|
-
LLM_TENSOR_OUTPUT,
|
|
2431
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2432
|
-
LLM_TENSOR_ATTN_Q,
|
|
2433
|
-
LLM_TENSOR_ATTN_K,
|
|
2434
|
-
LLM_TENSOR_ATTN_V,
|
|
2435
|
-
LLM_TENSOR_ATTN_SINKS,
|
|
2436
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2437
|
-
LLM_TENSOR_FFN_NORM,
|
|
2438
|
-
LLM_TENSOR_FFN_GATE,
|
|
2439
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2440
|
-
LLM_TENSOR_FFN_UP,
|
|
2441
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2442
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2443
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2444
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2445
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2446
|
-
};
|
|
2447
|
-
case LLM_ARCH_STEP35:
|
|
2448
|
-
return {
|
|
2449
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2450
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2451
|
-
LLM_TENSOR_OUTPUT,
|
|
2452
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
2453
|
-
LLM_TENSOR_ROPE_FACTORS_LONG,
|
|
2454
|
-
LLM_TENSOR_ROPE_FACTORS_SHORT,
|
|
2455
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2456
|
-
LLM_TENSOR_ATTN_Q,
|
|
2457
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2458
|
-
LLM_TENSOR_ATTN_K,
|
|
2459
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2460
|
-
LLM_TENSOR_ATTN_V,
|
|
2461
|
-
LLM_TENSOR_ATTN_GATE,
|
|
2462
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2463
|
-
LLM_TENSOR_FFN_NORM,
|
|
2464
|
-
LLM_TENSOR_FFN_GATE,
|
|
2465
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2466
|
-
LLM_TENSOR_FFN_UP,
|
|
2467
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2468
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2469
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2470
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2471
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2472
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2473
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2474
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2475
|
-
};
|
|
2476
|
-
case LLM_ARCH_GPTJ:
|
|
2477
|
-
case LLM_ARCH_UNKNOWN:
|
|
2478
|
-
return {
|
|
2479
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2480
|
-
};
|
|
2481
|
-
case LLM_ARCH_MAINCODER:
|
|
2482
|
-
return {
|
|
2483
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2484
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2485
|
-
LLM_TENSOR_OUTPUT,
|
|
2486
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2487
|
-
LLM_TENSOR_ATTN_Q,
|
|
2488
|
-
LLM_TENSOR_ATTN_Q_NORM,
|
|
2489
|
-
LLM_TENSOR_ATTN_K,
|
|
2490
|
-
LLM_TENSOR_ATTN_K_NORM,
|
|
2491
|
-
LLM_TENSOR_ATTN_V,
|
|
2492
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2493
|
-
LLM_TENSOR_FFN_NORM,
|
|
2494
|
-
LLM_TENSOR_FFN_GATE,
|
|
2495
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2496
|
-
LLM_TENSOR_FFN_UP,
|
|
2497
|
-
};
|
|
2498
|
-
case LLM_ARCH_KIMI_LINEAR:
|
|
2499
|
-
return {
|
|
2500
|
-
LLM_TENSOR_TOKEN_EMBD,
|
|
2501
|
-
LLM_TENSOR_OUTPUT_NORM,
|
|
2502
|
-
LLM_TENSOR_OUTPUT,
|
|
2503
|
-
LLM_TENSOR_ROPE_FREQS,
|
|
2504
|
-
LLM_TENSOR_ATTN_NORM,
|
|
2505
|
-
LLM_TENSOR_ATTN_Q,
|
|
2506
|
-
LLM_TENSOR_ATTN_K,
|
|
2507
|
-
LLM_TENSOR_ATTN_V,
|
|
2508
|
-
LLM_TENSOR_ATTN_OUT,
|
|
2509
|
-
LLM_TENSOR_FFN_NORM,
|
|
2510
|
-
// Dense FFN (layer 0 only)
|
|
2511
|
-
LLM_TENSOR_FFN_GATE,
|
|
2512
|
-
LLM_TENSOR_FFN_DOWN,
|
|
2513
|
-
LLM_TENSOR_FFN_UP,
|
|
2514
|
-
// MoE FFN (layers 1+)
|
|
2515
|
-
LLM_TENSOR_FFN_GATE_INP,
|
|
2516
|
-
LLM_TENSOR_FFN_GATE_EXPS,
|
|
2517
|
-
LLM_TENSOR_FFN_DOWN_EXPS,
|
|
2518
|
-
LLM_TENSOR_FFN_UP_EXPS,
|
|
2519
|
-
LLM_TENSOR_FFN_EXP_PROBS_B,
|
|
2520
|
-
// Shared experts
|
|
2521
|
-
LLM_TENSOR_FFN_GATE_SHEXP,
|
|
2522
|
-
LLM_TENSOR_FFN_DOWN_SHEXP,
|
|
2523
|
-
LLM_TENSOR_FFN_UP_SHEXP,
|
|
2524
|
-
// KDA (using SSM_ enum prefix, keeping GGUF names for backward compat)
|
|
2525
|
-
LLM_TENSOR_SSM_CONV1D_Q,
|
|
2526
|
-
LLM_TENSOR_SSM_CONV1D_K,
|
|
2527
|
-
LLM_TENSOR_SSM_CONV1D_V,
|
|
2528
|
-
LLM_TENSOR_SSM_F_A,
|
|
2529
|
-
LLM_TENSOR_SSM_F_B,
|
|
2530
|
-
LLM_TENSOR_SSM_BETA,
|
|
2531
|
-
LLM_TENSOR_SSM_A,
|
|
2532
|
-
LLM_TENSOR_SSM_G_A,
|
|
2533
|
-
LLM_TENSOR_SSM_G_B,
|
|
2534
|
-
LLM_TENSOR_SSM_DT,
|
|
2535
|
-
LLM_TENSOR_SSM_NORM,
|
|
2536
|
-
// MLA
|
|
2537
|
-
LLM_TENSOR_ATTN_Q_A,
|
|
2538
|
-
LLM_TENSOR_ATTN_Q_B,
|
|
2539
|
-
LLM_TENSOR_ATTN_Q_A_NORM,
|
|
2540
|
-
LLM_TENSOR_ATTN_KV_A_MQA,
|
|
2541
|
-
LLM_TENSOR_ATTN_KV_B,
|
|
2542
|
-
LLM_TENSOR_ATTN_K_B,
|
|
2543
|
-
LLM_TENSOR_ATTN_V_B,
|
|
2544
|
-
LLM_TENSOR_ATTN_KV_A_NORM,
|
|
2545
|
-
};
|
|
2546
|
-
default:
|
|
2547
|
-
GGML_ABORT("unknown architecture for tensor mapping");
|
|
2548
|
-
}
|
|
2549
|
-
}
|
|
2550
|
-
|
|
2551
|
-
// declare information about the model weight tensors:
|
|
2552
|
-
// - the layer in which the tensor is going to be used. this is needed in order to assign the correct buffer type for the weight
|
|
2553
|
-
// - the operator which is going to use the weight. this is needed to determine if the respective backend supports the operator
|
|
2554
|
-
//
|
|
2555
|
-
// for example, input layers are usually assigned to CPU/host buffer types
|
|
2556
|
-
//
|
|
2557
|
-
// a mismatch between the declared information and the actual layer/op in which the tensor is used can lead to sub-optimal
|
|
2558
|
-
// assignment of the buffer types and extra overhead during computation
|
|
2559
|
-
// example: https://github.com/ggml-org/llama.cpp/pull/17548
|
|
2560
|
-
//
|
|
2561
|
-
static const std::map<llm_tensor, llm_tensor_info> LLM_TENSOR_INFOS = {
|
|
2562
|
-
{LLM_TENSOR_TOKEN_EMBD, {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
|
|
2563
|
-
{LLM_TENSOR_POS_EMBD, {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
|
|
2564
|
-
{LLM_TENSOR_TOKEN_TYPES, {LLM_TENSOR_LAYER_INPUT, GGML_OP_GET_ROWS}},
|
|
2565
|
-
{LLM_TENSOR_TOKEN_EMBD_NORM, {LLM_TENSOR_LAYER_INPUT, GGML_OP_MUL}},
|
|
2566
|
-
{LLM_TENSOR_OUTPUT, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2567
|
-
{LLM_TENSOR_CLS, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2568
|
-
{LLM_TENSOR_CLS_OUT, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2569
|
-
{LLM_TENSOR_CLS_NORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2570
|
-
{LLM_TENSOR_DENSE_2_OUT, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}}, // Dense layer output
|
|
2571
|
-
{LLM_TENSOR_DENSE_3_OUT, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}}, // Dense layer output
|
|
2572
|
-
{LLM_TENSOR_OUTPUT_NORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2573
|
-
{LLM_TENSOR_OUTPUT_NORM_LFM2, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2574
|
-
{LLM_TENSOR_DEC_OUTPUT_NORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2575
|
-
{LLM_TENSOR_ENC_OUTPUT_NORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2576
|
-
{LLM_TENSOR_ROPE_FREQS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
|
|
2577
|
-
{LLM_TENSOR_ROPE_FACTORS_LONG, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
|
|
2578
|
-
{LLM_TENSOR_ROPE_FACTORS_SHORT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ROPE}},
|
|
2579
|
-
{LLM_TENSOR_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2580
|
-
{LLM_TENSOR_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2581
|
-
{LLM_TENSOR_ATTN_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2582
|
-
{LLM_TENSOR_ATTN_QKV, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2583
|
-
{LLM_TENSOR_ATTN_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2584
|
-
{LLM_TENSOR_ATTN_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2585
|
-
{LLM_TENSOR_FFN_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2586
|
-
{LLM_TENSOR_FFN_DOWN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2587
|
-
{LLM_TENSOR_FFN_UP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2588
|
-
{LLM_TENSOR_FFN_DOWN_SHEXP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2589
|
-
{LLM_TENSOR_FFN_GATE_SHEXP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2590
|
-
{LLM_TENSOR_FFN_UP_SHEXP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2591
|
-
{LLM_TENSOR_ATTN_Q_A, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2592
|
-
{LLM_TENSOR_ATTN_Q_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2593
|
-
{LLM_TENSOR_ATTN_KV_A_MQA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2594
|
-
{LLM_TENSOR_ATTN_KV_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2595
|
-
{LLM_TENSOR_ATTN_K_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2596
|
-
{LLM_TENSOR_ATTN_V_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2597
|
-
{LLM_TENSOR_ATTN_SINKS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SCALE}},
|
|
2598
|
-
{LLM_TENSOR_DEC_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2599
|
-
{LLM_TENSOR_DEC_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2600
|
-
{LLM_TENSOR_DEC_ATTN_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2601
|
-
{LLM_TENSOR_DEC_ATTN_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2602
|
-
{LLM_TENSOR_DEC_CROSS_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2603
|
-
{LLM_TENSOR_DEC_CROSS_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2604
|
-
{LLM_TENSOR_DEC_CROSS_ATTN_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2605
|
-
{LLM_TENSOR_DEC_CROSS_ATTN_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2606
|
-
{LLM_TENSOR_DEC_FFN_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2607
|
-
{LLM_TENSOR_DEC_FFN_DOWN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2608
|
-
{LLM_TENSOR_DEC_FFN_UP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2609
|
-
{LLM_TENSOR_ENC_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2610
|
-
{LLM_TENSOR_ENC_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2611
|
-
{LLM_TENSOR_ENC_ATTN_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2612
|
-
{LLM_TENSOR_ENC_ATTN_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2613
|
-
{LLM_TENSOR_ENC_FFN_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2614
|
-
{LLM_TENSOR_ENC_FFN_DOWN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2615
|
-
{LLM_TENSOR_ENC_FFN_UP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2616
|
-
{LLM_TENSOR_FFN_GATE_INP_SHEXP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2617
|
-
{LLM_TENSOR_FFN_GATE_INP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2618
|
-
{LLM_TENSOR_SSM_IN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2619
|
-
{LLM_TENSOR_SSM_X, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2620
|
-
{LLM_TENSOR_SSM_DT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2621
|
-
{LLM_TENSOR_SSM_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2622
|
-
{LLM_TENSOR_SSM_ALPHA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2623
|
-
{LLM_TENSOR_SSM_BETA_ALPHA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2624
|
-
{LLM_TENSOR_TIME_MIX_W1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2625
|
-
{LLM_TENSOR_TIME_MIX_W2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2626
|
-
{LLM_TENSOR_TIME_MIX_A1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2627
|
-
{LLM_TENSOR_TIME_MIX_A2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2628
|
-
{LLM_TENSOR_TIME_MIX_V1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2629
|
-
{LLM_TENSOR_TIME_MIX_V2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2630
|
-
{LLM_TENSOR_TIME_MIX_G1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2631
|
-
{LLM_TENSOR_TIME_MIX_G2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2632
|
-
{LLM_TENSOR_TIME_MIX_DECAY_W1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2633
|
-
{LLM_TENSOR_TIME_MIX_DECAY_W2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2634
|
-
{LLM_TENSOR_TIME_MIX_KEY, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2635
|
-
{LLM_TENSOR_TIME_MIX_VALUE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2636
|
-
{LLM_TENSOR_TIME_MIX_RECEPTANCE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2637
|
-
{LLM_TENSOR_TIME_MIX_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2638
|
-
{LLM_TENSOR_TIME_MIX_OUTPUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2639
|
-
{LLM_TENSOR_CHANNEL_MIX_KEY, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2640
|
-
{LLM_TENSOR_CHANNEL_MIX_RECEPTANCE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2641
|
-
{LLM_TENSOR_CHANNEL_MIX_VALUE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2642
|
-
{LLM_TENSOR_FFN_ACT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_DIV}},
|
|
2643
|
-
{LLM_TENSOR_SSM_CONV1D, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_CONV}},
|
|
2644
|
-
{LLM_TENSOR_SSM_A, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_SCAN}},
|
|
2645
|
-
{LLM_TENSOR_SSM_A_NOSCAN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}}, // a version of SSM_A used for MUL instead of SSM_SCAN
|
|
2646
|
-
{LLM_TENSOR_SSM_DT_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2647
|
-
{LLM_TENSOR_SSM_B_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2648
|
-
{LLM_TENSOR_SSM_C_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2649
|
-
{LLM_TENSOR_SSM_D, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2650
|
-
{LLM_TENSOR_SSM_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2651
|
-
// Kimi KDA - Conv tensors are 4D [d_conv, 1, d_inner, 1], reshaped to 2D at runtime
|
|
2652
|
-
{LLM_TENSOR_SSM_CONV1D_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2653
|
-
{LLM_TENSOR_SSM_CONV1D_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2654
|
-
{LLM_TENSOR_SSM_CONV1D_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2655
|
-
{LLM_TENSOR_SSM_F_A, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2656
|
-
{LLM_TENSOR_SSM_F_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2657
|
-
{LLM_TENSOR_SSM_BETA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2658
|
-
{LLM_TENSOR_SSM_G_A, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2659
|
-
{LLM_TENSOR_SSM_G_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2660
|
-
{LLM_TENSOR_TIME_MIX_LERP_X, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2661
|
-
{LLM_TENSOR_TIME_MIX_LN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2662
|
-
{LLM_TENSOR_CHANNEL_MIX_LERP_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2663
|
-
{LLM_TENSOR_CHANNEL_MIX_LERP_R, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2664
|
-
{LLM_TENSOR_TIME_MIX_K_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2665
|
-
{LLM_TENSOR_TIME_MIX_K_A, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2666
|
-
{LLM_TENSOR_TIME_MIX_R_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2667
|
-
{LLM_TENSOR_TIME_MIX_LERP_W, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2668
|
-
{LLM_TENSOR_TIME_MIX_LERP_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2669
|
-
{LLM_TENSOR_TIME_MIX_LERP_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2670
|
-
{LLM_TENSOR_TIME_MIX_LERP_R, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2671
|
-
{LLM_TENSOR_TIME_MIX_LERP_G, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2672
|
-
{LLM_TENSOR_TIME_MIX_LERP_FUSED, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2673
|
-
{LLM_TENSOR_TIME_MIX_DECAY, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2674
|
-
{LLM_TENSOR_TIME_MIX_W0, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2675
|
-
{LLM_TENSOR_TIME_MIX_A0, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2676
|
-
{LLM_TENSOR_TIME_MIX_V0, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2677
|
-
{LLM_TENSOR_TIME_MIX_FIRST, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_RWKV_WKV6}},
|
|
2678
|
-
{LLM_TENSOR_ATTN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2679
|
-
{LLM_TENSOR_ATTN_NORM_2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2680
|
-
{LLM_TENSOR_ATTN_OUT_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2681
|
-
{LLM_TENSOR_ATTN_POST_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2682
|
-
{LLM_TENSOR_FFN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2683
|
-
{LLM_TENSOR_FFN_POST_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2684
|
-
{LLM_TENSOR_FFN_NORM_EXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2685
|
-
{LLM_TENSOR_ATTN_Q_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2686
|
-
{LLM_TENSOR_ATTN_K_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2687
|
-
{LLM_TENSOR_LAYER_OUT_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2688
|
-
{LLM_TENSOR_ATTN_Q_A_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2689
|
-
{LLM_TENSOR_ATTN_KV_A_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2690
|
-
{LLM_TENSOR_ATTN_SUB_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2691
|
-
{LLM_TENSOR_FFN_SUB_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2692
|
-
{LLM_TENSOR_DEC_ATTN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2693
|
-
{LLM_TENSOR_DEC_CROSS_ATTN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2694
|
-
{LLM_TENSOR_DEC_FFN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2695
|
-
{LLM_TENSOR_ENC_ATTN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2696
|
-
{LLM_TENSOR_ENC_FFN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2697
|
-
{LLM_TENSOR_DEC_ATTN_REL_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}},
|
|
2698
|
-
{LLM_TENSOR_ENC_ATTN_REL_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_GET_ROWS}},
|
|
2699
|
-
{LLM_TENSOR_FFN_DOWN_EXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2700
|
-
{LLM_TENSOR_FFN_GATE_EXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2701
|
-
{LLM_TENSOR_FFN_UP_EXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2702
|
-
{LLM_TENSOR_FFN_GATE_UP_EXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2703
|
-
{LLM_TENSOR_FFN_DOWN_CHEXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2704
|
-
{LLM_TENSOR_FFN_GATE_CHEXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2705
|
-
{LLM_TENSOR_FFN_UP_CHEXPS, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT_ID}},
|
|
2706
|
-
{LLM_TENSOR_FFN_EXP_PROBS_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_ADD}},
|
|
2707
|
-
// altup / laurel (gemma 3n)
|
|
2708
|
-
{LLM_TENSOR_PER_LAYER_TOKEN_EMBD, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_GET_ROWS}},
|
|
2709
|
-
{LLM_TENSOR_PER_LAYER_MODEL_PROJ, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2710
|
-
{LLM_TENSOR_PER_LAYER_PROJ_NORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2711
|
-
{LLM_TENSOR_ALTUP_PROJ, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2712
|
-
{LLM_TENSOR_ALTUP_UNEMBD_PROJ, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2713
|
-
{LLM_TENSOR_PER_LAYER_INP_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2714
|
-
{LLM_TENSOR_PER_LAYER_PROJ, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2715
|
-
{LLM_TENSOR_PER_LAYER_POST_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2716
|
-
{LLM_TENSOR_ALTUP_CORRECT_COEF, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2717
|
-
{LLM_TENSOR_ALTUP_CORRECT_SCALE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2718
|
-
{LLM_TENSOR_ALTUP_PREDICT_COEF, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2719
|
-
{LLM_TENSOR_ALTUP_ROUTER, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2720
|
-
{LLM_TENSOR_ALTUP_ROUTER_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2721
|
-
{LLM_TENSOR_LAUREL_L, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2722
|
-
{LLM_TENSOR_LAUREL_R, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2723
|
-
{LLM_TENSOR_LAUREL_POST_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2724
|
-
// this tensor is loaded for T5, but never used
|
|
2725
|
-
{LLM_TENSOR_DEC_CROSS_ATTN_REL_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_NONE}},
|
|
2726
|
-
{LLM_TENSOR_CONV1D, {LLM_TENSOR_LAYER_INPUT, GGML_OP_IM2COL}},
|
|
2727
|
-
{LLM_TENSOR_POS_NET_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2728
|
-
{LLM_TENSOR_POS_NET_NORM1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2729
|
-
{LLM_TENSOR_POS_NET_NORM2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2730
|
-
{LLM_TENSOR_POS_NET_CONV1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_IM2COL}},
|
|
2731
|
-
{LLM_TENSOR_POS_NET_CONV2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_IM2COL}},
|
|
2732
|
-
{LLM_TENSOR_POS_NET_ATTN_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2733
|
-
{LLM_TENSOR_POS_NET_ATTN_Q, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2734
|
-
{LLM_TENSOR_POS_NET_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2735
|
-
{LLM_TENSOR_POS_NET_ATTN_V, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2736
|
-
{LLM_TENSOR_POS_NET_ATTN_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2737
|
-
{LLM_TENSOR_CONVNEXT_DW, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_IM2COL}},
|
|
2738
|
-
{LLM_TENSOR_CONVNEXT_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2739
|
-
{LLM_TENSOR_CONVNEXT_PW1, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2740
|
-
{LLM_TENSOR_CONVNEXT_PW2, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2741
|
-
{LLM_TENSOR_CONVNEXT_GAMMA, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2742
|
-
{LLM_TENSOR_SHORTCONV_CONV, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_SSM_CONV}},
|
|
2743
|
-
{LLM_TENSOR_SHORTCONV_INPROJ, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2744
|
-
{LLM_TENSOR_SHORTCONV_OUTPROJ, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2745
|
-
{LLM_TENSOR_VISEXP_ATTN_QKV, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2746
|
-
{LLM_TENSOR_VISEXP_ATTN_OUT, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2747
|
-
{LLM_TENSOR_VISEXP_FFN_GATE, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2748
|
-
{LLM_TENSOR_VISEXP_FFN_DOWN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2749
|
-
{LLM_TENSOR_VISEXP_FFN_UP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2750
|
-
{LLM_TENSOR_INDEXER_K_NORM, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2751
|
-
{LLM_TENSOR_INDEXER_PROJ, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2752
|
-
{LLM_TENSOR_INDEXER_ATTN_K, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2753
|
-
{LLM_TENSOR_INDEXER_ATTN_Q_B, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL_MAT}},
|
|
2754
|
-
// NextN/MTP tensors are currently ignored (reserved for future MTP support)
|
|
2755
|
-
// These tensors only exist in the last layer(s) and are treated as output tensors
|
|
2756
|
-
{LLM_TENSOR_NEXTN_EH_PROJ, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2757
|
-
{LLM_TENSOR_NEXTN_EMBED_TOKENS, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_GET_ROWS}},
|
|
2758
|
-
{LLM_TENSOR_NEXTN_ENORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_GET_ROWS}},
|
|
2759
|
-
{LLM_TENSOR_NEXTN_HNORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2760
|
-
{LLM_TENSOR_NEXTN_SHARED_HEAD_HEAD, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL_MAT}},
|
|
2761
|
-
{LLM_TENSOR_NEXTN_SHARED_HEAD_NORM, {LLM_TENSOR_LAYER_OUTPUT, GGML_OP_MUL}},
|
|
2762
|
-
// Nemotron 3 Super
|
|
2763
|
-
{LLM_TENSOR_FFN_LATENT_DOWN, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2764
|
-
{LLM_TENSOR_FFN_LATENT_UP, {LLM_TENSOR_LAYER_REPEATING, GGML_OP_MUL}},
|
|
2765
|
-
};
|
|
2766
|
-
|
|
2767
|
-
LLM_KV::LLM_KV(llm_arch arch, const char * suffix) : arch(arch), suffix(suffix) {}
|
|
2768
|
-
|
|
2769
|
-
std::string LLM_KV::operator()(llm_kv kv) const {
|
|
2770
|
-
std::string name = ::format(LLM_KV_NAMES.at(kv), LLM_ARCH_NAMES.at(arch));
|
|
2771
|
-
|
|
2772
|
-
if (suffix != nullptr) {
|
|
2773
|
-
name += ".";
|
|
2774
|
-
name += suffix;
|
|
2775
|
-
}
|
|
2776
|
-
|
|
2777
|
-
return name;
|
|
2778
|
-
}
|
|
2779
|
-
|
|
2780
|
-
LLM_TN_IMPL::LLM_TN_IMPL(llm_arch arch, llm_tensor tensor, const char * suffix, int bid, int xid)
|
|
2781
|
-
: arch(arch), tensor(tensor), suffix(suffix), bid(bid), xid(xid),
|
|
2782
|
-
model_tensors(llm_get_tensor_names(arch)) {}
|
|
2783
|
-
|
|
2784
|
-
std::string LLM_TN_IMPL::str() const {
|
|
2785
|
-
if (LLM_TENSOR_NAMES.find(tensor) == LLM_TENSOR_NAMES.end()) {
|
|
2786
|
-
GGML_ABORT("unknown tensor name for tensor id %d", static_cast<int>(tensor));
|
|
2787
|
-
}
|
|
2788
|
-
|
|
2789
|
-
if (model_tensors.find(tensor) == model_tensors.end()) {
|
|
2790
|
-
return LLM_TENSOR_NAMES.at(tensor);
|
|
2791
|
-
}
|
|
2792
|
-
|
|
2793
|
-
std::string name = ::format(LLM_TENSOR_NAMES.at(tensor), bid, xid);
|
|
2794
|
-
if (suffix != nullptr) {
|
|
2795
|
-
name += ".";
|
|
2796
|
-
name += suffix;
|
|
2797
|
-
}
|
|
2798
|
-
|
|
2799
|
-
return name;
|
|
2800
|
-
}
|
|
2801
|
-
|
|
2802
|
-
std::vector<llm_arch> llm_arch_all() {
|
|
2803
|
-
std::vector<llm_arch> ret;
|
|
2804
|
-
ret.reserve(LLM_ARCH_NAMES.size());
|
|
2805
|
-
for (const auto & [arch, _] : LLM_ARCH_NAMES) {
|
|
2806
|
-
ret.push_back(arch);
|
|
2807
|
-
}
|
|
2808
|
-
return ret;
|
|
2809
|
-
}
|
|
2810
|
-
|
|
2811
|
-
const char * llm_arch_name(llm_arch arch) {
|
|
2812
|
-
auto it = LLM_ARCH_NAMES.find(arch);
|
|
2813
|
-
if (it == LLM_ARCH_NAMES.end()) {
|
|
2814
|
-
return "unknown";
|
|
2815
|
-
}
|
|
2816
|
-
return it->second;
|
|
2817
|
-
}
|
|
2818
|
-
|
|
2819
|
-
llm_arch llm_arch_from_string(const std::string & name) {
|
|
2820
|
-
for (const auto & kv : LLM_ARCH_NAMES) { // NOLINT
|
|
2821
|
-
if (kv.second == name) {
|
|
2822
|
-
return kv.first;
|
|
2823
|
-
}
|
|
2824
|
-
}
|
|
2825
|
-
|
|
2826
|
-
return LLM_ARCH_UNKNOWN;
|
|
2827
|
-
}
|
|
2828
|
-
|
|
2829
|
-
const llm_tensor_info & llm_tensor_info_for(llm_tensor tensor) {
|
|
2830
|
-
return LLM_TENSOR_INFOS.at(tensor);
|
|
2831
|
-
}
|
|
2832
|
-
|
|
2833
|
-
bool llm_arch_is_recurrent(const llm_arch & arch) {
|
|
2834
|
-
switch (arch) {
|
|
2835
|
-
case LLM_ARCH_MAMBA:
|
|
2836
|
-
case LLM_ARCH_MAMBA2:
|
|
2837
|
-
case LLM_ARCH_RWKV6:
|
|
2838
|
-
case LLM_ARCH_RWKV6QWEN2:
|
|
2839
|
-
case LLM_ARCH_RWKV7:
|
|
2840
|
-
case LLM_ARCH_ARWKV7:
|
|
2841
|
-
return true;
|
|
2842
|
-
default:
|
|
2843
|
-
return false;
|
|
2844
|
-
}
|
|
2845
|
-
}
|
|
2846
|
-
|
|
2847
|
-
bool llm_arch_is_hybrid(const llm_arch & arch) {
|
|
2848
|
-
switch (arch) {
|
|
2849
|
-
case LLM_ARCH_JAMBA:
|
|
2850
|
-
case LLM_ARCH_FALCON_H1:
|
|
2851
|
-
case LLM_ARCH_PLAMO2:
|
|
2852
|
-
case LLM_ARCH_GRANITE_HYBRID:
|
|
2853
|
-
case LLM_ARCH_LFM2:
|
|
2854
|
-
case LLM_ARCH_LFM2MOE:
|
|
2855
|
-
case LLM_ARCH_NEMOTRON_H:
|
|
2856
|
-
case LLM_ARCH_NEMOTRON_H_MOE:
|
|
2857
|
-
case LLM_ARCH_QWEN3NEXT:
|
|
2858
|
-
case LLM_ARCH_KIMI_LINEAR:
|
|
2859
|
-
case LLM_ARCH_QWEN35:
|
|
2860
|
-
case LLM_ARCH_QWEN35MOE:
|
|
2861
|
-
return true;
|
|
2862
|
-
default:
|
|
2863
|
-
return false;
|
|
2864
|
-
}
|
|
2865
|
-
}
|
|
2866
|
-
|
|
2867
|
-
bool llm_arch_is_diffusion(const llm_arch & arch) {
|
|
2868
|
-
switch (arch) {
|
|
2869
|
-
case LLM_ARCH_DREAM:
|
|
2870
|
-
case LLM_ARCH_LLADA:
|
|
2871
|
-
case LLM_ARCH_LLADA_MOE:
|
|
2872
|
-
case LLM_ARCH_RND1:
|
|
2873
|
-
return true;
|
|
2874
|
-
default:
|
|
2875
|
-
return false;
|
|
2876
|
-
}
|
|
2877
|
-
}
|