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
|
@@ -0,0 +1,95 @@
|
|
|
1
|
+
{
|
|
2
|
+
"version": 4,
|
|
3
|
+
"configurePresets": [
|
|
4
|
+
{
|
|
5
|
+
"name": "base",
|
|
6
|
+
"hidden": true,
|
|
7
|
+
"generator": "Ninja",
|
|
8
|
+
"binaryDir": "${sourceDir}/build-${presetName}",
|
|
9
|
+
"cacheVariables": {
|
|
10
|
+
"CMAKE_EXPORT_COMPILE_COMMANDS": "ON",
|
|
11
|
+
"CMAKE_INSTALL_RPATH": "$ORIGIN;$ORIGIN/.."
|
|
12
|
+
}
|
|
13
|
+
},
|
|
14
|
+
{
|
|
15
|
+
"name": "sycl-base",
|
|
16
|
+
"hidden": true,
|
|
17
|
+
"generator": "Ninja",
|
|
18
|
+
"binaryDir": "${sourceDir}/build-${presetName}",
|
|
19
|
+
"cacheVariables": {
|
|
20
|
+
"CMAKE_EXPORT_COMPILE_COMMANDS": "ON",
|
|
21
|
+
"CMAKE_CXX_COMPILER": "icx",
|
|
22
|
+
"CMAKE_C_COMPILER": "cl",
|
|
23
|
+
"GGML_SYCL": "ON",
|
|
24
|
+
"CMAKE_INSTALL_RPATH": "$ORIGIN;$ORIGIN/.."
|
|
25
|
+
}
|
|
26
|
+
},
|
|
27
|
+
{ "name": "debug", "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "Debug" } },
|
|
28
|
+
{ "name": "release", "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "Release" } },
|
|
29
|
+
{ "name": "reldbg", "hidden": true, "cacheVariables": { "CMAKE_BUILD_TYPE": "RelWithDebInfo" } },
|
|
30
|
+
{ "name": "static", "hidden": true, "cacheVariables": { "GGML_STATIC": "ON" } },
|
|
31
|
+
{ "name": "sycl_f16", "hidden": true, "cacheVariables": { "GGML_SYCL_F16": "ON" } },
|
|
32
|
+
{ "name": "vulkan", "hidden": true, "cacheVariables": { "GGML_VULKAN": "ON" } },
|
|
33
|
+
|
|
34
|
+
{
|
|
35
|
+
"name": "x64-windows-llvm", "hidden": true,
|
|
36
|
+
"cacheVariables": {
|
|
37
|
+
"CMAKE_TOOLCHAIN_FILE": "${sourceDir}/cmake/x64-windows-llvm.cmake"
|
|
38
|
+
}
|
|
39
|
+
},
|
|
40
|
+
|
|
41
|
+
{
|
|
42
|
+
"name": "arm64-windows-llvm", "hidden": true,
|
|
43
|
+
"architecture": { "value": "arm64", "strategy": "external" },
|
|
44
|
+
"toolset": { "value": "host=x64", "strategy": "external" },
|
|
45
|
+
"cacheVariables": {
|
|
46
|
+
"CMAKE_TOOLCHAIN_FILE": "${sourceDir}/cmake/arm64-windows-llvm.cmake"
|
|
47
|
+
}
|
|
48
|
+
},
|
|
49
|
+
|
|
50
|
+
{
|
|
51
|
+
"name": "arm64-apple-clang", "hidden": true,
|
|
52
|
+
"architecture": { "value": "arm64", "strategy": "external" },
|
|
53
|
+
"toolset": { "value": "host=x64", "strategy": "external" },
|
|
54
|
+
"cacheVariables": {
|
|
55
|
+
"CMAKE_TOOLCHAIN_FILE": "${sourceDir}/cmake/arm64-apple-clang.cmake"
|
|
56
|
+
}
|
|
57
|
+
},
|
|
58
|
+
{
|
|
59
|
+
"name": "x64-linux-gcc", "hidden": true,
|
|
60
|
+
"cacheVariables": {
|
|
61
|
+
"CMAKE_C_COMPILER": "gcc",
|
|
62
|
+
"CMAKE_CXX_COMPILER": "g++"
|
|
63
|
+
}
|
|
64
|
+
},
|
|
65
|
+
{ "name": "x64-linux-gcc-debug", "inherits": [ "base", "x64-linux-gcc", "debug" ] },
|
|
66
|
+
{ "name": "x64-linux-gcc-release", "inherits": [ "base", "x64-linux-gcc", "release" ] },
|
|
67
|
+
{ "name": "x64-linux-gcc-reldbg", "inherits": [ "base", "x64-linux-gcc", "reldbg" ] },
|
|
68
|
+
{ "name": "x64-linux-gcc+static-release", "inherits": [ "base", "x64-linux-gcc", "release", "static" ] },
|
|
69
|
+
|
|
70
|
+
{ "name": "arm64-windows-llvm-debug", "inherits": [ "base", "arm64-windows-llvm", "debug" ] },
|
|
71
|
+
{ "name": "arm64-windows-llvm-release", "inherits": [ "base", "arm64-windows-llvm", "reldbg" ] },
|
|
72
|
+
{ "name": "arm64-windows-llvm+static-release", "inherits": [ "base", "arm64-windows-llvm", "reldbg", "static" ] },
|
|
73
|
+
|
|
74
|
+
{ "name": "arm64-apple-clang-debug", "inherits": [ "base", "arm64-apple-clang", "debug" ] },
|
|
75
|
+
{ "name": "arm64-apple-clang-release", "inherits": [ "base", "arm64-apple-clang", "reldbg" ] },
|
|
76
|
+
{ "name": "arm64-apple-clang+static-release", "inherits": [ "base", "arm64-apple-clang", "reldbg", "static" ] },
|
|
77
|
+
|
|
78
|
+
{ "name": "x64-windows-llvm-debug", "inherits": [ "base", "x64-windows-llvm", "debug" ] },
|
|
79
|
+
{ "name": "x64-windows-llvm-release", "inherits": [ "base", "x64-windows-llvm", "release" ] },
|
|
80
|
+
{ "name": "x64-windows-llvm-reldbg", "inherits": [ "base", "x64-windows-llvm", "reldbg" ] },
|
|
81
|
+
{ "name": "x64-windows-llvm+static-release", "inherits": [ "base", "x64-windows-llvm", "reldbg", "static" ] },
|
|
82
|
+
|
|
83
|
+
{ "name": "x64-windows-msvc-debug", "inherits": [ "base", "debug" ] },
|
|
84
|
+
{ "name": "x64-windows-msvc-release", "inherits": [ "base", "reldbg" ] },
|
|
85
|
+
{ "name": "x64-windows-msvc+static-release", "inherits": [ "base", "reldbg", "static" ] },
|
|
86
|
+
|
|
87
|
+
{ "name": "x64-windows-sycl-debug", "inherits": [ "sycl-base", "debug" ] },
|
|
88
|
+
{ "name": "x64-windows-sycl-debug-f16", "inherits": [ "sycl-base", "debug", "sycl_f16" ] },
|
|
89
|
+
{ "name": "x64-windows-sycl-release", "inherits": [ "sycl-base", "release" ] },
|
|
90
|
+
{ "name": "x64-windows-sycl-release-f16", "inherits": [ "sycl-base", "release", "sycl_f16" ] },
|
|
91
|
+
|
|
92
|
+
{ "name": "x64-windows-vulkan-debug", "inherits": [ "base", "vulkan", "debug" ] },
|
|
93
|
+
{ "name": "x64-windows-vulkan-release", "inherits": [ "base", "vulkan", "release" ] }
|
|
94
|
+
]
|
|
95
|
+
}
|
|
@@ -0,0 +1,30 @@
|
|
|
1
|
+
set(PARAKEET_VERSION @WHISPER_INSTALL_VERSION@)
|
|
2
|
+
set(PARAKEET_BUILD_COMMIT @WHISPER_BUILD_COMMIT@)
|
|
3
|
+
set(PARAKEET_BUILD_NUMBER @WHISPER_BUILD_NUMBER@)
|
|
4
|
+
set(PARAKEET_SHARED_LIB @BUILD_SHARED_LIBS@)
|
|
5
|
+
|
|
6
|
+
@PACKAGE_INIT@
|
|
7
|
+
|
|
8
|
+
set_and_check(PARAKEET_INCLUDE_DIR "@PACKAGE_PARAKEET_INCLUDE_INSTALL_DIR@")
|
|
9
|
+
set_and_check(PARAKEET_LIB_DIR "@PACKAGE_PARAKEET_LIB_INSTALL_DIR@")
|
|
10
|
+
set_and_check(PARAKEET_BIN_DIR "@PACKAGE_PARAKEET_BIN_INSTALL_DIR@")
|
|
11
|
+
|
|
12
|
+
find_package(ggml REQUIRED HINTS ${PARAKEET_LIB_DIR}/cmake)
|
|
13
|
+
|
|
14
|
+
find_library(parakeet_LIBRARY parakeet
|
|
15
|
+
REQUIRED
|
|
16
|
+
HINTS ${PARAKEET_LIB_DIR}
|
|
17
|
+
NO_CMAKE_FIND_ROOT_PATH
|
|
18
|
+
)
|
|
19
|
+
|
|
20
|
+
add_library(parakeet UNKNOWN IMPORTED)
|
|
21
|
+
set_target_properties(parakeet
|
|
22
|
+
PROPERTIES
|
|
23
|
+
INTERFACE_INCLUDE_DIRECTORIES "${PARAKEET_INCLUDE_DIR}"
|
|
24
|
+
INTERFACE_LINK_LIBRARIES "ggml::ggml;ggml::ggml-base;"
|
|
25
|
+
IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
|
|
26
|
+
IMPORTED_LOCATION "${parakeet_LIBRARY}"
|
|
27
|
+
INTERFACE_COMPILE_FEATURES cxx_std_11
|
|
28
|
+
POSITION_INDEPENDENT_CODE ON)
|
|
29
|
+
|
|
30
|
+
check_required_components(parakeet)
|
|
@@ -0,0 +1,10 @@
|
|
|
1
|
+
prefix=@CMAKE_INSTALL_PREFIX@
|
|
2
|
+
exec_prefix=${prefix}
|
|
3
|
+
libdir=${prefix}/@CMAKE_INSTALL_LIBDIR@
|
|
4
|
+
includedir=${prefix}/include
|
|
5
|
+
|
|
6
|
+
Name: parakeet
|
|
7
|
+
Description: Port of NVIDIA's Parakeet model in C/C++
|
|
8
|
+
Version: @PROJECT_VERSION@
|
|
9
|
+
Libs: -L${libdir} -lggml -lggml-base -lparakeet
|
|
10
|
+
Cflags: -I${includedir}
|
|
@@ -20,7 +20,7 @@ set(TARGET common)
|
|
|
20
20
|
|
|
21
21
|
unset(COMMON_EXTRA_LIBS)
|
|
22
22
|
|
|
23
|
-
if (
|
|
23
|
+
if (WHISPER_COMMON_FFMPEG)
|
|
24
24
|
# As of cmake 3.27, there is no official cmake support for FindFFmpeg.
|
|
25
25
|
# Consequnelty we added a FindFFmpeg.cmake script the cmake subfolder:
|
|
26
26
|
# whisper.cpp does not need the full ffmpeg libs, just AVFORMAT AVCODEC AVUTIL SWRESAMPLE
|
|
@@ -39,7 +39,7 @@ if (WHISPER_FFMPEG)
|
|
|
39
39
|
message(STATUS "Found avformat ${AVFORMAT_VERSION}")
|
|
40
40
|
|
|
41
41
|
include_directories(${FFMPEG_INCLUDE_DIRS})
|
|
42
|
-
add_compile_definitions(
|
|
42
|
+
add_compile_definitions(WHISPER_COMMON_FFMPEG)
|
|
43
43
|
|
|
44
44
|
list(APPEND COMMON_EXTRA_LIBS ${FFMPEG_LIBRARIES})
|
|
45
45
|
|
|
@@ -107,6 +107,8 @@ else()
|
|
|
107
107
|
add_subdirectory(server)
|
|
108
108
|
add_subdirectory(quantize)
|
|
109
109
|
add_subdirectory(vad-speech-segments)
|
|
110
|
+
add_subdirectory(parakeet-cli)
|
|
111
|
+
add_subdirectory(parakeet-quantize)
|
|
110
112
|
if (WHISPER_SDL2)
|
|
111
113
|
add_subdirectory(stream)
|
|
112
114
|
add_subdirectory(command)
|
|
@@ -157,7 +157,7 @@ static int whisper_bench_full(const whisper_params & params) {
|
|
|
157
157
|
fprintf(stderr, "\n");
|
|
158
158
|
fprintf(stderr, "If you wish, you can submit these results here:\n");
|
|
159
159
|
fprintf(stderr, "\n");
|
|
160
|
-
fprintf(stderr, " https://github.com/
|
|
160
|
+
fprintf(stderr, " https://github.com/ggml-org/whisper.cpp/issues/89\n");
|
|
161
161
|
fprintf(stderr, "\n");
|
|
162
162
|
fprintf(stderr, "Please include the following information:\n");
|
|
163
163
|
fprintf(stderr, "\n");
|
|
@@ -151,6 +151,10 @@ static bool whisper_params_parse(int argc, char ** argv, whisper_params & params
|
|
|
151
151
|
whisper_print_usage(argc, argv, params);
|
|
152
152
|
exit(0);
|
|
153
153
|
}
|
|
154
|
+
if (arg == "--version") {
|
|
155
|
+
fprintf(stdout, "whisper.cpp version: %s\n", whisper_version());
|
|
156
|
+
exit(0);
|
|
157
|
+
}
|
|
154
158
|
#define ARGV_NEXT (((i + 1) < argc) ? argv[++i] : requires_value_error(arg))
|
|
155
159
|
else if (arg == "-t" || arg == "--threads") { params.n_threads = std::stoi(ARGV_NEXT); }
|
|
156
160
|
else if (arg == "-p" || arg == "--processors") { params.n_processors = std::stoi(ARGV_NEXT); }
|
|
@@ -234,6 +238,7 @@ static void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params
|
|
|
234
238
|
fprintf(stderr, "\n");
|
|
235
239
|
fprintf(stderr, "options:\n");
|
|
236
240
|
fprintf(stderr, " -h, --help [default] show this help message and exit\n");
|
|
241
|
+
fprintf(stderr, " --version show version information and exit\n");
|
|
237
242
|
fprintf(stderr, " -t N, --threads N [%-7d] number of threads to use during computation\n", params.n_threads);
|
|
238
243
|
fprintf(stderr, " -p N, --processors N [%-7d] number of processors to use during computation\n", params.n_processors);
|
|
239
244
|
fprintf(stderr, " -ot N, --offset-t N [%-7d] time offset in milliseconds\n", params.offset_t_ms);
|
|
@@ -738,18 +743,47 @@ static void output_json(
|
|
|
738
743
|
if (full) {
|
|
739
744
|
start_arr("tokens");
|
|
740
745
|
const int n = whisper_full_n_tokens(ctx, i);
|
|
741
|
-
|
|
742
|
-
|
|
746
|
+
|
|
747
|
+
// Merge adjacent tokens whose bytes together form a
|
|
748
|
+
// single UTF-8 codepoint. Multi-byte characters (CJK
|
|
749
|
+
// in particular) can end up split across whisper
|
|
750
|
+
// tokens, which used to produce invalid UTF-8 in the
|
|
751
|
+
// JSON string. Refs issue #1798.
|
|
752
|
+
struct merged_token {
|
|
753
|
+
std::string text;
|
|
754
|
+
whisper_token_data data;
|
|
755
|
+
int64_t t1;
|
|
756
|
+
};
|
|
757
|
+
std::vector<merged_token> merged;
|
|
758
|
+
merged.reserve(n);
|
|
759
|
+
for (int j = 0; j < n; ) {
|
|
760
|
+
auto tok = whisper_full_get_token_data(ctx, i, j);
|
|
761
|
+
merged_token m{ whisper_token_to_str(ctx, tok.id), tok, tok.t1 };
|
|
762
|
+
++j;
|
|
763
|
+
while (j < n && utf8_trailing_bytes_needed(m.text) > 0) {
|
|
764
|
+
auto tok_next = whisper_full_get_token_data(ctx, i, j);
|
|
765
|
+
m.text += whisper_token_to_str(ctx, tok_next.id);
|
|
766
|
+
if (tok_next.t1 > -1) {
|
|
767
|
+
m.t1 = tok_next.t1;
|
|
768
|
+
}
|
|
769
|
+
++j;
|
|
770
|
+
}
|
|
771
|
+
merged.push_back(std::move(m));
|
|
772
|
+
}
|
|
773
|
+
|
|
774
|
+
const int nm = (int) merged.size();
|
|
775
|
+
for (int j = 0; j < nm; ++j) {
|
|
776
|
+
const auto & mt = merged[j];
|
|
743
777
|
start_obj(nullptr);
|
|
744
|
-
value_s("text",
|
|
745
|
-
if(
|
|
778
|
+
value_s("text", mt.text.c_str(), false);
|
|
779
|
+
if (mt.data.t0 > -1 && mt.t1 > -1) {
|
|
746
780
|
// If we have per-token timestamps, write them out
|
|
747
|
-
times_o(
|
|
781
|
+
times_o(mt.data.t0, mt.t1, false);
|
|
748
782
|
}
|
|
749
|
-
value_i("id",
|
|
750
|
-
value_f("p",
|
|
751
|
-
value_f("t_dtw",
|
|
752
|
-
end_obj(j == (
|
|
783
|
+
value_i("id", mt.data.id, false);
|
|
784
|
+
value_f("p", mt.data.p, false);
|
|
785
|
+
value_f("t_dtw", mt.data.t_dtw, true);
|
|
786
|
+
end_obj(j == (nm - 1));
|
|
753
787
|
}
|
|
754
788
|
end_arr(!params.diarize && !params.tinydiarize);
|
|
755
789
|
}
|
|
@@ -74,6 +74,7 @@ bool ggml_common_quantize_0(
|
|
|
74
74
|
case GGML_FTYPE_MOSTLY_BF16:
|
|
75
75
|
case GGML_FTYPE_MOSTLY_MXFP4:
|
|
76
76
|
case GGML_FTYPE_MOSTLY_NVFP4:
|
|
77
|
+
case GGML_FTYPE_MOSTLY_Q1_0:
|
|
77
78
|
{
|
|
78
79
|
fprintf(stderr, "%s: invalid model type %d\n", __func__, ftype);
|
|
79
80
|
return false;
|
|
@@ -215,6 +216,7 @@ bool ggml_common_quantize_0(
|
|
|
215
216
|
case GGML_TYPE_TQ2_0:
|
|
216
217
|
case GGML_TYPE_MXFP4:
|
|
217
218
|
case GGML_TYPE_NVFP4:
|
|
219
|
+
case GGML_TYPE_Q1_0:
|
|
218
220
|
case GGML_TYPE_COUNT:
|
|
219
221
|
{
|
|
220
222
|
fprintf(stderr, "%s: unsupported quantization type %d (%s)\n", __func__, ttype, ggml_type_name((ggml_type) ttype));
|
|
@@ -34,91 +34,35 @@
|
|
|
34
34
|
#include <cstring>
|
|
35
35
|
#include <fstream>
|
|
36
36
|
|
|
37
|
-
#ifdef
|
|
38
|
-
// as implemented in
|
|
39
|
-
extern bool ffmpeg_decode_audio(const std::string & ifname, std::vector<uint8_t> & wav_data);
|
|
37
|
+
#ifdef WHISPER_COMMON_FFMPEG
|
|
38
|
+
// as implemented in ffmpeg-trancode.cpp only embedded in common lib if whisper built with ffmpeg support
|
|
39
|
+
extern bool ffmpeg_decode_audio(const std::string & ifname, std::vector<uint8_t> & wav_data, int out_sample_rate = WHISPER_SAMPLE_RATE);
|
|
40
40
|
#endif
|
|
41
41
|
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
|
|
42
|
+
// extract f32 PCM frames from an initialized decoder, downmix to mono and keep the stereo split
|
|
43
|
+
static bool read_audio_from_decoder(ma_decoder & decoder, std::vector<float> & pcmf32, std::vector<std::vector<float>> & pcmf32s, bool stereo) {
|
|
45
44
|
ma_result result;
|
|
46
|
-
ma_decoder_config decoder_config;
|
|
47
|
-
ma_decoder decoder;
|
|
48
|
-
|
|
49
|
-
decoder_config = ma_decoder_config_init(ma_format_f32, stereo ? 2 : 1, WHISPER_SAMPLE_RATE);
|
|
50
|
-
|
|
51
|
-
if (fname == "-") {
|
|
52
|
-
#ifdef _WIN32
|
|
53
|
-
_setmode(_fileno(stdin), _O_BINARY);
|
|
54
|
-
#endif
|
|
55
|
-
|
|
56
|
-
uint8_t buf[1024];
|
|
57
|
-
while (true)
|
|
58
|
-
{
|
|
59
|
-
const size_t n = fread(buf, 1, sizeof(buf), stdin);
|
|
60
|
-
if (n == 0) {
|
|
61
|
-
break;
|
|
62
|
-
}
|
|
63
|
-
audio_data.insert(audio_data.end(), buf, buf + n);
|
|
64
|
-
}
|
|
65
|
-
|
|
66
|
-
if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
|
|
67
|
-
|
|
68
|
-
fprintf(stderr, "Error: failed to open audio data from stdin (%s)\n", ma_result_description(result));
|
|
69
|
-
|
|
70
|
-
return false;
|
|
71
|
-
}
|
|
72
|
-
|
|
73
|
-
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, audio_data.size());
|
|
74
|
-
}
|
|
75
|
-
else if (((result = ma_decoder_init_file(fname.c_str(), &decoder_config, &decoder)) != MA_SUCCESS)) {
|
|
76
|
-
#if defined(WHISPER_FFMPEG)
|
|
77
|
-
if (ffmpeg_decode_audio(fname, audio_data) != 0) {
|
|
78
|
-
fprintf(stderr, "error: failed to ffmpeg decode '%s'\n", fname.c_str());
|
|
79
|
-
|
|
80
|
-
return false;
|
|
81
|
-
}
|
|
82
|
-
|
|
83
|
-
if ((result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
|
|
84
|
-
fprintf(stderr, "error: failed to read audio data as wav (%s)\n", ma_result_description(result));
|
|
85
|
-
|
|
86
|
-
return false;
|
|
87
|
-
}
|
|
88
|
-
#else
|
|
89
|
-
if ((result = ma_decoder_init_memory(fname.c_str(), fname.size(), &decoder_config, &decoder)) != MA_SUCCESS) {
|
|
90
|
-
fprintf(stderr, "error: failed to read audio data as wav (%s)\n", ma_result_description(result));
|
|
91
|
-
|
|
92
|
-
return false;
|
|
93
|
-
}
|
|
94
|
-
#endif
|
|
95
|
-
}
|
|
96
|
-
|
|
97
45
|
ma_uint64 frame_count;
|
|
98
46
|
ma_uint64 frames_read;
|
|
99
47
|
|
|
100
48
|
if ((result = ma_decoder_get_length_in_pcm_frames(&decoder, &frame_count)) != MA_SUCCESS) {
|
|
101
|
-
|
|
102
|
-
|
|
103
|
-
return false;
|
|
49
|
+
fprintf(stderr, "error: failed to retrieve the length of the audio data (%s)\n", ma_result_description(result));
|
|
50
|
+
return false;
|
|
104
51
|
}
|
|
105
52
|
|
|
106
53
|
pcmf32.resize(stereo ? frame_count*2 : frame_count);
|
|
107
54
|
|
|
108
55
|
if ((result = ma_decoder_read_pcm_frames(&decoder, pcmf32.data(), frame_count, &frames_read)) != MA_SUCCESS) {
|
|
109
|
-
|
|
110
|
-
|
|
111
|
-
return false;
|
|
56
|
+
fprintf(stderr, "error: failed to read the frames of the audio data (%s)\n", ma_result_description(result));
|
|
57
|
+
return false;
|
|
112
58
|
}
|
|
113
59
|
|
|
114
60
|
if (stereo) {
|
|
115
61
|
std::vector<float> stereo_data = pcmf32;
|
|
116
62
|
pcmf32.resize(frame_count);
|
|
117
|
-
|
|
118
63
|
for (uint64_t i = 0; i < frame_count; i++) {
|
|
119
64
|
pcmf32[i] = (stereo_data[2*i] + stereo_data[2*i + 1]);
|
|
120
65
|
}
|
|
121
|
-
|
|
122
66
|
pcmf32s.resize(2);
|
|
123
67
|
pcmf32s[0].resize(frame_count);
|
|
124
68
|
pcmf32s[1].resize(frame_count);
|
|
@@ -128,11 +72,111 @@ bool read_audio_data(const std::string & fname, std::vector<float>& pcmf32, std:
|
|
|
128
72
|
}
|
|
129
73
|
}
|
|
130
74
|
|
|
131
|
-
ma_decoder_uninit(&decoder);
|
|
132
|
-
|
|
133
75
|
return true;
|
|
134
76
|
}
|
|
135
77
|
|
|
78
|
+
bool read_audio_data(const std::string & fname, std::vector<float> & pcmf32, std::vector<std::vector<float>> & pcmf32s, bool stereo) {
|
|
79
|
+
std::vector<uint8_t> audio_data; // used for pipe input from stdin or ffmpeg decoding output
|
|
80
|
+
|
|
81
|
+
ma_result result;
|
|
82
|
+
ma_decoder_config decoder_config;
|
|
83
|
+
|
|
84
|
+
struct decoder_guard {
|
|
85
|
+
ma_decoder decoder;
|
|
86
|
+
bool initialized = false;
|
|
87
|
+
ma_decoder * operator&() { return &decoder; }
|
|
88
|
+
~decoder_guard() {
|
|
89
|
+
if (initialized) {
|
|
90
|
+
ma_decoder_uninit(&decoder);
|
|
91
|
+
}
|
|
92
|
+
}
|
|
93
|
+
};
|
|
94
|
+
decoder_guard decoder{};
|
|
95
|
+
|
|
96
|
+
decoder_config = ma_decoder_config_init(ma_format_f32, stereo ? 2 : 1, WHISPER_SAMPLE_RATE);
|
|
97
|
+
|
|
98
|
+
if (fname == "-") {
|
|
99
|
+
#ifdef _WIN32
|
|
100
|
+
_setmode(_fileno(stdin), _O_BINARY);
|
|
101
|
+
#endif
|
|
102
|
+
|
|
103
|
+
uint8_t buf[1024];
|
|
104
|
+
while (true)
|
|
105
|
+
{
|
|
106
|
+
const size_t n = fread(buf, 1, sizeof(buf), stdin);
|
|
107
|
+
if (n == 0) {
|
|
108
|
+
break;
|
|
109
|
+
}
|
|
110
|
+
audio_data.insert(audio_data.end(), buf, buf + n);
|
|
111
|
+
}
|
|
112
|
+
|
|
113
|
+
result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder);
|
|
114
|
+
if (result != MA_SUCCESS) {
|
|
115
|
+
fprintf(stderr, "%s: failed to open audio data from stdin (%s)\n", __func__, ma_result_description(result));
|
|
116
|
+
return false;
|
|
117
|
+
}
|
|
118
|
+
decoder.initialized = true;
|
|
119
|
+
|
|
120
|
+
fprintf(stderr, "%s: read %zu bytes from stdin\n", __func__, audio_data.size());
|
|
121
|
+
} else {
|
|
122
|
+
fprintf(stderr, "%s: reading audio data from '%s' ...\n", __func__, fname.c_str());
|
|
123
|
+
|
|
124
|
+
// first try miniaudio. if it fails (or skipped) - try ffmpeg
|
|
125
|
+
{
|
|
126
|
+
const char * skip = getenv("WHISPER_COMMON_MINIAUDIO_SKIP");
|
|
127
|
+
if (!skip || strlen(skip) == 0 || strcmp(skip, "0") == 0) {
|
|
128
|
+
fprintf(stderr, "%s: trying to decode with miniaudio\n", __func__);
|
|
129
|
+
|
|
130
|
+
result = ma_decoder_init_file(fname.c_str(), &decoder_config, &decoder);
|
|
131
|
+
if (result == MA_SUCCESS) {
|
|
132
|
+
decoder.initialized = true;
|
|
133
|
+
}
|
|
134
|
+
} else {
|
|
135
|
+
fprintf(stderr, "%s: skipping miniaudio\n", __func__);
|
|
136
|
+
}
|
|
137
|
+
}
|
|
138
|
+
|
|
139
|
+
#if defined(WHISPER_COMMON_FFMPEG)
|
|
140
|
+
if (!decoder.initialized) {
|
|
141
|
+
fprintf(stderr, "%s: trying to decode with ffmpeg\n", __func__);
|
|
142
|
+
|
|
143
|
+
if (ffmpeg_decode_audio(fname, audio_data) != 0) {
|
|
144
|
+
fprintf(stderr, "%s: failed to ffmpeg decode\n", __func__);
|
|
145
|
+
return false;
|
|
146
|
+
}
|
|
147
|
+
result = ma_decoder_init_memory(audio_data.data(), audio_data.size(), &decoder_config, &decoder);
|
|
148
|
+
if (result != MA_SUCCESS) {
|
|
149
|
+
fprintf(stderr, "%s: failed to read audio data as wav (%s)\n", __func__, ma_result_description(result));
|
|
150
|
+
return false;
|
|
151
|
+
}
|
|
152
|
+
decoder.initialized = true;
|
|
153
|
+
}
|
|
154
|
+
#endif
|
|
155
|
+
|
|
156
|
+
if (!decoder.initialized) {
|
|
157
|
+
fprintf(stderr, "%s: failed to read audio data\n", __func__);
|
|
158
|
+
return false;
|
|
159
|
+
}
|
|
160
|
+
}
|
|
161
|
+
|
|
162
|
+
return read_audio_from_decoder(decoder.decoder, pcmf32, pcmf32s, stereo);
|
|
163
|
+
}
|
|
164
|
+
|
|
165
|
+
// decode audio bytes already held in memory
|
|
166
|
+
bool read_audio_data(const char * buffer, size_t buffer_size, std::vector<float> & pcmf32, std::vector<std::vector<float>> & pcmf32s, bool stereo) {
|
|
167
|
+
ma_decoder_config decoder_config = ma_decoder_config_init(ma_format_f32, stereo ? 2 : 1, WHISPER_SAMPLE_RATE);
|
|
168
|
+
ma_decoder decoder;
|
|
169
|
+
|
|
170
|
+
if (ma_decoder_init_memory(buffer, buffer_size, &decoder_config, &decoder) != MA_SUCCESS) {
|
|
171
|
+
fprintf(stderr, "error: failed to decode audio data from memory buffer\n");
|
|
172
|
+
return false;
|
|
173
|
+
}
|
|
174
|
+
|
|
175
|
+
bool ok = read_audio_from_decoder(decoder, pcmf32, pcmf32s, stereo);
|
|
176
|
+
ma_decoder_uninit(&decoder);
|
|
177
|
+
return ok;
|
|
178
|
+
}
|
|
179
|
+
|
|
136
180
|
// 500 -> 00:05.000
|
|
137
181
|
// 6000 -> 01:00.000
|
|
138
182
|
std::string to_timestamp(int64_t t, bool comma) {
|
|
@@ -154,6 +198,34 @@ int timestamp_to_sample(int64_t t, int n_samples, int whisper_sample_rate) {
|
|
|
154
198
|
return std::max(0, std::min((int) n_samples - 1, (int) ((t*whisper_sample_rate)/100)));
|
|
155
199
|
}
|
|
156
200
|
|
|
201
|
+
int utf8_trailing_bytes_needed(const std::string & s) {
|
|
202
|
+
const int n = (int) s.size();
|
|
203
|
+
int i = n - 1;
|
|
204
|
+
while (i >= 0 && ((unsigned char) s[i] & 0xC0) == 0x80) {
|
|
205
|
+
--i;
|
|
206
|
+
}
|
|
207
|
+
if (i < 0) {
|
|
208
|
+
return 0;
|
|
209
|
+
}
|
|
210
|
+
|
|
211
|
+
const unsigned char c = (unsigned char) s[i];
|
|
212
|
+
int expected;
|
|
213
|
+
if ((c & 0x80) == 0x00) {
|
|
214
|
+
expected = 1;
|
|
215
|
+
} else if ((c & 0xE0) == 0xC0) {
|
|
216
|
+
expected = 2;
|
|
217
|
+
} else if ((c & 0xF0) == 0xE0) {
|
|
218
|
+
expected = 3;
|
|
219
|
+
} else if ((c & 0xF8) == 0xF0) {
|
|
220
|
+
expected = 4;
|
|
221
|
+
} else {
|
|
222
|
+
return 0;
|
|
223
|
+
}
|
|
224
|
+
|
|
225
|
+
const int have = n - i;
|
|
226
|
+
return have >= expected ? 0 : (expected - have);
|
|
227
|
+
}
|
|
228
|
+
|
|
157
229
|
bool speak_with_file(const std::string & command, const std::string & text, const std::string & path, int voice_id) {
|
|
158
230
|
std::ofstream speak_file(path.c_str());
|
|
159
231
|
if (speak_file.fail()) {
|
|
@@ -14,11 +14,22 @@ bool read_audio_data(
|
|
|
14
14
|
std::vector<std::vector<float>> & pcmf32s,
|
|
15
15
|
bool stereo);
|
|
16
16
|
|
|
17
|
+
// decode audio bytes already held in memory (uploaded file, network buffer)
|
|
18
|
+
bool read_audio_data(
|
|
19
|
+
const char * buffer,
|
|
20
|
+
size_t buffer_size,
|
|
21
|
+
std::vector<float> & pcmf32,
|
|
22
|
+
std::vector<std::vector<float>> & pcmf32s,
|
|
23
|
+
bool stereo);
|
|
24
|
+
|
|
17
25
|
// convert timestamp to string, 6000 -> 01:00.000
|
|
18
26
|
std::string to_timestamp(int64_t t, bool comma = false);
|
|
19
27
|
|
|
20
28
|
// given a timestamp get the sample
|
|
21
29
|
int timestamp_to_sample(int64_t t, int n_samples, int whisper_sample_rate);
|
|
22
30
|
|
|
31
|
+
// Returns the number of trailing bytes still needed for s to end on a complete UTF-8 codepoint.
|
|
32
|
+
int utf8_trailing_bytes_needed(const std::string & s);
|
|
33
|
+
|
|
23
34
|
// write text to file, and call system("command voice_id file")
|
|
24
35
|
bool speak_with_file(const std::string & command, const std::string & text, const std::string & path, int voice_id);
|