whispercpp 1.3.1 → 1.3.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- checksums.yaml +4 -4
- data/.gitignore +4 -3
- data/README.md +92 -31
- data/Rakefile +26 -7
- data/ext/.gitignore +5 -7
- data/ext/dependencies.rb +61 -0
- data/ext/extconf.rb +21 -198
- data/ext/options.rb +221 -0
- data/ext/ruby_whisper.c +159 -0
- data/ext/ruby_whisper.h +17 -2
- data/ext/ruby_whisper_context.c +641 -0
- data/ext/ruby_whisper_error.c +52 -0
- data/ext/ruby_whisper_model.c +232 -0
- data/ext/ruby_whisper_params.c +1301 -0
- data/ext/ruby_whisper_segment.c +143 -0
- data/ext/ruby_whisper_transcribe.cpp +87 -0
- data/ext/ruby_whisper_vad_params.c +288 -0
- data/ext/sources/.dockerignore +3 -0
- data/ext/sources/.github/workflows/bindings-ruby.yml +21 -0
- data/ext/sources/CMakeGraphVizOptions.cmake +8 -0
- data/ext/sources/CMakeLists.txt +251 -0
- data/ext/sources/bindings/javascript/CMakeLists.txt +41 -0
- data/ext/sources/bindings/javascript/emscripten.cpp +93 -0
- data/ext/sources/bindings/javascript/libwhisper.worker.js +1 -0
- data/ext/sources/bindings/javascript/package-tmpl.json +26 -0
- data/ext/sources/bindings/javascript/package.json +26 -0
- data/ext/sources/bindings/javascript/whisper.js +19 -0
- data/ext/sources/build-xcframework.sh +547 -0
- data/ext/sources/ci/run.sh +336 -0
- data/ext/sources/close-issue.yml +28 -0
- data/ext/sources/cmake/DefaultTargetOptions.cmake +16 -0
- data/ext/sources/cmake/FindFFmpeg.cmake +163 -0
- data/ext/sources/cmake/build-info.cmake +60 -0
- data/ext/sources/cmake/git-vars.cmake +22 -0
- data/ext/sources/cmake/whisper-config.cmake.in +65 -0
- data/ext/sources/cmake/whisper.pc.in +10 -0
- data/ext/sources/examples/CMakeLists.txt +124 -0
- data/ext/sources/examples/addon.node/CMakeLists.txt +31 -0
- data/ext/sources/examples/addon.node/__test__/whisper.spec.js +37 -0
- data/ext/sources/examples/addon.node/addon.cpp +438 -0
- data/ext/sources/examples/addon.node/index.js +54 -0
- data/ext/sources/examples/addon.node/package.json +16 -0
- data/ext/sources/examples/bench/CMakeLists.txt +8 -0
- data/ext/sources/examples/bench/bench.cpp +175 -0
- data/ext/sources/examples/bench.wasm/CMakeLists.txt +49 -0
- data/ext/sources/examples/bench.wasm/emscripten.cpp +87 -0
- data/ext/sources/examples/bench.wasm/index-tmpl.html +284 -0
- data/ext/sources/examples/cli/CMakeLists.txt +8 -0
- data/ext/sources/examples/cli/cli.cpp +1294 -0
- data/ext/sources/examples/coi-serviceworker.js +146 -0
- data/ext/sources/examples/command/CMakeLists.txt +10 -0
- data/ext/sources/examples/command/command.cpp +776 -0
- data/ext/sources/examples/command/commands.txt +9 -0
- data/ext/sources/examples/command.wasm/CMakeLists.txt +50 -0
- data/ext/sources/examples/command.wasm/emscripten.cpp +327 -0
- data/ext/sources/examples/command.wasm/index-tmpl.html +414 -0
- data/ext/sources/examples/common-ggml.cpp +238 -0
- data/ext/sources/examples/common-ggml.h +18 -0
- data/ext/sources/examples/common-sdl.cpp +227 -0
- data/ext/sources/examples/common-sdl.h +49 -0
- data/ext/sources/examples/common-whisper.cpp +168 -0
- data/ext/sources/examples/common-whisper.h +24 -0
- data/ext/sources/examples/common.cpp +675 -0
- data/ext/sources/examples/common.h +322 -0
- data/ext/sources/examples/deprecation-warning/CMakeLists.txt +6 -0
- data/ext/sources/examples/deprecation-warning/deprecation-warning.cpp +38 -0
- data/ext/sources/examples/ffmpeg-transcode.cpp +368 -0
- data/ext/sources/examples/generate-karaoke.sh +57 -0
- data/ext/sources/examples/grammar-parser.cpp +423 -0
- data/ext/sources/examples/grammar-parser.h +29 -0
- data/ext/sources/examples/helpers.js +191 -0
- data/ext/sources/examples/json.hpp +24596 -0
- data/ext/sources/examples/livestream.sh +112 -0
- data/ext/sources/examples/lsp/CMakeLists.txt +9 -0
- data/ext/sources/examples/lsp/lsp.cpp +467 -0
- data/ext/sources/examples/lsp/whisper.vim +362 -0
- data/ext/sources/examples/miniaudio.h +93468 -0
- data/ext/sources/examples/python/test_whisper_processor.py +7 -0
- data/ext/sources/examples/python/whisper_processor.py +54 -0
- data/ext/sources/examples/quantize/CMakeLists.txt +6 -0
- data/ext/sources/examples/quantize/quantize.cpp +223 -0
- data/ext/sources/examples/server/CMakeLists.txt +12 -0
- data/ext/sources/examples/server/bench.js +29 -0
- data/ext/sources/examples/server/httplib.h +10497 -0
- data/ext/sources/examples/server/server.cpp +1091 -0
- data/ext/sources/examples/server.py +115 -0
- data/ext/sources/examples/stb_vorbis.c +5584 -0
- data/ext/sources/examples/stream/CMakeLists.txt +10 -0
- data/ext/sources/examples/stream/stream.cpp +429 -0
- data/ext/sources/examples/stream.wasm/CMakeLists.txt +49 -0
- data/ext/sources/examples/stream.wasm/emscripten.cpp +216 -0
- data/ext/sources/examples/stream.wasm/index-tmpl.html +414 -0
- data/ext/sources/examples/sycl/CMakeLists.txt +9 -0
- data/ext/sources/examples/sycl/build.sh +22 -0
- data/ext/sources/examples/sycl/ls-sycl-device.cpp +11 -0
- data/ext/sources/examples/sycl/run-whisper.sh +17 -0
- data/ext/sources/examples/talk-llama/CMakeLists.txt +40 -0
- data/ext/sources/examples/talk-llama/eleven-labs.py +80 -0
- data/ext/sources/examples/talk-llama/llama-adapter.cpp +388 -0
- data/ext/sources/examples/talk-llama/llama-adapter.h +76 -0
- data/ext/sources/examples/talk-llama/llama-arch.cpp +1746 -0
- data/ext/sources/examples/talk-llama/llama-arch.h +437 -0
- data/ext/sources/examples/talk-llama/llama-batch.cpp +374 -0
- data/ext/sources/examples/talk-llama/llama-batch.h +89 -0
- data/ext/sources/examples/talk-llama/llama-chat.cpp +663 -0
- data/ext/sources/examples/talk-llama/llama-chat.h +58 -0
- data/ext/sources/examples/talk-llama/llama-context.cpp +2676 -0
- data/ext/sources/examples/talk-llama/llama-context.h +276 -0
- data/ext/sources/examples/talk-llama/llama-cparams.cpp +5 -0
- data/ext/sources/examples/talk-llama/llama-cparams.h +41 -0
- data/ext/sources/examples/talk-llama/llama-grammar.cpp +1229 -0
- data/ext/sources/examples/talk-llama/llama-grammar.h +173 -0
- data/ext/sources/examples/talk-llama/llama-graph.cpp +1618 -0
- data/ext/sources/examples/talk-llama/llama-graph.h +640 -0
- data/ext/sources/examples/talk-llama/llama-hparams.cpp +95 -0
- data/ext/sources/examples/talk-llama/llama-hparams.h +190 -0
- data/ext/sources/examples/talk-llama/llama-impl.cpp +167 -0
- data/ext/sources/examples/talk-llama/llama-impl.h +61 -0
- data/ext/sources/examples/talk-llama/llama-io.cpp +15 -0
- data/ext/sources/examples/talk-llama/llama-io.h +35 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache.cpp +2739 -0
- data/ext/sources/examples/talk-llama/llama-kv-cache.h +502 -0
- data/ext/sources/examples/talk-llama/llama-kv-cells.h +379 -0
- data/ext/sources/examples/talk-llama/llama-memory.cpp +1 -0
- data/ext/sources/examples/talk-llama/llama-memory.h +32 -0
- data/ext/sources/examples/talk-llama/llama-mmap.cpp +600 -0
- data/ext/sources/examples/talk-llama/llama-mmap.h +68 -0
- data/ext/sources/examples/talk-llama/llama-model-loader.cpp +1138 -0
- data/ext/sources/examples/talk-llama/llama-model-loader.h +169 -0
- data/ext/sources/examples/talk-llama/llama-model-saver.cpp +281 -0
- data/ext/sources/examples/talk-llama/llama-model-saver.h +37 -0
- data/ext/sources/examples/talk-llama/llama-model.cpp +13814 -0
- data/ext/sources/examples/talk-llama/llama-model.h +425 -0
- data/ext/sources/examples/talk-llama/llama-quant.cpp +966 -0
- data/ext/sources/examples/talk-llama/llama-quant.h +1 -0
- data/ext/sources/examples/talk-llama/llama-sampling.cpp +2575 -0
- data/ext/sources/examples/talk-llama/llama-sampling.h +32 -0
- data/ext/sources/examples/talk-llama/llama-vocab.cpp +3340 -0
- data/ext/sources/examples/talk-llama/llama-vocab.h +131 -0
- data/ext/sources/examples/talk-llama/llama.cpp +354 -0
- data/ext/sources/examples/talk-llama/llama.h +1377 -0
- data/ext/sources/examples/talk-llama/prompts/talk-alpaca.txt +23 -0
- data/ext/sources/examples/talk-llama/speak +40 -0
- data/ext/sources/examples/talk-llama/speak.bat +1 -0
- data/ext/sources/examples/talk-llama/speak.ps1 +14 -0
- data/ext/sources/examples/talk-llama/talk-llama.cpp +808 -0
- data/ext/sources/examples/talk-llama/unicode-data.cpp +7034 -0
- data/ext/sources/examples/talk-llama/unicode-data.h +20 -0
- data/ext/sources/examples/talk-llama/unicode.cpp +849 -0
- data/ext/sources/examples/talk-llama/unicode.h +66 -0
- data/ext/sources/examples/vad-speech-segments/CMakeLists.txt +8 -0
- data/ext/sources/examples/vad-speech-segments/speech.cpp +143 -0
- data/ext/sources/examples/wchess/CMakeLists.txt +10 -0
- data/ext/sources/examples/wchess/libwchess/CMakeLists.txt +19 -0
- data/ext/sources/examples/wchess/libwchess/Chessboard.cpp +803 -0
- data/ext/sources/examples/wchess/libwchess/Chessboard.h +33 -0
- data/ext/sources/examples/wchess/libwchess/WChess.cpp +193 -0
- data/ext/sources/examples/wchess/libwchess/WChess.h +63 -0
- data/ext/sources/examples/wchess/libwchess/test-chessboard.cpp +117 -0
- data/ext/sources/examples/wchess/wchess.cmd/CMakeLists.txt +8 -0
- data/ext/sources/examples/wchess/wchess.cmd/wchess.cmd.cpp +249 -0
- data/ext/sources/examples/whisper.wasm/CMakeLists.txt +50 -0
- data/ext/sources/examples/whisper.wasm/emscripten.cpp +118 -0
- data/ext/sources/examples/whisper.wasm/index-tmpl.html +658 -0
- data/ext/sources/ggml/CMakeLists.txt +390 -0
- data/ext/sources/ggml/cmake/BuildTypes.cmake +54 -0
- data/ext/sources/ggml/cmake/GitVars.cmake +22 -0
- data/ext/sources/ggml/cmake/common.cmake +26 -0
- data/ext/sources/ggml/cmake/ggml-config.cmake.in +152 -0
- data/ext/{ggml → sources/ggml}/include/ggml-alloc.h +1 -1
- data/ext/{ggml → sources/ggml}/include/ggml-backend.h +9 -7
- data/ext/{ggml → sources/ggml}/include/ggml-cpp.h +2 -1
- data/ext/{ggml → sources/ggml}/include/ggml-cpu.h +9 -1
- data/ext/{ggml → sources/ggml}/include/ggml-metal.h +1 -1
- data/ext/{ggml → sources/ggml}/include/ggml-opt.h +49 -28
- data/ext/{ggml → sources/ggml}/include/ggml-rpc.h +6 -1
- data/ext/{ggml → sources/ggml}/include/ggml-vulkan.h +0 -2
- data/ext/{ggml → sources/ggml}/include/ggml.h +182 -265
- data/ext/sources/ggml/include/gguf.h +202 -0
- data/ext/sources/ggml/src/CMakeLists.txt +346 -0
- data/ext/{ggml → sources/ggml}/src/ggml-alloc.c +34 -29
- data/ext/sources/ggml/src/ggml-amx/CMakeLists.txt +107 -0
- data/ext/{ggml → sources/ggml}/src/ggml-backend-impl.h +1 -2
- data/ext/{ggml → sources/ggml}/src/ggml-backend-reg.cpp +87 -53
- data/ext/{ggml → sources/ggml}/src/ggml-backend.cpp +26 -14
- data/ext/sources/ggml/src/ggml-blas/CMakeLists.txt +87 -0
- data/ext/sources/ggml/src/ggml-cann/CMakeLists.txt +74 -0
- data/ext/sources/ggml/src/ggml-cann/Doxyfile +2579 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cann/acl_tensor.cpp +10 -4
- data/ext/{ggml → sources/ggml}/src/ggml-cann/acl_tensor.h +5 -5
- data/ext/{ggml → sources/ggml}/src/ggml-cann/aclnn_ops.cpp +1272 -1506
- data/ext/sources/ggml/src/ggml-cann/aclnn_ops.h +1125 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cann/common.h +135 -1
- data/ext/{ggml → sources/ggml}/src/ggml-cann/ggml-cann.cpp +564 -146
- data/ext/sources/ggml/src/ggml-cann/kernels/CMakeLists.txt +30 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/dup.cpp +3 -5
- data/ext/{ggml → sources/ggml}/src/ggml-common.h +12 -8
- data/ext/sources/ggml/src/ggml-cpu/CMakeLists.txt +504 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cpu/amx/amx.cpp +2 -1
- data/ext/sources/ggml/src/ggml-cpu/binary-ops.cpp +158 -0
- data/ext/sources/ggml/src/ggml-cpu/binary-ops.h +16 -0
- data/ext/sources/ggml/src/ggml-cpu/cmake/FindSIMD.cmake +100 -0
- data/ext/sources/ggml/src/ggml-cpu/common.h +72 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cpu/cpu-feats-x86.cpp +5 -1
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +6431 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-impl.h +163 -41
- data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-quants.c +4029 -1117
- data/ext/sources/ggml/src/ggml-cpu/ggml-cpu.c +3510 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu.cpp +67 -18
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.cpp +337 -0
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kernels.h +95 -0
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +482 -0
- data/ext/sources/ggml/src/ggml-cpu/kleidiai/kleidiai.h +17 -0
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.cpp +3544 -0
- data/ext/sources/ggml/src/ggml-cpu/llamafile/sgemm.h +14 -0
- data/ext/sources/ggml/src/ggml-cpu/ops.cpp +8903 -0
- data/ext/sources/ggml/src/ggml-cpu/ops.h +110 -0
- data/ext/sources/ggml/src/ggml-cpu/simd-mappings.h +892 -0
- data/ext/sources/ggml/src/ggml-cpu/unary-ops.cpp +186 -0
- data/ext/sources/ggml/src/ggml-cpu/unary-ops.h +28 -0
- data/ext/sources/ggml/src/ggml-cpu/vec.cpp +252 -0
- data/ext/sources/ggml/src/ggml-cpu/vec.h +818 -0
- data/ext/sources/ggml/src/ggml-cuda/CMakeLists.txt +184 -0
- data/ext/sources/ggml/src/ggml-cuda/acc.cu +61 -0
- data/ext/sources/ggml/src/ggml-cuda/acc.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/arange.cu +34 -0
- data/ext/sources/ggml/src/ggml-cuda/arange.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/argmax.cu +91 -0
- data/ext/sources/ggml/src/ggml-cuda/argmax.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/argsort.cu +104 -0
- data/ext/sources/ggml/src/ggml-cuda/argsort.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cu +363 -0
- data/ext/sources/ggml/src/ggml-cuda/binbcast.cuh +9 -0
- data/ext/sources/ggml/src/ggml-cuda/clamp.cu +45 -0
- data/ext/sources/ggml/src/ggml-cuda/clamp.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/common.cuh +828 -0
- data/ext/sources/ggml/src/ggml-cuda/concat.cu +221 -0
- data/ext/sources/ggml/src/ggml-cuda/concat.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/conv-transpose-1d.cu +89 -0
- data/ext/sources/ggml/src/ggml-cuda/conv-transpose-1d.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cu +730 -0
- data/ext/sources/ggml/src/ggml-cuda/convert.cuh +26 -0
- data/ext/sources/ggml/src/ggml-cuda/count-equal.cu +64 -0
- data/ext/sources/ggml/src/ggml-cuda/count-equal.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/cp-async.cuh +57 -0
- data/ext/sources/ggml/src/ggml-cuda/cpy.cu +705 -0
- data/ext/sources/ggml/src/ggml-cuda/cpy.cuh +11 -0
- data/ext/sources/ggml/src/ggml-cuda/cross-entropy-loss.cu +189 -0
- data/ext/sources/ggml/src/ggml-cuda/cross-entropy-loss.cuh +7 -0
- data/ext/sources/ggml/src/ggml-cuda/dequantize.cuh +103 -0
- data/ext/sources/ggml/src/ggml-cuda/diagmask.cu +40 -0
- data/ext/sources/ggml/src/ggml-cuda/diagmask.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-common.cuh +881 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-mma-f16.cuh +1471 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cu +357 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f16.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cu +365 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-tile-f32.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f16.cuh +482 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-vec-f32.cuh +472 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cu +634 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn.cu +346 -0
- data/ext/sources/ggml/src/ggml-cuda/fattn.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/getrows.cu +275 -0
- data/ext/sources/ggml/src/ggml-cuda/getrows.cuh +15 -0
- data/ext/sources/ggml/src/ggml-cuda/ggml-cuda.cu +3505 -0
- data/ext/sources/ggml/src/ggml-cuda/gla.cu +93 -0
- data/ext/sources/ggml/src/ggml-cuda/gla.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/im2col.cu +103 -0
- data/ext/sources/ggml/src/ggml-cuda/im2col.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/mma.cuh +396 -0
- data/ext/sources/ggml/src/ggml-cuda/mmq.cu +324 -0
- data/ext/sources/ggml/src/ggml-cuda/mmq.cuh +3217 -0
- data/ext/sources/ggml/src/ggml-cuda/mmv.cu +336 -0
- data/ext/sources/ggml/src/ggml-cuda/mmv.cuh +12 -0
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cu +595 -0
- data/ext/sources/ggml/src/ggml-cuda/mmvq.cuh +12 -0
- data/ext/sources/ggml/src/ggml-cuda/norm.cu +458 -0
- data/ext/sources/ggml/src/ggml-cuda/norm.cuh +11 -0
- data/ext/sources/ggml/src/ggml-cuda/opt-step-adamw.cu +78 -0
- data/ext/sources/ggml/src/ggml-cuda/opt-step-adamw.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/out-prod.cu +68 -0
- data/ext/sources/ggml/src/ggml-cuda/out-prod.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/pad.cu +49 -0
- data/ext/sources/ggml/src/ggml-cuda/pad.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/pool2d.cu +94 -0
- data/ext/sources/ggml/src/ggml-cuda/pool2d.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/quantize.cu +190 -0
- data/ext/sources/ggml/src/ggml-cuda/quantize.cuh +27 -0
- data/ext/sources/ggml/src/ggml-cuda/rope.cu +456 -0
- data/ext/sources/ggml/src/ggml-cuda/rope.cuh +7 -0
- data/ext/sources/ggml/src/ggml-cuda/scale.cu +31 -0
- data/ext/sources/ggml/src/ggml-cuda/scale.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/softmax.cu +283 -0
- data/ext/sources/ggml/src/ggml-cuda/softmax.cuh +7 -0
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cu +148 -0
- data/ext/sources/ggml/src/ggml-cuda/ssm-conv.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cu +153 -0
- data/ext/sources/ggml/src/ggml-cuda/ssm-scan.cuh +3 -0
- data/ext/sources/ggml/src/ggml-cuda/sum.cu +45 -0
- data/ext/sources/ggml/src/ggml-cuda/sum.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cu +39 -0
- data/ext/sources/ggml/src/ggml-cuda/sumrows.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_1-ncols2_8.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_1.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_2.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_16-ncols2_4.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_4.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_2-ncols2_8.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_32-ncols2_1.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_32-ncols2_2.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_2.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_4.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_4-ncols2_8.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_64-ncols2_1.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_1.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_2.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_4.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-mma-f16-instance-ncols1_8-ncols2_8.cu +10 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +78 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq1_s.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_s.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xs.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq2_xxs.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_s.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq3_xxs.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_nl.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-iq4_xs.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q2_k.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q3_k.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q4_k.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_1.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q5_k.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q6_k.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/template-instances/mmq-instance-q8_0.cu +5 -0
- data/ext/sources/ggml/src/ggml-cuda/tsembd.cu +47 -0
- data/ext/sources/ggml/src/ggml-cuda/tsembd.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/unary.cu +289 -0
- data/ext/sources/ggml/src/ggml-cuda/unary.cuh +59 -0
- data/ext/sources/ggml/src/ggml-cuda/upscale.cu +51 -0
- data/ext/sources/ggml/src/ggml-cuda/upscale.cuh +5 -0
- data/ext/sources/ggml/src/ggml-cuda/vecdotq.cuh +1135 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cuda/vendors/cuda.h +1 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cuda/vendors/hip.h +57 -0
- data/ext/{ggml → sources/ggml}/src/ggml-cuda/vendors/musa.h +7 -1
- data/ext/sources/ggml/src/ggml-cuda/wkv.cu +199 -0
- data/ext/sources/ggml/src/ggml-cuda/wkv.cuh +7 -0
- data/ext/sources/ggml/src/ggml-hip/CMakeLists.txt +131 -0
- data/ext/{ggml → sources/ggml}/src/ggml-impl.h +64 -19
- data/ext/sources/ggml/src/ggml-kompute/CMakeLists.txt +166 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/common.comp +112 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_add.comp +58 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_addrow.comp +25 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f16.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f32.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f16.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f32.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_diagmask.comp +30 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_gelu.comp +22 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows.comp +17 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f16.comp +31 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f32.comp +31 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_0.comp +38 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_1.comp +39 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q6_k.comp +44 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_f16.comp +69 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_mat_f32.comp +51 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_0.comp +33 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_1.comp +35 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_k.comp +140 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q6_k.comp +106 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q8_0.comp +73 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n_pre.comp +28 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_norm.comp +84 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_relu.comp +21 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rmsnorm.comp +53 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f16.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f32.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f16.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f32.comp +52 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale.comp +19 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_scale_8.comp +23 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_silu.comp +22 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/op_softmax.comp +72 -0
- data/ext/sources/ggml/src/ggml-kompute/kompute-shaders/rope_common.comp +71 -0
- data/ext/sources/ggml/src/ggml-metal/CMakeLists.txt +120 -0
- data/ext/sources/ggml/src/ggml-metal/ggml-metal-impl.h +622 -0
- data/ext/{ggml → sources/ggml}/src/ggml-metal/ggml-metal.m +2178 -1064
- data/ext/{ggml → sources/ggml}/src/ggml-metal/ggml-metal.metal +1575 -1218
- data/ext/sources/ggml/src/ggml-musa/CMakeLists.txt +113 -0
- data/ext/sources/ggml/src/ggml-musa/mudnn.cu +112 -0
- data/ext/sources/ggml/src/ggml-musa/mudnn.cuh +12 -0
- data/ext/sources/ggml/src/ggml-opencl/CMakeLists.txt +96 -0
- data/ext/sources/ggml/src/ggml-opencl/ggml-opencl.cpp +5124 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/add.cl +83 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/clamp.cl +20 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cpy.cl +184 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/cvt.cl +118 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/diag_mask_inf.cl +58 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/embed_kernel.py +26 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gelu.cl +62 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle.cl +268 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/gemv_noshuffle_general.cl +274 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/get_rows.cl +163 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f16.cl +57 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/im2col_f32.cl +57 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul.cl +79 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mat_Ab_Bi_8x4.cl +139 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_f16_f16.cl +118 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32.cl +118 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_1row.cl +94 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_f16_f32_l4.cl +84 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_f32_f32.cl +118 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32.cl +192 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_1d_16x_flat.cl +307 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_1d_8x_flat.cl +265 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_8x_flat.cl +272 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q4_0_f32_v.cl +254 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/mul_mv_q6_k.cl +190 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/norm.cl +81 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/relu.cl +16 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/rms_norm.cl +96 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/rope.cl +721 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/scale.cl +16 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/silu.cl +30 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +87 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +87 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f16.cl +86 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/softmax_f32.cl +86 -0
- data/ext/sources/ggml/src/ggml-opencl/kernels/transpose.cl +84 -0
- data/ext/{ggml → sources/ggml}/src/ggml-opt.cpp +373 -190
- data/ext/{ggml → sources/ggml}/src/ggml-quants.c +114 -120
- data/ext/sources/ggml/src/ggml-rpc/CMakeLists.txt +9 -0
- data/ext/{ggml → sources/ggml}/src/ggml-rpc/ggml-rpc.cpp +480 -73
- data/ext/sources/ggml/src/ggml-sycl/CMakeLists.txt +189 -0
- data/ext/sources/ggml/src/ggml-sycl/backend.hpp +37 -0
- data/ext/sources/ggml/src/ggml-sycl/binbcast.cpp +345 -0
- data/ext/sources/ggml/src/ggml-sycl/binbcast.hpp +39 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/common.cpp +20 -32
- data/ext/sources/ggml/src/ggml-sycl/common.hpp +589 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/concat.cpp +32 -33
- data/ext/sources/ggml/src/ggml-sycl/concat.hpp +20 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/conv.cpp +4 -2
- data/ext/sources/ggml/src/ggml-sycl/conv.hpp +20 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/convert.cpp +104 -28
- data/ext/sources/ggml/src/ggml-sycl/convert.hpp +34 -0
- data/ext/sources/ggml/src/ggml-sycl/cpy.cpp +700 -0
- data/ext/sources/ggml/src/ggml-sycl/cpy.hpp +11 -0
- data/ext/sources/ggml/src/ggml-sycl/dequantize.hpp +791 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/dmmv.cpp +156 -17
- data/ext/sources/ggml/src/ggml-sycl/dmmv.hpp +27 -0
- data/ext/sources/ggml/src/ggml-sycl/dpct/helper.hpp +2957 -0
- data/ext/sources/ggml/src/ggml-sycl/element_wise.cpp +1511 -0
- data/ext/sources/ggml/src/ggml-sycl/element_wise.hpp +75 -0
- data/ext/sources/ggml/src/ggml-sycl/gemm.hpp +99 -0
- data/ext/sources/ggml/src/ggml-sycl/getrows.cpp +309 -0
- data/ext/sources/ggml/src/ggml-sycl/getrows.hpp +20 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/ggml-sycl.cpp +1004 -1240
- data/ext/sources/ggml/src/ggml-sycl/gla.cpp +106 -0
- data/ext/sources/ggml/src/ggml-sycl/gla.hpp +8 -0
- data/ext/sources/ggml/src/ggml-sycl/im2col.cpp +136 -0
- data/ext/sources/ggml/src/ggml-sycl/im2col.hpp +21 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/mmq.cpp +0 -1
- data/ext/sources/ggml/src/ggml-sycl/mmq.hpp +33 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/mmvq.cpp +261 -166
- data/ext/sources/ggml/src/ggml-sycl/mmvq.hpp +27 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/norm.cpp +204 -81
- data/ext/sources/ggml/src/ggml-sycl/norm.hpp +26 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/outprod.cpp +8 -17
- data/ext/sources/ggml/src/ggml-sycl/outprod.hpp +10 -0
- data/ext/sources/ggml/src/ggml-sycl/presets.hpp +74 -0
- data/ext/sources/ggml/src/ggml-sycl/quants.hpp +83 -0
- data/ext/sources/ggml/src/ggml-sycl/rope.cpp +361 -0
- data/ext/sources/ggml/src/ggml-sycl/rope.hpp +20 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/softmax.cpp +35 -25
- data/ext/sources/ggml/src/ggml-sycl/softmax.hpp +20 -0
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.cpp +13 -0
- data/ext/sources/ggml/src/ggml-sycl/sycl_hw.hpp +23 -0
- data/ext/{ggml → sources/ggml}/src/ggml-sycl/tsembd.cpp +3 -3
- data/ext/sources/ggml/src/ggml-sycl/tsembd.hpp +20 -0
- data/ext/sources/ggml/src/ggml-sycl/vecdotq.hpp +1215 -0
- data/ext/sources/ggml/src/ggml-sycl/wkv.cpp +293 -0
- data/ext/sources/ggml/src/ggml-sycl/wkv.hpp +10 -0
- data/ext/sources/ggml/src/ggml-vulkan/CMakeLists.txt +196 -0
- data/ext/sources/ggml/src/ggml-vulkan/cmake/host-toolchain.cmake.in +15 -0
- data/ext/{ggml → sources/ggml}/src/ggml-vulkan/ggml-vulkan.cpp +3130 -1087
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +39 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +29 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +29 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +51 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +69 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp +17 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/concat.comp +41 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +49 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_dw.comp +105 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +23 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +51 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +242 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/cos.comp +17 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/count_equal.comp +31 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_f32.comp +20 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs.comp +462 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_funcs_cm2.comp +699 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_head.comp +13 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +42 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_s.comp +35 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +44 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp +43 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +48 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +39 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +49 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_nl.comp +32 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_xs.comp +34 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +34 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +42 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_0.comp +30 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp +32 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +68 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_0.comp +34 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_1.comp +35 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +70 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +33 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q8_0.comp +31 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/diag_mask_inf.comp +34 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/div.comp +27 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +337 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +162 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +360 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +267 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +59 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp +25 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/gelu_quick.comp +23 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_binary_head.comp +64 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_head.comp +9 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/generic_unary_head.comp +76 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +33 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +41 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/group_norm.comp +66 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +100 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +41 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/leaky_relu.comp +22 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul.comp +27 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_split_k_reduce.comp +48 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +169 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +118 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_m.comp +82 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_s.comp +79 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp +90 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp +87 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp +87 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp +90 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp +88 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +118 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp +154 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +130 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q3_k.comp +132 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +136 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +167 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q6_k.comp +130 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +868 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +441 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +442 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +99 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/norm.comp +44 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_adamw.comp +42 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +28 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/pool2d.comp +74 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +77 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/relu.comp +21 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp +26 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/repeat_back.comp +37 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +52 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_back.comp +55 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +58 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +60 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +43 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +43 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +47 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +24 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sigmoid.comp +20 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/silu.comp +22 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/silu_back.comp +26 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sin.comp +17 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +173 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +50 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/square.comp +17 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sub.comp +29 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +37 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/tanh.comp +20 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/test_bfloat16_support.comp +7 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/test_coopmat2_support.comp +7 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/test_coopmat_support.comp +7 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/test_integer_dot_support.comp +7 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +41 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/types.comp +1373 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +36 -0
- data/ext/{ggml → sources/ggml}/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +193 -35
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/wkv6.comp +87 -0
- data/ext/sources/ggml/src/ggml-vulkan/vulkan-shaders/wkv7.comp +91 -0
- data/ext/{ggml → sources/ggml}/src/ggml.c +676 -1820
- data/ext/sources/ggml/src/gguf.cpp +1330 -0
- data/ext/{include → sources/include}/whisper.h +68 -2
- data/ext/sources/src/CMakeLists.txt +143 -0
- data/ext/{src → sources/src}/coreml/whisper-decoder-impl.h +27 -15
- data/ext/{src → sources/src}/coreml/whisper-decoder-impl.m +35 -10
- data/ext/{src → sources/src}/coreml/whisper-encoder-impl.h +21 -9
- data/ext/{src → sources/src}/coreml/whisper-encoder-impl.m +28 -3
- data/ext/sources/src/coreml/whisper-encoder.mm +73 -0
- data/ext/sources/src/whisper-arch.h +197 -0
- data/ext/{src → sources/src}/whisper.cpp +1905 -374
- data/ext/sources/tests/CMakeLists.txt +105 -0
- data/ext/sources/tests/earnings21/eval.mk +58 -0
- data/ext/sources/tests/earnings21/eval.py +68 -0
- data/ext/sources/tests/earnings21/normalizers/__init__.py +2 -0
- data/ext/sources/tests/earnings21/normalizers/basic.py +80 -0
- data/ext/sources/tests/earnings21/normalizers/english.json +1741 -0
- data/ext/sources/tests/earnings21/normalizers/english.py +550 -0
- data/ext/sources/tests/earnings21/requirements.txt +6 -0
- data/ext/sources/tests/en-0-ref.txt +1 -0
- data/ext/sources/tests/en-1-ref.txt +1 -0
- data/ext/sources/tests/en-2-ref.txt +1 -0
- data/ext/sources/tests/es-0-ref.txt +1 -0
- data/ext/sources/tests/librispeech/eval.mk +39 -0
- data/ext/sources/tests/librispeech/eval.py +47 -0
- data/ext/sources/tests/librispeech/normalizers/__init__.py +2 -0
- data/ext/sources/tests/librispeech/normalizers/basic.py +80 -0
- data/ext/sources/tests/librispeech/normalizers/english.json +1741 -0
- data/ext/sources/tests/librispeech/normalizers/english.py +550 -0
- data/ext/sources/tests/librispeech/requirements.txt +6 -0
- data/ext/sources/tests/run-tests.sh +130 -0
- data/ext/sources/tests/test-c.c +3 -0
- data/ext/sources/tests/test-vad-full.cpp +54 -0
- data/ext/sources/tests/test-vad.cpp +83 -0
- data/ext/sources/tests/test-whisper.js +58 -0
- data/extsources.rb +33 -5
- data/lib/whisper/model/uri.rb +149 -128
- data/sig/whisper.rbs +480 -0
- data/tests/helper.rb +28 -0
- data/tests/test_callback.rb +45 -3
- data/tests/test_error.rb +2 -2
- data/tests/test_model.rb +38 -0
- data/tests/test_package.rb +18 -3
- data/tests/test_params.rb +145 -8
- data/tests/test_segment.rb +10 -19
- data/tests/test_vad.rb +19 -0
- data/tests/test_vad_params.rb +103 -0
- data/tests/test_whisper.rb +37 -37
- data/whispercpp.gemspec +5 -4
- metadata +766 -111
- data/ext/cpu.mk +0 -9
- data/ext/examples/dr_wav.h +0 -8815
- data/ext/ggml/src/ggml-cann/aclnn_ops.h +0 -592
- data/ext/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +0 -4262
- data/ext/ggml/src/ggml-cpu/ggml-cpu.c +0 -14123
- data/ext/ggml/src/ggml-cpu/llamafile/sgemm.cpp +0 -1884
- data/ext/ggml/src/ggml-cpu/llamafile/sgemm.h +0 -14
- data/ext/ggml/src/ggml-metal/ggml-metal-impl.h +0 -288
- data/ext/ggml/src/ggml-sycl/element_wise.cpp +0 -1030
- data/ext/ggml/src/ggml-sycl/im2col.cpp +0 -126
- data/ext/ggml/src/ggml-sycl/rope.cpp +0 -276
- data/ext/ggml/src/ggml-sycl/wkv6.cpp +0 -141
- data/ext/metal-embed.mk +0 -17
- data/ext/metal.mk +0 -6
- data/ext/ruby_whisper.cpp +0 -1909
- data/ext/scripts/get-flags.mk +0 -38
- data/lib/whisper.rb +0 -2
- /data/ext/{ggml → sources/ggml}/include/ggml-blas.h +0 -0
- /data/ext/{ggml → sources/ggml}/include/ggml-cann.h +0 -0
- /data/ext/{ggml → sources/ggml}/include/ggml-cuda.h +0 -0
- /data/ext/{ggml → sources/ggml}/include/ggml-kompute.h +0 -0
- /data/ext/{ggml → sources/ggml}/include/ggml-opencl.h +0 -0
- /data/ext/{ggml → sources/ggml}/include/ggml-sycl.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-amx/common.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-amx/ggml-amx.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-amx/mmq.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-amx/mmq.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-blas/ggml-blas.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/ascendc_kernels.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/get_row_f16.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/get_row_f32.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/amx/amx.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/amx/common.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/amx/mmq.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/amx/mmq.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-aarch64.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-hbm.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-hbm.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-quants.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-traits.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-cpu/ggml-cpu-traits.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-kompute/ggml-kompute.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-quants.h +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-threading.cpp +0 -0
- /data/ext/{ggml → sources/ggml}/src/ggml-threading.h +0 -0
- /data/ext/{src → sources/src}/coreml/whisper-encoder.h +0 -0
- /data/ext/{src → sources/src}/openvino/whisper-openvino-encoder.cpp +0 -0
- /data/ext/{src → sources/src}/openvino/whisper-openvino-encoder.h +0 -0
@@ -0,0 +1,33 @@
|
|
1
|
+
#pragma once
|
2
|
+
#include <string>
|
3
|
+
#include <set>
|
4
|
+
#include <memory>
|
5
|
+
|
6
|
+
// just basic validation
|
7
|
+
// fixme: missing en passant, castling, promotion, etc.
|
8
|
+
struct State;
|
9
|
+
class Piece;
|
10
|
+
class Chessboard {
|
11
|
+
public:
|
12
|
+
Chessboard();
|
13
|
+
~Chessboard();
|
14
|
+
std::string process(const std::string& command);
|
15
|
+
std::string stringifyBoard();
|
16
|
+
const std::string& grammar() { return m_grammar; }
|
17
|
+
const std::string& prompt() { return m_prompt; }
|
18
|
+
void setPrompt(const std::string& prompt);
|
19
|
+
private:
|
20
|
+
bool parseCommand(const std::string& command, Piece*& piece, char& pos_to);
|
21
|
+
bool move(Piece& piece, char pos);
|
22
|
+
void flagUpdates(char pos_from, char pos_to);
|
23
|
+
void updatePins(Piece& piece);
|
24
|
+
void detectChecks();
|
25
|
+
void setGrammar();
|
26
|
+
|
27
|
+
std::unique_ptr<State> m_state;
|
28
|
+
std::set<char> m_allowedInCheck;
|
29
|
+
bool m_inCheck = false;
|
30
|
+
int m_moveCounter = 0;
|
31
|
+
std::string m_grammar;
|
32
|
+
std::string m_prompt;
|
33
|
+
};
|
@@ -0,0 +1,193 @@
|
|
1
|
+
#include "WChess.h"
|
2
|
+
#include "Chessboard.h"
|
3
|
+
#include "grammar-parser.h"
|
4
|
+
#include "common.h"
|
5
|
+
#include <chrono>
|
6
|
+
|
7
|
+
WChess::WChess(whisper_context * ctx,
|
8
|
+
const whisper_full_params & wparams,
|
9
|
+
callbacks cb,
|
10
|
+
settings s)
|
11
|
+
: m_ctx(ctx)
|
12
|
+
, m_wparams(wparams)
|
13
|
+
, m_cb(cb)
|
14
|
+
, m_settings(s)
|
15
|
+
, m_board(new Chessboard())
|
16
|
+
{}
|
17
|
+
|
18
|
+
WChess::~WChess() = default;
|
19
|
+
|
20
|
+
void WChess::set_move(const std::string& moves, float prob) const {
|
21
|
+
if (m_cb.set_move) (*m_cb.set_move)(moves, prob);
|
22
|
+
}
|
23
|
+
|
24
|
+
void WChess::set_grammar(const std::string& grammar) const {
|
25
|
+
if (m_cb.set_grammar) (*m_cb.set_grammar)(grammar);
|
26
|
+
}
|
27
|
+
|
28
|
+
bool WChess::get_audio(std::vector<float>& pcmf32) const {
|
29
|
+
if (m_cb.get_audio) return (*m_cb.get_audio)(pcmf32);
|
30
|
+
return false;
|
31
|
+
}
|
32
|
+
|
33
|
+
std::string WChess::stringify_board() const {
|
34
|
+
return m_board->stringifyBoard();
|
35
|
+
}
|
36
|
+
|
37
|
+
std::string WChess::get_grammar() const {
|
38
|
+
return m_board->grammar();
|
39
|
+
}
|
40
|
+
|
41
|
+
void WChess::run() {
|
42
|
+
bool have_prompt = true;
|
43
|
+
bool ask_prompt = !have_prompt;
|
44
|
+
|
45
|
+
float logprob_min = 0.0f;
|
46
|
+
|
47
|
+
float logprob_sum = 0.0f;
|
48
|
+
|
49
|
+
int n_tokens = 0;
|
50
|
+
|
51
|
+
std::vector<float> pcmf32_cur;
|
52
|
+
std::vector<float> pcmf32_prompt;
|
53
|
+
|
54
|
+
const std::string k_prompt = have_prompt ? "" : "rook to d4, f3";
|
55
|
+
int64_t t_ms = 0;
|
56
|
+
|
57
|
+
if (ask_prompt) {
|
58
|
+
fprintf(stdout, "\n");
|
59
|
+
fprintf(stdout, "%s: Say the following phrase: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
|
60
|
+
fprintf(stdout, "\n");
|
61
|
+
|
62
|
+
ask_prompt = false;
|
63
|
+
}
|
64
|
+
|
65
|
+
while (get_audio(pcmf32_cur)) {
|
66
|
+
if (!pcmf32_cur.empty()) {
|
67
|
+
// fprintf(stdout, "%s: Processing ...\n", __func__);
|
68
|
+
|
69
|
+
if (!have_prompt) {
|
70
|
+
const auto txt = ::trim(transcribe(pcmf32_cur, logprob_min, logprob_sum, n_tokens, t_ms));
|
71
|
+
|
72
|
+
fprintf(stdout, "%s: Heard '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", txt.c_str(), "\033[0m", (int) t_ms);
|
73
|
+
|
74
|
+
const float sim = similarity(txt, k_prompt);
|
75
|
+
|
76
|
+
if (txt.length() < 0.8*k_prompt.length() || txt.length() > 1.2*k_prompt.length() || sim < 0.8f) {
|
77
|
+
fprintf(stdout, "%s: WARNING: prompt not recognized, try again\n", __func__);
|
78
|
+
ask_prompt = true;
|
79
|
+
} else {
|
80
|
+
fprintf(stdout, "\n");
|
81
|
+
fprintf(stdout, "%s: The prompt has been recognized!\n", __func__);
|
82
|
+
fprintf(stdout, "%s: Waiting for voice commands ...\n", __func__);
|
83
|
+
fprintf(stdout, "\n");
|
84
|
+
|
85
|
+
// save the audio for the prompt
|
86
|
+
pcmf32_prompt = pcmf32_cur;
|
87
|
+
have_prompt = true;
|
88
|
+
m_board->setPrompt(k_prompt);
|
89
|
+
}
|
90
|
+
} else {
|
91
|
+
if (!pcmf32_prompt.empty()) pcmf32_cur.insert(pcmf32_cur.begin(), pcmf32_prompt.begin(), pcmf32_prompt.end());
|
92
|
+
constexpr size_t MIN_SIZE = 1.2 * WHISPER_SAMPLE_RATE;
|
93
|
+
if (MIN_SIZE > pcmf32_cur.size()) pcmf32_cur.insert(pcmf32_cur.begin(), MIN_SIZE - pcmf32_cur.size(), 0.0f);
|
94
|
+
|
95
|
+
// fprintf(stdout, "%s: grammar rules:\n'%s'\n", __func__, m_board->grammar().c_str());
|
96
|
+
|
97
|
+
auto grammar_parsed = grammar_parser::parse(m_board->grammar().c_str());
|
98
|
+
auto grammar_rules = grammar_parsed.c_rules();
|
99
|
+
|
100
|
+
m_wparams.grammar_rules = grammar_rules.data();
|
101
|
+
m_wparams.n_grammar_rules = grammar_rules.size();
|
102
|
+
|
103
|
+
m_wparams.i_start_rule = grammar_parsed.symbol_ids.at("move");
|
104
|
+
auto txt = ::trim(transcribe(pcmf32_cur, logprob_min, logprob_sum, n_tokens, t_ms));
|
105
|
+
|
106
|
+
const float p = 100.0f * std::exp(logprob_min);
|
107
|
+
|
108
|
+
fprintf(stdout, "%s: heard '%s'\n", __func__, txt.c_str());
|
109
|
+
|
110
|
+
// find the prompt in the text
|
111
|
+
float best_sim = 0.0f;
|
112
|
+
size_t best_len = 0;
|
113
|
+
for (int n = 0.8*k_prompt.size(); n <= 1.2*k_prompt.size(); ++n) {
|
114
|
+
const auto prompt = txt.substr(0, n);
|
115
|
+
|
116
|
+
const float sim = similarity(prompt, k_prompt);
|
117
|
+
|
118
|
+
//fprintf(stderr, "%s: prompt = '%s', sim = %f\n", __func__, prompt.c_str(), sim);
|
119
|
+
|
120
|
+
if (sim > best_sim) {
|
121
|
+
best_sim = sim;
|
122
|
+
best_len = n;
|
123
|
+
}
|
124
|
+
}
|
125
|
+
|
126
|
+
fprintf(stdout, "%s: DEBUG: txt = '%s', prob = %.2f%%\n", __func__, txt.c_str(), p);
|
127
|
+
std::string command = ::trim(txt.substr(best_len));
|
128
|
+
|
129
|
+
fprintf(stdout, "%s: Command '%s%s%s', (t = %d ms)\n", __func__, "\033[1m", command.c_str(), "\033[0m", (int) t_ms);
|
130
|
+
fprintf(stdout, "\n");
|
131
|
+
|
132
|
+
if (!command.empty()) {
|
133
|
+
set_move(m_board->process(command), p);
|
134
|
+
set_grammar(m_board->grammar());
|
135
|
+
}
|
136
|
+
if (m_board->grammar().empty()) {
|
137
|
+
fprintf(stdout, "%s: No more moves possible\n", __func__);
|
138
|
+
break;
|
139
|
+
}
|
140
|
+
}
|
141
|
+
}
|
142
|
+
|
143
|
+
if (ask_prompt) {
|
144
|
+
fprintf(stdout, "\n");
|
145
|
+
fprintf(stdout, "%s: Say the following phrase: '%s%s%s'\n", __func__, "\033[1m", k_prompt.c_str(), "\033[0m");
|
146
|
+
fprintf(stdout, "\n");
|
147
|
+
|
148
|
+
ask_prompt = false;
|
149
|
+
}
|
150
|
+
}
|
151
|
+
}
|
152
|
+
|
153
|
+
std::string WChess::transcribe(
|
154
|
+
const std::vector<float> & pcmf32,
|
155
|
+
float & logprob_min,
|
156
|
+
float & logprob_sum,
|
157
|
+
int & n_tokens,
|
158
|
+
int64_t & t_ms) {
|
159
|
+
const auto t_start = std::chrono::high_resolution_clock::now();
|
160
|
+
|
161
|
+
logprob_min = 0.0f;
|
162
|
+
logprob_sum = 0.0f;
|
163
|
+
n_tokens = 0;
|
164
|
+
t_ms = 0;
|
165
|
+
|
166
|
+
if (whisper_full(m_ctx, m_wparams, pcmf32.data(), pcmf32.size()) != 0) {
|
167
|
+
return {};
|
168
|
+
}
|
169
|
+
|
170
|
+
std::string result;
|
171
|
+
|
172
|
+
const int n_segments = whisper_full_n_segments(m_ctx);
|
173
|
+
for (int i = 0; i < n_segments; ++i) {
|
174
|
+
const char * text = whisper_full_get_segment_text(m_ctx, i);
|
175
|
+
|
176
|
+
result += text;
|
177
|
+
|
178
|
+
const int n = whisper_full_n_tokens(m_ctx, i);
|
179
|
+
for (int j = 0; j < n; ++j) {
|
180
|
+
const auto token = whisper_full_get_token_data(m_ctx, i, j);
|
181
|
+
|
182
|
+
if(token.plog > 0.0f) return {};
|
183
|
+
logprob_min = std::min(logprob_min, token.plog);
|
184
|
+
logprob_sum += token.plog;
|
185
|
+
++n_tokens;
|
186
|
+
}
|
187
|
+
}
|
188
|
+
|
189
|
+
const auto t_end = std::chrono::high_resolution_clock::now();
|
190
|
+
t_ms = std::chrono::duration_cast<std::chrono::milliseconds>(t_end - t_start).count();
|
191
|
+
|
192
|
+
return result;
|
193
|
+
}
|
@@ -0,0 +1,63 @@
|
|
1
|
+
#pragma once
|
2
|
+
#include "whisper.h"
|
3
|
+
#include <string>
|
4
|
+
#include <vector>
|
5
|
+
#include <memory>
|
6
|
+
|
7
|
+
class Chessboard;
|
8
|
+
|
9
|
+
class WChess {
|
10
|
+
public:
|
11
|
+
using CheckRunningCb = bool (*)();
|
12
|
+
using GetAudioCb = bool (*)(std::vector<float> &);
|
13
|
+
using SetMovesCb = void (*)(const std::string &, float);
|
14
|
+
using SetGrammarCb = void (*)(const std::string &);
|
15
|
+
using ClearAudioCb = void (*)();
|
16
|
+
|
17
|
+
struct callbacks {
|
18
|
+
GetAudioCb get_audio = nullptr;
|
19
|
+
SetMovesCb set_move = nullptr;
|
20
|
+
SetGrammarCb set_grammar = nullptr;
|
21
|
+
};
|
22
|
+
|
23
|
+
struct settings {
|
24
|
+
int32_t vad_ms = 2000;
|
25
|
+
int32_t prompt_ms = 5000;
|
26
|
+
int32_t command_ms = 4000;
|
27
|
+
float vad_thold = 0.2f;
|
28
|
+
float freq_thold = 100.0f;
|
29
|
+
bool print_energy = false;
|
30
|
+
};
|
31
|
+
|
32
|
+
WChess(
|
33
|
+
whisper_context * ctx,
|
34
|
+
const whisper_full_params & wparams,
|
35
|
+
callbacks cb,
|
36
|
+
settings s
|
37
|
+
);
|
38
|
+
~WChess();
|
39
|
+
|
40
|
+
void run();
|
41
|
+
|
42
|
+
std::string stringify_board() const;
|
43
|
+
|
44
|
+
std::string get_grammar() const;
|
45
|
+
|
46
|
+
private:
|
47
|
+
bool get_audio(std::vector<float>& pcmf32) const;
|
48
|
+
void set_move(const std::string& moves, float prob) const;
|
49
|
+
void set_grammar(const std::string& grammar) const;
|
50
|
+
|
51
|
+
std::string transcribe(
|
52
|
+
const std::vector<float> & pcmf32,
|
53
|
+
float & logprob_min,
|
54
|
+
float & logprob_sum,
|
55
|
+
int & n_tokens,
|
56
|
+
int64_t & t_ms);
|
57
|
+
|
58
|
+
whisper_context * m_ctx;
|
59
|
+
whisper_full_params m_wparams;
|
60
|
+
const callbacks m_cb;
|
61
|
+
const settings m_settings;
|
62
|
+
std::unique_ptr<Chessboard> m_board;
|
63
|
+
};
|
@@ -0,0 +1,117 @@
|
|
1
|
+
#include "Chessboard.h"
|
2
|
+
|
3
|
+
#define ASSERT(x) \
|
4
|
+
do { \
|
5
|
+
if (!(x)) { \
|
6
|
+
fprintf(stderr, "ASSERT: %s:%d: %s\n", __FILE__, __LINE__, #x); \
|
7
|
+
fflush(stderr); \
|
8
|
+
exit(1); \
|
9
|
+
} \
|
10
|
+
} while (0)
|
11
|
+
|
12
|
+
|
13
|
+
int main() {
|
14
|
+
{
|
15
|
+
Chessboard chess;
|
16
|
+
|
17
|
+
ASSERT(chess.process("pawn to d4") == "d2-d4");
|
18
|
+
ASSERT(chess.process("e5") == "e7-e5");
|
19
|
+
ASSERT(chess.process("c1 h6") == "c1-h6");
|
20
|
+
ASSERT(chess.process("queen h4") == "d8-h4");
|
21
|
+
ASSERT(chess.process("bishop to g5") == "h6-g5");
|
22
|
+
ASSERT(chess.process("bishop to b4") == "f8-b4");
|
23
|
+
ASSERT(chess.process("c4") == "");
|
24
|
+
ASSERT(chess.process("knight c3") == "b1-c3");
|
25
|
+
ASSERT(chess.process("knight c6") == "b8-c6");
|
26
|
+
ASSERT(chess.process("f3") == "");
|
27
|
+
}
|
28
|
+
|
29
|
+
{
|
30
|
+
Chessboard chess;
|
31
|
+
|
32
|
+
ASSERT(chess.process("d4") == "d2-d4");
|
33
|
+
ASSERT(chess.process("e5") == "e7-e5");
|
34
|
+
ASSERT(chess.process("e4") == "e2-e4");
|
35
|
+
ASSERT(chess.process("queen h4") == "d8-h4");
|
36
|
+
ASSERT(chess.process("queen h5") == "d1-h5");
|
37
|
+
ASSERT(chess.process("f5") == "");
|
38
|
+
ASSERT(chess.process("g6") == "g7-g6");
|
39
|
+
ASSERT(chess.process("knight e2") == "g1-e2");
|
40
|
+
ASSERT(chess.process("f5") == "f7-f5");
|
41
|
+
ASSERT(chess.process("knight g3") == "e2-g3");
|
42
|
+
ASSERT(chess.process("g5") == "");
|
43
|
+
ASSERT(chess.process("king e7") == "e8-e7");
|
44
|
+
ASSERT(chess.process("f4") == "f2-f4");
|
45
|
+
ASSERT(chess.process("g5") == "g6-g5");
|
46
|
+
}
|
47
|
+
|
48
|
+
{
|
49
|
+
Chessboard chess;
|
50
|
+
|
51
|
+
ASSERT(chess.process("e4") == "e2-e4");
|
52
|
+
ASSERT(chess.process("c5") == "c7-c5");
|
53
|
+
ASSERT(chess.process("e5") == "e4-e5");
|
54
|
+
ASSERT(chess.process("c4") == "c5-c4");
|
55
|
+
ASSERT(chess.process("e6") == "e5-e6");
|
56
|
+
ASSERT(chess.process("c3") == "c4-c3");
|
57
|
+
ASSERT(chess.process("e7") == "");
|
58
|
+
ASSERT(chess.process("f7") == "e6-f7");
|
59
|
+
ASSERT(chess.process("d2") == "");
|
60
|
+
ASSERT(chess.process("king to f7") == "e8-f7");
|
61
|
+
ASSERT(chess.process("f4") == "f2-f4");
|
62
|
+
ASSERT(chess.process("d2") == "c3-d2");
|
63
|
+
ASSERT(chess.process("f5") == "");
|
64
|
+
ASSERT(chess.process("king to e2") == "e1-e2");
|
65
|
+
ASSERT(chess.process("king to g6") == "f7-g6");
|
66
|
+
ASSERT(chess.process("f5") == "f4-f5");
|
67
|
+
ASSERT(chess.process("e6") == "");
|
68
|
+
ASSERT(chess.process("king to h5") == "g6-h5");
|
69
|
+
ASSERT(chess.process("g4") == "g2-g4");
|
70
|
+
ASSERT(chess.process("king to g5") == "h5-g5");
|
71
|
+
ASSERT(chess.process("h4") == "h2-h4");
|
72
|
+
ASSERT(chess.process("king to h5") == "");
|
73
|
+
ASSERT(chess.process("king to g6") == "");
|
74
|
+
ASSERT(chess.process("king to h6") == "g5-h6");
|
75
|
+
ASSERT(chess.process("bishop to d2") == "c1-d2");
|
76
|
+
ASSERT(chess.process("king to g5") == "");
|
77
|
+
ASSERT(chess.process("g5") == "g7-g5");
|
78
|
+
}
|
79
|
+
|
80
|
+
{
|
81
|
+
Chessboard chess;
|
82
|
+
ASSERT(chess.process("f4") == "f2-f4");
|
83
|
+
ASSERT(chess.process("e5") == "e7-e5");
|
84
|
+
ASSERT(chess.process("g4") == "g2-g4");
|
85
|
+
ASSERT(chess.process("queen to h4") == "d8-h4#");
|
86
|
+
ASSERT(chess.process("knight f3") == "");
|
87
|
+
ASSERT(chess.grammar().empty());
|
88
|
+
}
|
89
|
+
|
90
|
+
{
|
91
|
+
Chessboard chess;
|
92
|
+
ASSERT(chess.process("f4") == "f2-f4");
|
93
|
+
ASSERT(chess.process("e5") == "e7-e5");
|
94
|
+
ASSERT(chess.process("g4") == "g2-g4");
|
95
|
+
ASSERT(chess.process("d5") == "d7-d5");
|
96
|
+
ASSERT(chess.process("g1 f3") == "g1-f3");
|
97
|
+
ASSERT(chess.process("queen to h4") == "d8-h4");
|
98
|
+
ASSERT(!chess.grammar().empty());
|
99
|
+
}
|
100
|
+
|
101
|
+
{
|
102
|
+
Chessboard chess;
|
103
|
+
ASSERT(chess.process("knight c3") == "b1-c3");
|
104
|
+
ASSERT(chess.process("knight c6") == "b8-c6");
|
105
|
+
ASSERT(chess.process("knight b5") == "c3-b5");
|
106
|
+
ASSERT(chess.process("knight f6") == "g8-f6");
|
107
|
+
ASSERT(chess.process("knight d6") == "b5-d6");
|
108
|
+
ASSERT(chess.process("knight d4") == "");
|
109
|
+
ASSERT(chess.process("d6") == "c7-d6");
|
110
|
+
ASSERT(chess.process("e4") == "e2-e4");
|
111
|
+
ASSERT(chess.process("knight d4") == "c6-d4");
|
112
|
+
ASSERT(chess.process("d3") == "d2-d3");
|
113
|
+
ASSERT(chess.process("knight e4") == "f6-e4");
|
114
|
+
ASSERT(chess.process("king to e2") == "");
|
115
|
+
ASSERT(chess.process("king to d2") == "");
|
116
|
+
}
|
117
|
+
}
|
@@ -0,0 +1,249 @@
|
|
1
|
+
// Command line voice assisted chess
|
2
|
+
//
|
3
|
+
// Speak chess move commands to the microphone.
|
4
|
+
// The moves will translated to chessboard positions.
|
5
|
+
//
|
6
|
+
//
|
7
|
+
|
8
|
+
#include "WChess.h"
|
9
|
+
#include "common-sdl.h"
|
10
|
+
#include <iostream>
|
11
|
+
|
12
|
+
#include <memory>
|
13
|
+
#include <thread>
|
14
|
+
|
15
|
+
// command-line parameters
|
16
|
+
struct whisper_params {
|
17
|
+
int32_t n_threads = std::min(4, (int32_t) std::thread::hardware_concurrency());
|
18
|
+
int32_t prompt_ms = 5000;
|
19
|
+
int32_t command_ms = 8000;
|
20
|
+
int32_t capture_id = -1;
|
21
|
+
int32_t max_tokens = 32;
|
22
|
+
int32_t audio_ctx = 0;
|
23
|
+
|
24
|
+
float vad_thold = 0.6f;
|
25
|
+
float freq_thold = 100.0f;
|
26
|
+
|
27
|
+
float grammar_penalty = 100.0f;
|
28
|
+
|
29
|
+
bool translate = false;
|
30
|
+
bool print_special = false;
|
31
|
+
bool print_energy = false;
|
32
|
+
bool no_timestamps = true;
|
33
|
+
bool use_gpu = true;
|
34
|
+
bool flash_attn = false;
|
35
|
+
|
36
|
+
std::string language = "en";
|
37
|
+
std::string model = "models/ggml-base.en.bin";
|
38
|
+
std::string fname_out;
|
39
|
+
std::string commands;
|
40
|
+
std::string prompt;
|
41
|
+
std::string context;
|
42
|
+
std::string grammar;
|
43
|
+
};
|
44
|
+
|
45
|
+
void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & params) {
|
46
|
+
fprintf(stderr, "\n");
|
47
|
+
fprintf(stderr, "usage: %s [options]\n", argv[0]);
|
48
|
+
fprintf(stderr, "\n");
|
49
|
+
fprintf(stderr, "options:\n");
|
50
|
+
fprintf(stderr, " -h, --help [default] show this help message and exit\n");
|
51
|
+
fprintf(stderr, " -t N, --threads N [%-7d] number of threads to use during computation\n", params.n_threads);
|
52
|
+
fprintf(stderr, " -pms N, --prompt-ms N [%-7d] prompt duration in milliseconds\n", params.prompt_ms);
|
53
|
+
fprintf(stderr, " -cms N, --command-ms N [%-7d] command duration in milliseconds\n", params.command_ms);
|
54
|
+
fprintf(stderr, " -c ID, --capture ID [%-7d] capture device ID\n", params.capture_id);
|
55
|
+
fprintf(stderr, " -mt N, --max-tokens N [%-7d] maximum number of tokens per audio chunk\n", params.max_tokens);
|
56
|
+
fprintf(stderr, " -ac N, --audio-ctx N [%-7d] audio context size (0 - all)\n", params.audio_ctx);
|
57
|
+
fprintf(stderr, " -vth N, --vad-thold N [%-7.2f] voice activity detection threshold\n", params.vad_thold);
|
58
|
+
fprintf(stderr, " -fth N, --freq-thold N [%-7.2f] high-pass frequency cutoff\n", params.freq_thold);
|
59
|
+
fprintf(stderr, " -tr, --translate [%-7s] translate from source language to english\n", params.translate ? "true" : "false");
|
60
|
+
fprintf(stderr, " -ps, --print-special [%-7s] print special tokens\n", params.print_special ? "true" : "false");
|
61
|
+
fprintf(stderr, " -pe, --print-energy [%-7s] print sound energy (for debugging)\n", params.print_energy ? "true" : "false");
|
62
|
+
fprintf(stderr, " -ng, --no-gpu [%-7s] disable GPU\n", params.use_gpu ? "false" : "true");
|
63
|
+
fprintf(stderr, " -fa, --flash-attn [%-7s] flash attention during decoding\n", params.flash_attn ? "true" : "false");
|
64
|
+
fprintf(stderr, " -l LANG, --language LANG [%-7s] spoken language\n", params.language.c_str());
|
65
|
+
fprintf(stderr, " -m FNAME, --model FNAME [%-7s] model path\n", params.model.c_str());
|
66
|
+
fprintf(stderr, " -f FNAME, --file FNAME [%-7s] text output file name\n", params.fname_out.c_str());
|
67
|
+
fprintf(stderr, " -cmd FNAME, --commands FNAME [%-7s] text file with allowed commands\n", params.commands.c_str());
|
68
|
+
fprintf(stderr, " -p, --prompt [%-7s] the required activation prompt\n", params.prompt.c_str());
|
69
|
+
fprintf(stderr, " -ctx, --context [%-7s] sample text to help the transcription\n", params.context.c_str());
|
70
|
+
fprintf(stderr, " --grammar-penalty N [%-7.1f] scales down logits of nongrammar tokens\n", params.grammar_penalty);
|
71
|
+
fprintf(stderr, "\n");
|
72
|
+
}
|
73
|
+
|
74
|
+
bool whisper_params_parse(int argc, char ** argv, whisper_params & params) {
|
75
|
+
for (int i = 1; i < argc; i++) {
|
76
|
+
std::string arg = argv[i];
|
77
|
+
|
78
|
+
if (arg == "-h" || arg == "--help") {
|
79
|
+
whisper_print_usage(argc, argv, params);
|
80
|
+
exit(0);
|
81
|
+
}
|
82
|
+
else if (arg == "-t" || arg == "--threads") { params.n_threads = std::stoi(argv[++i]); }
|
83
|
+
else if (arg == "-pms" || arg == "--prompt-ms") { params.prompt_ms = std::stoi(argv[++i]); }
|
84
|
+
else if (arg == "-cms" || arg == "--command-ms") { params.command_ms = std::stoi(argv[++i]); }
|
85
|
+
else if (arg == "-c" || arg == "--capture") { params.capture_id = std::stoi(argv[++i]); }
|
86
|
+
else if (arg == "-mt" || arg == "--max-tokens") { params.max_tokens = std::stoi(argv[++i]); }
|
87
|
+
else if (arg == "-ac" || arg == "--audio-ctx") { params.audio_ctx = std::stoi(argv[++i]); }
|
88
|
+
else if (arg == "-vth" || arg == "--vad-thold") { params.vad_thold = std::stof(argv[++i]); }
|
89
|
+
else if (arg == "-fth" || arg == "--freq-thold") { params.freq_thold = std::stof(argv[++i]); }
|
90
|
+
else if (arg == "-tr" || arg == "--translate") { params.translate = true; }
|
91
|
+
else if (arg == "-ps" || arg == "--print-special") { params.print_special = true; }
|
92
|
+
else if (arg == "-pe" || arg == "--print-energy") { params.print_energy = true; }
|
93
|
+
else if (arg == "-ng" || arg == "--no-gpu") { params.use_gpu = false; }
|
94
|
+
else if (arg == "-fa" || arg == "--flash-attn") { params.flash_attn = true; }
|
95
|
+
else if (arg == "-l" || arg == "--language") { params.language = argv[++i]; }
|
96
|
+
else if (arg == "-m" || arg == "--model") { params.model = argv[++i]; }
|
97
|
+
else if (arg == "-f" || arg == "--file") { params.fname_out = argv[++i]; }
|
98
|
+
else if (arg == "-cmd" || arg == "--commands") { params.commands = argv[++i]; }
|
99
|
+
else if (arg == "-p" || arg == "--prompt") { params.prompt = argv[++i]; }
|
100
|
+
else if (arg == "-ctx" || arg == "--context") { params.context = argv[++i]; }
|
101
|
+
else if ( arg == "--grammar-penalty") { params.grammar_penalty = std::stof(argv[++i]); }
|
102
|
+
else {
|
103
|
+
fprintf(stderr, "error: unknown argument: %s\n", arg.c_str());
|
104
|
+
whisper_print_usage(argc, argv, params);
|
105
|
+
exit(0);
|
106
|
+
}
|
107
|
+
}
|
108
|
+
|
109
|
+
return true;
|
110
|
+
}
|
111
|
+
|
112
|
+
std::unique_ptr<WChess> g_wchess;
|
113
|
+
int g_moveCount = 0;
|
114
|
+
void set_move(const std::string & move, float) {
|
115
|
+
if (!move.empty()) {
|
116
|
+
g_moveCount++;
|
117
|
+
fprintf(stdout, "Move: %s\n\n", move.c_str());
|
118
|
+
}
|
119
|
+
else fprintf(stdout, "Move rejected\n\n");
|
120
|
+
fprintf(stdout, "%s\n", g_wchess->stringify_board().c_str());
|
121
|
+
fprintf(stdout, "%s\n", g_moveCount ? "White's turn" : "Black's turn");
|
122
|
+
}
|
123
|
+
|
124
|
+
audio_async g_audio(30*1000);
|
125
|
+
bool g_listening = false;
|
126
|
+
std::vector<float> g_pcmf32;
|
127
|
+
|
128
|
+
bool read_input() {
|
129
|
+
std::string input;
|
130
|
+
while (true) {
|
131
|
+
fprintf(stdout, "[(l)isten/(p)ause/(q)uit]: ");
|
132
|
+
std::cin >> input;
|
133
|
+
fprintf(stdout, "\n");
|
134
|
+
if (input[0] == 'q') {
|
135
|
+
fprintf(stdout, "Quitting\n");
|
136
|
+
return false;
|
137
|
+
}
|
138
|
+
if (input[0] == 'l') {
|
139
|
+
if (!g_listening) {
|
140
|
+
fprintf(stdout, "Listening\n");
|
141
|
+
g_listening = true;
|
142
|
+
g_pcmf32.clear();
|
143
|
+
g_audio.resume();
|
144
|
+
g_audio.clear();
|
145
|
+
}
|
146
|
+
else fprintf(stdout, "Still listening\n");
|
147
|
+
return true;
|
148
|
+
}
|
149
|
+
else {
|
150
|
+
if (g_listening) {
|
151
|
+
g_listening = false;
|
152
|
+
g_audio.get(0, g_pcmf32);
|
153
|
+
g_audio.pause();
|
154
|
+
fprintf(stdout, "Processing\n");
|
155
|
+
}
|
156
|
+
else fprintf(stdout, "Not listening\n");
|
157
|
+
return true;
|
158
|
+
}
|
159
|
+
}
|
160
|
+
return true;
|
161
|
+
}
|
162
|
+
|
163
|
+
bool get_audio(std::vector<float> & pcmf32_cur) {
|
164
|
+
if (!read_input()) return false;
|
165
|
+
if (!g_pcmf32.empty()) pcmf32_cur = std::move(g_pcmf32);
|
166
|
+
else pcmf32_cur.clear();
|
167
|
+
return true;
|
168
|
+
}
|
169
|
+
|
170
|
+
int main(int argc, char ** argv) {
|
171
|
+
whisper_params params;
|
172
|
+
|
173
|
+
if (whisper_params_parse(argc, argv, params) == false) {
|
174
|
+
return 1;
|
175
|
+
}
|
176
|
+
|
177
|
+
if (whisper_lang_id(params.language.c_str()) == -1) {
|
178
|
+
fprintf(stderr, "error: unknown language '%s'\n", params.language.c_str());
|
179
|
+
whisper_print_usage(argc, argv, params);
|
180
|
+
exit(0);
|
181
|
+
}
|
182
|
+
|
183
|
+
// whisper init
|
184
|
+
|
185
|
+
struct whisper_context_params cparams = whisper_context_default_params();
|
186
|
+
|
187
|
+
cparams.use_gpu = params.use_gpu;
|
188
|
+
cparams.flash_attn = params.flash_attn;
|
189
|
+
|
190
|
+
struct whisper_context * ctx = whisper_init_from_file_with_params(params.model.c_str(), cparams);
|
191
|
+
if (!ctx) {
|
192
|
+
fprintf(stderr, "%s: whisper_init_from_file_with_params() failed!\n", __func__);
|
193
|
+
return 1;
|
194
|
+
}
|
195
|
+
|
196
|
+
// init audio
|
197
|
+
|
198
|
+
if (!g_audio.init(params.capture_id, WHISPER_SAMPLE_RATE)) {
|
199
|
+
fprintf(stderr, "%s: audio.init() failed!\n", __func__);
|
200
|
+
return 1;
|
201
|
+
}
|
202
|
+
|
203
|
+
struct whisper_full_params wparams = whisper_full_default_params(whisper_sampling_strategy::WHISPER_SAMPLING_GREEDY);
|
204
|
+
wparams.offset_ms = 0;
|
205
|
+
wparams.translate = false;
|
206
|
+
wparams.no_context = true;
|
207
|
+
wparams.single_segment = true;
|
208
|
+
wparams.print_realtime = false;
|
209
|
+
wparams.print_progress = false;
|
210
|
+
wparams.print_timestamps = true;
|
211
|
+
wparams.print_special = false;
|
212
|
+
wparams.no_timestamps = true;
|
213
|
+
|
214
|
+
wparams.max_tokens = 32;
|
215
|
+
wparams.audio_ctx = 768; // partial encoder context for better performance
|
216
|
+
|
217
|
+
wparams.temperature = 0.0f;
|
218
|
+
wparams.temperature_inc = 2.0f;
|
219
|
+
wparams.greedy.best_of = 1;
|
220
|
+
|
221
|
+
wparams.beam_search.beam_size = 1;
|
222
|
+
|
223
|
+
wparams.language = "en";
|
224
|
+
|
225
|
+
wparams.grammar_penalty = 100.0;
|
226
|
+
|
227
|
+
wparams.initial_prompt = params.context.data();
|
228
|
+
|
229
|
+
WChess::callbacks cb;
|
230
|
+
cb.get_audio = get_audio;
|
231
|
+
cb.set_move = set_move;
|
232
|
+
|
233
|
+
WChess::settings s;
|
234
|
+
s.vad_ms = 2000;
|
235
|
+
s.prompt_ms = params.prompt_ms;
|
236
|
+
s.command_ms = params.command_ms;
|
237
|
+
s.vad_thold = params.vad_thold;
|
238
|
+
s.freq_thold = params.freq_thold;
|
239
|
+
s.print_energy = params.print_energy;
|
240
|
+
|
241
|
+
g_wchess.reset(new WChess(ctx, wparams, cb, s));
|
242
|
+
set_move("start", 0);
|
243
|
+
g_wchess->run();
|
244
|
+
|
245
|
+
whisper_print_timings(ctx);
|
246
|
+
whisper_free(ctx);
|
247
|
+
|
248
|
+
return 0;
|
249
|
+
}
|