@simulatte/doppler 0.1.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/BRANDING.md +14 -0
- package/LICENSE +201 -0
- package/NOTICE +5 -0
- package/README.md +85 -0
- package/SECURITY.md +19 -0
- package/package.json +144 -0
- package/src/adapters/adapter-manager.d.ts +200 -0
- package/src/adapters/adapter-manager.js +509 -0
- package/src/adapters/adapter-manifest.d.ts +290 -0
- package/src/adapters/adapter-manifest.js +320 -0
- package/src/adapters/adapter-registry.d.ts +192 -0
- package/src/adapters/adapter-registry.js +466 -0
- package/src/adapters/index.d.ts +89 -0
- package/src/adapters/index.js +42 -0
- package/src/adapters/lora-loader.d.ts +105 -0
- package/src/adapters/lora-loader.js +397 -0
- package/src/bootstrap.d.ts +1 -0
- package/src/bootstrap.js +30 -0
- package/src/bridge/extension/background.d.ts +14 -0
- package/src/bridge/extension/background.js +168 -0
- package/src/bridge/extension/manifest.json +34 -0
- package/src/bridge/extension-client.d.ts +109 -0
- package/src/bridge/extension-client.js +369 -0
- package/src/bridge/index.d.ts +68 -0
- package/src/bridge/index.js +51 -0
- package/src/bridge/protocol.d.ts +96 -0
- package/src/bridge/protocol.js +130 -0
- package/src/browser/browser-converter.d.ts +71 -0
- package/src/browser/browser-converter.js +947 -0
- package/src/browser/file-picker.d.ts +63 -0
- package/src/browser/file-picker.js +275 -0
- package/src/browser/gguf-importer.d.ts +136 -0
- package/src/browser/gguf-importer.js +532 -0
- package/src/browser/gguf-parser-browser.d.ts +14 -0
- package/src/browser/gguf-parser-browser.js +17 -0
- package/src/browser/quantization.d.ts +69 -0
- package/src/browser/quantization.js +328 -0
- package/src/browser/safetensors-parser-browser.d.ts +193 -0
- package/src/browser/safetensors-parser-browser.js +264 -0
- package/src/browser/shard-io-browser.d.ts +57 -0
- package/src/browser/shard-io-browser.js +89 -0
- package/src/browser/tensor-source-download.d.ts +27 -0
- package/src/browser/tensor-source-download.js +239 -0
- package/src/browser/tensor-source-file.d.ts +26 -0
- package/src/browser/tensor-source-file.js +53 -0
- package/src/browser/tensor-source-http.d.ts +28 -0
- package/src/browser/tensor-source-http.js +126 -0
- package/src/client/doppler-provider/generation.d.ts +25 -0
- package/src/client/doppler-provider/generation.js +114 -0
- package/src/client/doppler-provider/index.d.ts +2 -0
- package/src/client/doppler-provider/index.js +3 -0
- package/src/client/doppler-provider/model-manager.d.ts +61 -0
- package/src/client/doppler-provider/model-manager.js +667 -0
- package/src/client/doppler-provider/provider.d.ts +5 -0
- package/src/client/doppler-provider/provider.js +102 -0
- package/src/client/doppler-provider/source-runtime.d.ts +22 -0
- package/src/client/doppler-provider/source-runtime.js +522 -0
- package/src/client/doppler-provider/types.d.ts +127 -0
- package/src/client/doppler-provider/types.js +17 -0
- package/src/client/doppler-provider.d.ts +46 -0
- package/src/client/doppler-provider.js +36 -0
- package/src/config/README.md +69 -0
- package/src/config/backward-registry-loader.d.ts +3 -0
- package/src/config/backward-registry-loader.js +8 -0
- package/src/config/index.d.ts +63 -0
- package/src/config/index.js +31 -0
- package/src/config/kernel-path-loader.d.ts +149 -0
- package/src/config/kernel-path-loader.js +534 -0
- package/src/config/kernels/backward-registry.json +99 -0
- package/src/config/kernels/kernel-ref-digests.d.ts +1 -0
- package/src/config/kernels/kernel-ref-digests.js +214 -0
- package/src/config/kernels/kernel-ref.d.ts +17 -0
- package/src/config/kernels/kernel-ref.js +75 -0
- package/src/config/kernels/moe/gpt-oss.paths.json +49 -0
- package/src/config/kernels/registry.d.ts +86 -0
- package/src/config/kernels/registry.js +103 -0
- package/src/config/kernels/registry.json +6771 -0
- package/src/config/loader.d.ts +57 -0
- package/src/config/loader.js +513 -0
- package/src/config/merge.d.ts +142 -0
- package/src/config/merge.js +389 -0
- package/src/config/param-categories.d.ts +17 -0
- package/src/config/param-categories.js +72 -0
- package/src/config/param-validator.d.ts +26 -0
- package/src/config/param-validator.js +235 -0
- package/src/config/platforms/amd-rdna3.json +16 -0
- package/src/config/platforms/apple-m1.json +16 -0
- package/src/config/platforms/apple-m2.json +16 -0
- package/src/config/platforms/apple-m3.json +16 -0
- package/src/config/platforms/generic.json +14 -0
- package/src/config/platforms/loader.d.ts +65 -0
- package/src/config/platforms/loader.js +153 -0
- package/src/config/platforms/nvidia-rtx30.json +16 -0
- package/src/config/platforms/nvidia-rtx40.json +16 -0
- package/src/config/presets/kernel-paths/embeddinggemma-f16-f32a.json +60 -0
- package/src/config/presets/kernel-paths/embeddinggemma-f32-f32a.json +60 -0
- package/src/config/presets/kernel-paths/embeddinggemma-q4k-dequant-f32a.json +60 -0
- package/src/config/presets/kernel-paths/gemma2-f16-f16a.json +61 -0
- package/src/config/presets/kernel-paths/gemma2-f16-f32a.json +60 -0
- package/src/config/presets/kernel-paths/gemma2-q4k-dequant-f16a.json +61 -0
- package/src/config/presets/kernel-paths/gemma2-q4k-dequant-f32a.json +60 -0
- package/src/config/presets/kernel-paths/gemma2-q4k-fused-f32a.json +57 -0
- package/src/config/presets/kernel-paths/gemma3-f16-fused-f16a-online.json +200 -0
- package/src/config/presets/kernel-paths/gemma3-f16-fused-f32a-online.json +223 -0
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f16a-online.json +60 -0
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f32a-online.json +61 -0
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f32a.json +61 -0
- package/src/config/presets/kernel-paths/lfm2-q4k-dequant-f32a-online.json +61 -0
- package/src/config/presets/kernel-paths/registry.json +103 -0
- package/src/config/presets/models/deepseek.json +20 -0
- package/src/config/presets/models/diffusion.json +10 -0
- package/src/config/presets/models/embeddinggemma.json +74 -0
- package/src/config/presets/models/functiongemma.json +31 -0
- package/src/config/presets/models/gemma2.json +59 -0
- package/src/config/presets/models/gemma3.json +75 -0
- package/src/config/presets/models/gpt-oss.json +68 -0
- package/src/config/presets/models/kimi-k2.json +25 -0
- package/src/config/presets/models/lfm2.json +83 -0
- package/src/config/presets/models/llama3.json +40 -0
- package/src/config/presets/models/mamba.json +34 -0
- package/src/config/presets/models/mixtral.json +37 -0
- package/src/config/presets/models/modernbert.json +32 -0
- package/src/config/presets/models/qwen3.json +41 -0
- package/src/config/presets/models/transformer.json +73 -0
- package/src/config/presets/models/translategemma.json +30 -0
- package/src/config/presets/platforms/nvidia-gb200-8gpu.json +45 -0
- package/src/config/presets/platforms/nvidia-gb200-nvl72.json +45 -0
- package/src/config/presets/platforms/nvidia-gh200-nvl2.json +44 -0
- package/src/config/presets/platforms/nvidia-gh200.json +44 -0
- package/src/config/presets/runtime/compute/f16-activations.json +30 -0
- package/src/config/presets/runtime/compute/f16-batched.json +32 -0
- package/src/config/presets/runtime/default.json +101 -0
- package/src/config/presets/runtime/diagnostics/debug-logits.json +53 -0
- package/src/config/presets/runtime/experiments/bench/gemma3-bench-q4k.json +53 -0
- package/src/config/presets/runtime/experiments/debug/gemma3-debug-q4k.json +210 -0
- package/src/config/presets/runtime/experiments/verify/gemma3-verify.json +39 -0
- package/src/config/presets/runtime/kernels/dequant-f16-q4k.json +20 -0
- package/src/config/presets/runtime/kernels/dequant-f32-q4k.json +20 -0
- package/src/config/presets/runtime/kernels/fused-q4k.json +20 -0
- package/src/config/presets/runtime/kernels/safe-q4k.json +20 -0
- package/src/config/presets/runtime/model/gemma2-debug.json +77 -0
- package/src/config/presets/runtime/model/gemma2-pipeline-debug.json +66 -0
- package/src/config/presets/runtime/model/gemma2-pipeline.json +75 -0
- package/src/config/presets/runtime/model/gemma3-layer-probe.json +85 -0
- package/src/config/presets/runtime/modes/bench.json +37 -0
- package/src/config/presets/runtime/modes/debug.json +39 -0
- package/src/config/presets/runtime/modes/default.json +10 -0
- package/src/config/presets/runtime/modes/embedding-bench.json +28 -0
- package/src/config/presets/runtime/modes/embedding.json +54 -0
- package/src/config/presets/runtime/modes/low-memory.json +40 -0
- package/src/config/presets/runtime/modes/production.json +48 -0
- package/src/config/presets/runtime/modes/simulation.json +30 -0
- package/src/config/presets/runtime/modes/trace-layers.json +126 -0
- package/src/config/presets/runtime/platform/metal-apple-q4k.json +11 -0
- package/src/config/runtime-merge.d.ts +5 -0
- package/src/config/runtime-merge.js +21 -0
- package/src/config/runtime.d.ts +28 -0
- package/src/config/runtime.js +56 -0
- package/src/config/schema/adapter.schema.d.ts +53 -0
- package/src/config/schema/adapter.schema.js +60 -0
- package/src/config/schema/backward-registry.schema.d.ts +14 -0
- package/src/config/schema/backward-registry.schema.js +46 -0
- package/src/config/schema/benchmark.schema.d.ts +54 -0
- package/src/config/schema/benchmark.schema.js +74 -0
- package/src/config/schema/bridge.schema.d.ts +25 -0
- package/src/config/schema/bridge.schema.js +22 -0
- package/src/config/schema/buffer-pool.schema.d.ts +92 -0
- package/src/config/schema/buffer-pool.schema.js +50 -0
- package/src/config/schema/conversion.schema.d.ts +183 -0
- package/src/config/schema/conversion.schema.js +13 -0
- package/src/config/schema/converter.schema.d.ts +123 -0
- package/src/config/schema/converter.schema.js +136 -0
- package/src/config/schema/debug.schema.d.ts +245 -0
- package/src/config/schema/debug.schema.js +106 -0
- package/src/config/schema/diffusion.schema.d.ts +88 -0
- package/src/config/schema/diffusion.schema.js +62 -0
- package/src/config/schema/distill-training.schema.d.ts +48 -0
- package/src/config/schema/distill-training.schema.js +139 -0
- package/src/config/schema/distribution.schema.d.ts +155 -0
- package/src/config/schema/distribution.schema.js +81 -0
- package/src/config/schema/doppler.schema.d.ts +75 -0
- package/src/config/schema/doppler.schema.js +352 -0
- package/src/config/schema/ecosystem.schema.d.ts +255 -0
- package/src/config/schema/ecosystem.schema.js +534 -0
- package/src/config/schema/emulation.schema.d.ts +351 -0
- package/src/config/schema/emulation.schema.js +299 -0
- package/src/config/schema/energy.schema.d.ts +102 -0
- package/src/config/schema/energy.schema.js +72 -0
- package/src/config/schema/execution-v0.schema.d.ts +187 -0
- package/src/config/schema/execution-v0.schema.js +55 -0
- package/src/config/schema/gpu-cache.schema.d.ts +26 -0
- package/src/config/schema/gpu-cache.schema.js +8 -0
- package/src/config/schema/harness.schema.d.ts +32 -0
- package/src/config/schema/harness.schema.js +20 -0
- package/src/config/schema/hotswap.schema.d.ts +55 -0
- package/src/config/schema/hotswap.schema.js +18 -0
- package/src/config/schema/index.d.ts +863 -0
- package/src/config/schema/index.js +471 -0
- package/src/config/schema/inference-defaults.schema.d.ts +276 -0
- package/src/config/schema/inference-defaults.schema.js +185 -0
- package/src/config/schema/inference.schema.d.ts +289 -0
- package/src/config/schema/inference.schema.js +39 -0
- package/src/config/schema/intent-bundle.schema.d.ts +28 -0
- package/src/config/schema/intent-bundle.schema.js +12 -0
- package/src/config/schema/kernel-path.schema.d.ts +173 -0
- package/src/config/schema/kernel-path.schema.js +9 -0
- package/src/config/schema/kernel-registry.schema.d.ts +199 -0
- package/src/config/schema/kernel-registry.schema.js +46 -0
- package/src/config/schema/kernel-thresholds.schema.d.ts +302 -0
- package/src/config/schema/kernel-thresholds.schema.js +187 -0
- package/src/config/schema/kernel-warmup.schema.d.ts +19 -0
- package/src/config/schema/kernel-warmup.schema.js +5 -0
- package/src/config/schema/kvcache.schema.d.ts +131 -0
- package/src/config/schema/kvcache.schema.js +31 -0
- package/src/config/schema/loading.schema.d.ts +153 -0
- package/src/config/schema/loading.schema.js +84 -0
- package/src/config/schema/lora.schema.d.ts +12 -0
- package/src/config/schema/lora.schema.js +12 -0
- package/src/config/schema/manifest.schema.d.ts +500 -0
- package/src/config/schema/manifest.schema.js +130 -0
- package/src/config/schema/memory-limits.schema.d.ts +107 -0
- package/src/config/schema/memory-limits.schema.js +57 -0
- package/src/config/schema/moe.schema.d.ts +78 -0
- package/src/config/schema/moe.schema.js +31 -0
- package/src/config/schema/platform.schema.d.ts +121 -0
- package/src/config/schema/platform.schema.js +1 -0
- package/src/config/schema/preset.schema.d.ts +124 -0
- package/src/config/schema/preset.schema.js +1 -0
- package/src/config/schema/quantization-defaults.schema.d.ts +34 -0
- package/src/config/schema/quantization-defaults.schema.js +5 -0
- package/src/config/schema/quantization.schema.d.ts +10 -0
- package/src/config/schema/quantization.schema.js +33 -0
- package/src/config/schema/shared-runtime.schema.d.ts +75 -0
- package/src/config/schema/shared-runtime.schema.js +45 -0
- package/src/config/schema/speculative.schema.d.ts +21 -0
- package/src/config/schema/speculative.schema.js +11 -0
- package/src/config/schema/storage.schema.d.ts +123 -0
- package/src/config/schema/storage.schema.js +66 -0
- package/src/config/schema/tooling.schema.d.ts +29 -0
- package/src/config/schema/tooling.schema.js +12 -0
- package/src/config/schema/training-metrics.schema.d.ts +89 -0
- package/src/config/schema/training-metrics.schema.js +374 -0
- package/src/config/schema/training.schema.d.ts +88 -0
- package/src/config/schema/training.schema.js +106 -0
- package/src/config/schema/tuner.schema.d.ts +39 -0
- package/src/config/schema/tuner.schema.js +13 -0
- package/src/config/schema/ul-training.schema.d.ts +61 -0
- package/src/config/schema/ul-training.schema.js +140 -0
- package/src/config/schema/units.schema.d.ts +27 -0
- package/src/config/schema/units.schema.js +26 -0
- package/src/config/training-defaults.d.ts +24 -0
- package/src/config/training-defaults.js +91 -0
- package/src/converter/conversion-plan.d.ts +64 -0
- package/src/converter/conversion-plan.js +472 -0
- package/src/converter/core.d.ts +247 -0
- package/src/converter/core.js +1329 -0
- package/src/converter/execution-v0-manifest.d.ts +15 -0
- package/src/converter/execution-v0-manifest.js +146 -0
- package/src/converter/index.d.ts +98 -0
- package/src/converter/index.js +59 -0
- package/src/converter/manifest-inference.d.ts +20 -0
- package/src/converter/manifest-inference.js +492 -0
- package/src/converter/parsers/diffusion.d.ts +50 -0
- package/src/converter/parsers/diffusion.js +270 -0
- package/src/converter/parsers/gguf.d.ts +22 -0
- package/src/converter/parsers/gguf.js +46 -0
- package/src/converter/parsers/index.d.ts +21 -0
- package/src/converter/parsers/index.js +12 -0
- package/src/converter/parsers/transformer.d.ts +16 -0
- package/src/converter/parsers/transformer.js +25 -0
- package/src/converter/quantization-info.d.ts +37 -0
- package/src/converter/quantization-info.js +398 -0
- package/src/converter/quantizer.d.ts +96 -0
- package/src/converter/quantizer.js +422 -0
- package/src/converter/rope-config.d.ts +15 -0
- package/src/converter/rope-config.js +218 -0
- package/src/converter/shard-packer.d.ts +138 -0
- package/src/converter/shard-packer.js +422 -0
- package/src/converter/tokenizer-utils.d.ts +11 -0
- package/src/converter/tokenizer-utils.js +87 -0
- package/src/debug/config.d.ts +78 -0
- package/src/debug/config.js +235 -0
- package/src/debug/history.d.ts +65 -0
- package/src/debug/history.js +71 -0
- package/src/debug/index.d.ts +268 -0
- package/src/debug/index.js +192 -0
- package/src/debug/log.d.ts +46 -0
- package/src/debug/log.js +132 -0
- package/src/debug/perf.d.ts +33 -0
- package/src/debug/perf.js +51 -0
- package/src/debug/reference/README.md +114 -0
- package/src/debug/reference/hf_attn_debug.py +114 -0
- package/src/debug/reference/hf_embed_check.py +89 -0
- package/src/debug/reference/hf_layer_out.py +100 -0
- package/src/debug/reference/hf_rope_check.py +116 -0
- package/src/debug/reference/hf_weights.py +75 -0
- package/src/debug/signals.d.ts +63 -0
- package/src/debug/signals.js +33 -0
- package/src/debug/stats.d.ts +47 -0
- package/src/debug/stats.js +160 -0
- package/src/debug/tensor.d.ts +123 -0
- package/src/debug/tensor.js +257 -0
- package/src/debug/trace.d.ts +17 -0
- package/src/debug/trace.js +167 -0
- package/src/diffusion/image-regression.d.ts +31 -0
- package/src/diffusion/image-regression.js +107 -0
- package/src/diffusion/index.d.ts +8 -0
- package/src/diffusion/index.js +8 -0
- package/src/distribution/p2p-control-plane.d.ts +52 -0
- package/src/distribution/p2p-control-plane.js +232 -0
- package/src/distribution/p2p-observability.d.ts +116 -0
- package/src/distribution/p2p-observability.js +267 -0
- package/src/distribution/p2p-transport-contract.d.ts +57 -0
- package/src/distribution/p2p-transport-contract.js +310 -0
- package/src/distribution/p2p-webrtc-browser.d.ts +37 -0
- package/src/distribution/p2p-webrtc-browser.js +434 -0
- package/src/distribution/shard-delivery.d.ts +251 -0
- package/src/distribution/shard-delivery.js +2096 -0
- package/src/energy/index.d.ts +2 -0
- package/src/energy/index.js +2 -0
- package/src/errors/doppler-error.d.ts +21 -0
- package/src/errors/doppler-error.js +25 -0
- package/src/errors/index.d.ts +1 -0
- package/src/errors/index.js +1 -0
- package/src/formats/gguf/index.d.ts +8 -0
- package/src/formats/gguf/index.js +4 -0
- package/src/formats/gguf/types.d.ts +137 -0
- package/src/formats/gguf/types.js +443 -0
- package/src/formats/index.d.ts +51 -0
- package/src/formats/index.js +13 -0
- package/src/formats/rdrr/classification.d.ts +39 -0
- package/src/formats/rdrr/classification.js +275 -0
- package/src/formats/rdrr/groups.d.ts +27 -0
- package/src/formats/rdrr/groups.js +76 -0
- package/src/formats/rdrr/index.d.ts +25 -0
- package/src/formats/rdrr/index.js +19 -0
- package/src/formats/rdrr/manifest.d.ts +32 -0
- package/src/formats/rdrr/manifest.js +108 -0
- package/src/formats/rdrr/parsing.d.ts +23 -0
- package/src/formats/rdrr/parsing.js +101 -0
- package/src/formats/rdrr/tensor-config-validator.d.ts +42 -0
- package/src/formats/rdrr/tensor-config-validator.js +156 -0
- package/src/formats/rdrr/types.d.ts +200 -0
- package/src/formats/rdrr/types.js +16 -0
- package/src/formats/rdrr/validation.d.ts +9 -0
- package/src/formats/rdrr/validation.js +200 -0
- package/src/formats/safetensors/index.d.ts +8 -0
- package/src/formats/safetensors/index.js +4 -0
- package/src/formats/safetensors/types.d.ts +67 -0
- package/src/formats/safetensors/types.js +102 -0
- package/src/formats/tokenizer/index.d.ts +5 -0
- package/src/formats/tokenizer/index.js +3 -0
- package/src/formats/tokenizer/types.d.ts +9 -0
- package/src/formats/tokenizer/types.js +22 -0
- package/src/generation/index.d.ts +18 -0
- package/src/generation/index.js +12 -0
- package/src/gpu/command-recorder.d.ts +175 -0
- package/src/gpu/command-recorder.js +473 -0
- package/src/gpu/device.d.ts +141 -0
- package/src/gpu/device.js +350 -0
- package/src/gpu/kernel-runtime.d.ts +20 -0
- package/src/gpu/kernel-runtime.js +37 -0
- package/src/gpu/kernel-selection-cache.d.ts +13 -0
- package/src/gpu/kernel-selection-cache.js +13 -0
- package/src/gpu/kernel-selection-log.d.ts +12 -0
- package/src/gpu/kernel-selection-log.js +28 -0
- package/src/gpu/kernel-selector.d.ts +11 -0
- package/src/gpu/kernel-selector.js +10 -0
- package/src/gpu/kernel-tuner/benchmarks.d.ts +144 -0
- package/src/gpu/kernel-tuner/benchmarks.js +892 -0
- package/src/gpu/kernel-tuner/cache.d.ts +55 -0
- package/src/gpu/kernel-tuner/cache.js +66 -0
- package/src/gpu/kernel-tuner/index.d.ts +59 -0
- package/src/gpu/kernel-tuner/index.js +38 -0
- package/src/gpu/kernel-tuner/tuner.d.ts +82 -0
- package/src/gpu/kernel-tuner/tuner.js +229 -0
- package/src/gpu/kernel-tuner/types.d.ts +101 -0
- package/src/gpu/kernel-tuner/types.js +4 -0
- package/src/gpu/kernel-tuner.d.ts +33 -0
- package/src/gpu/kernel-tuner.js +12 -0
- package/src/gpu/kernels/README.md +127 -0
- package/src/gpu/kernels/attention.d.ts +236 -0
- package/src/gpu/kernels/attention.js +1359 -0
- package/src/gpu/kernels/attention.wgsl +249 -0
- package/src/gpu/kernels/attention_bdpa_decode_f16.wgsl +246 -0
- package/src/gpu/kernels/attention_decode.wgsl +233 -0
- package/src/gpu/kernels/attention_decode_chunked_f16.wgsl +183 -0
- package/src/gpu/kernels/attention_decode_chunked_f16kv.wgsl +208 -0
- package/src/gpu/kernels/attention_decode_f16.wgsl +202 -0
- package/src/gpu/kernels/attention_decode_f16kv.wgsl +224 -0
- package/src/gpu/kernels/attention_decode_online_f16.wgsl +223 -0
- package/src/gpu/kernels/attention_decode_online_f16kv.wgsl +225 -0
- package/src/gpu/kernels/attention_decode_optimized.wgsl +445 -0
- package/src/gpu/kernels/attention_decode_paged_f16.wgsl +172 -0
- package/src/gpu/kernels/attention_decode_paged_f16kv.wgsl +174 -0
- package/src/gpu/kernels/attention_decode_subgroup.wgsl +233 -0
- package/src/gpu/kernels/attention_decode_tiered_f16.wgsl +218 -0
- package/src/gpu/kernels/attention_decode_tiered_f16kv.wgsl +220 -0
- package/src/gpu/kernels/attention_decode_tiered_int4_f16kv.wgsl +242 -0
- package/src/gpu/kernels/attention_decode_tiered_int8_f16kv.wgsl +242 -0
- package/src/gpu/kernels/attention_f16.wgsl +214 -0
- package/src/gpu/kernels/attention_f16kv.wgsl +242 -0
- package/src/gpu/kernels/attention_small.wgsl +260 -0
- package/src/gpu/kernels/attention_small_f16.wgsl +240 -0
- package/src/gpu/kernels/attention_small_f16kv.wgsl +266 -0
- package/src/gpu/kernels/attention_streaming.wgsl +149 -0
- package/src/gpu/kernels/attention_streaming_f16.wgsl +147 -0
- package/src/gpu/kernels/attention_streaming_f16kv.wgsl +151 -0
- package/src/gpu/kernels/backward/adam.d.ts +28 -0
- package/src/gpu/kernels/backward/adam.js +199 -0
- package/src/gpu/kernels/backward/adam.wgsl +50 -0
- package/src/gpu/kernels/backward/attention_backward.d.ts +22 -0
- package/src/gpu/kernels/backward/attention_backward.js +276 -0
- package/src/gpu/kernels/backward/attention_backward.wgsl +49 -0
- package/src/gpu/kernels/backward/bias_add_backward.d.ts +17 -0
- package/src/gpu/kernels/backward/bias_add_backward.js +24 -0
- package/src/gpu/kernels/backward/bias_add_backward.wgsl +33 -0
- package/src/gpu/kernels/backward/conv2d_backward.d.ts +31 -0
- package/src/gpu/kernels/backward/conv2d_backward.js +135 -0
- package/src/gpu/kernels/backward/conv2d_backward_input.wgsl +83 -0
- package/src/gpu/kernels/backward/conv2d_backward_weight.wgsl +70 -0
- package/src/gpu/kernels/backward/cross_entropy_backward.d.ts +23 -0
- package/src/gpu/kernels/backward/cross_entropy_backward.js +29 -0
- package/src/gpu/kernels/backward/cross_entropy_backward.wgsl +39 -0
- package/src/gpu/kernels/backward/embed_backward.d.ts +29 -0
- package/src/gpu/kernels/backward/embed_backward.js +118 -0
- package/src/gpu/kernels/backward/embed_backward.wgsl +73 -0
- package/src/gpu/kernels/backward/gelu_backward.d.ts +16 -0
- package/src/gpu/kernels/backward/gelu_backward.js +39 -0
- package/src/gpu/kernels/backward/gelu_backward.wgsl +38 -0
- package/src/gpu/kernels/backward/groupnorm_backward.d.ts +24 -0
- package/src/gpu/kernels/backward/groupnorm_backward.js +29 -0
- package/src/gpu/kernels/backward/groupnorm_backward.wgsl +143 -0
- package/src/gpu/kernels/backward/index.d.ts +17 -0
- package/src/gpu/kernels/backward/index.js +23 -0
- package/src/gpu/kernels/backward/layernorm_backward.d.ts +22 -0
- package/src/gpu/kernels/backward/layernorm_backward.js +135 -0
- package/src/gpu/kernels/backward/layernorm_backward.wgsl +194 -0
- package/src/gpu/kernels/backward/matmul_backward.d.ts +32 -0
- package/src/gpu/kernels/backward/matmul_backward.js +124 -0
- package/src/gpu/kernels/backward/matmul_backward.wgsl +90 -0
- package/src/gpu/kernels/backward/matmul_transpose_a.wgsl +84 -0
- package/src/gpu/kernels/backward/pixel_shuffle_backward.d.ts +22 -0
- package/src/gpu/kernels/backward/pixel_shuffle_backward.js +30 -0
- package/src/gpu/kernels/backward/pixel_shuffle_backward.wgsl +54 -0
- package/src/gpu/kernels/backward/rmsnorm_backward.d.ts +24 -0
- package/src/gpu/kernels/backward/rmsnorm_backward.js +101 -0
- package/src/gpu/kernels/backward/rmsnorm_backward.wgsl +78 -0
- package/src/gpu/kernels/backward/rope_backward.d.ts +25 -0
- package/src/gpu/kernels/backward/rope_backward.js +109 -0
- package/src/gpu/kernels/backward/rope_backward.wgsl +59 -0
- package/src/gpu/kernels/backward/scale_backward.d.ts +16 -0
- package/src/gpu/kernels/backward/scale_backward.js +84 -0
- package/src/gpu/kernels/backward/scale_backward.wgsl +27 -0
- package/src/gpu/kernels/backward/silu_backward.d.ts +16 -0
- package/src/gpu/kernels/backward/silu_backward.js +39 -0
- package/src/gpu/kernels/backward/silu_backward.wgsl +31 -0
- package/src/gpu/kernels/backward/softmax_backward.d.ts +16 -0
- package/src/gpu/kernels/backward/softmax_backward.js +43 -0
- package/src/gpu/kernels/backward/softmax_backward.wgsl +44 -0
- package/src/gpu/kernels/backward/upsample2d_backward.d.ts +21 -0
- package/src/gpu/kernels/backward/upsample2d_backward.js +30 -0
- package/src/gpu/kernels/backward/upsample2d_backward.wgsl +59 -0
- package/src/gpu/kernels/backward/utils.d.ts +45 -0
- package/src/gpu/kernels/backward/utils.js +371 -0
- package/src/gpu/kernels/bf16_to_f16.wgsl +54 -0
- package/src/gpu/kernels/bf16_to_f32.wgsl +70 -0
- package/src/gpu/kernels/bias_add.wgsl +40 -0
- package/src/gpu/kernels/bias_add_f16.wgsl +44 -0
- package/src/gpu/kernels/cast.d.ts +67 -0
- package/src/gpu/kernels/cast.js +422 -0
- package/src/gpu/kernels/cast_f16_to_f32.wgsl +31 -0
- package/src/gpu/kernels/cast_f32_to_f16.wgsl +36 -0
- package/src/gpu/kernels/check-finiteness.d.ts +15 -0
- package/src/gpu/kernels/check-finiteness.js +149 -0
- package/src/gpu/kernels/check-stop.d.ts +31 -0
- package/src/gpu/kernels/check-stop.js +181 -0
- package/src/gpu/kernels/clamp.d.ts +22 -0
- package/src/gpu/kernels/clamp.js +42 -0
- package/src/gpu/kernels/clamp.wgsl +24 -0
- package/src/gpu/kernels/constants.d.ts +168 -0
- package/src/gpu/kernels/constants.js +129 -0
- package/src/gpu/kernels/conv2d.d.ts +34 -0
- package/src/gpu/kernels/conv2d.js +81 -0
- package/src/gpu/kernels/conv2d.wgsl +71 -0
- package/src/gpu/kernels/conv2d_f16.wgsl +73 -0
- package/src/gpu/kernels/cross_entropy_loss.d.ts +21 -0
- package/src/gpu/kernels/cross_entropy_loss.js +54 -0
- package/src/gpu/kernels/cross_entropy_loss.wgsl +39 -0
- package/src/gpu/kernels/dequant.d.ts +108 -0
- package/src/gpu/kernels/dequant.js +524 -0
- package/src/gpu/kernels/dequant_f16_out.wgsl +151 -0
- package/src/gpu/kernels/dequant_f16_out_vec4.wgsl +149 -0
- package/src/gpu/kernels/dequant_f16_rowwise.wgsl +139 -0
- package/src/gpu/kernels/dequant_f32_rowwise.wgsl +133 -0
- package/src/gpu/kernels/dequant_mxfp4.wgsl +120 -0
- package/src/gpu/kernels/dequant_mxfp4_expert.wgsl +129 -0
- package/src/gpu/kernels/dequant_mxfp4_expert_f16.wgsl +105 -0
- package/src/gpu/kernels/dequant_mxfp4_vec4.wgsl +116 -0
- package/src/gpu/kernels/dequant_q6k.wgsl +140 -0
- package/src/gpu/kernels/dequant_q8_0.wgsl +98 -0
- package/src/gpu/kernels/dequant_shared.wgsl +202 -0
- package/src/gpu/kernels/dequant_shared_vec4.wgsl +153 -0
- package/src/gpu/kernels/dequant_subgroup.wgsl +202 -0
- package/src/gpu/kernels/dispatch.d.ts +157 -0
- package/src/gpu/kernels/dispatch.js +235 -0
- package/src/gpu/kernels/energy.d.ts +131 -0
- package/src/gpu/kernels/energy.js +425 -0
- package/src/gpu/kernels/energy_eval.wgsl +26 -0
- package/src/gpu/kernels/energy_eval_f16.wgsl +30 -0
- package/src/gpu/kernels/energy_quintel_grad.wgsl +92 -0
- package/src/gpu/kernels/energy_quintel_grad_f16.wgsl +96 -0
- package/src/gpu/kernels/energy_quintel_reduce.wgsl +112 -0
- package/src/gpu/kernels/energy_quintel_reduce_f16.wgsl +116 -0
- package/src/gpu/kernels/energy_quintel_update.wgsl +92 -0
- package/src/gpu/kernels/energy_quintel_update_f16.wgsl +96 -0
- package/src/gpu/kernels/energy_update.wgsl +25 -0
- package/src/gpu/kernels/energy_update_f16.wgsl +30 -0
- package/src/gpu/kernels/feature-check.d.ts +42 -0
- package/src/gpu/kernels/feature-check.js +70 -0
- package/src/gpu/kernels/fused_ffn.d.ts +65 -0
- package/src/gpu/kernels/fused_ffn.js +318 -0
- package/src/gpu/kernels/fused_ffn.wgsl +420 -0
- package/src/gpu/kernels/fused_ffn_f16.wgsl +213 -0
- package/src/gpu/kernels/fused_ffn_q4k.wgsl +375 -0
- package/src/gpu/kernels/fused_matmul_q4.wgsl +404 -0
- package/src/gpu/kernels/fused_matmul_q4_batched.wgsl +194 -0
- package/src/gpu/kernels/fused_matmul_q4_batched_f16.wgsl +170 -0
- package/src/gpu/kernels/fused_matmul_q4_batched_f16a.wgsl +154 -0
- package/src/gpu/kernels/fused_matmul_q4_f16a.wgsl +219 -0
- package/src/gpu/kernels/fused_matmul_q4_multicol_f16.wgsl +216 -0
- package/src/gpu/kernels/fused_matmul_q4_multicol_f16a.wgsl +204 -0
- package/src/gpu/kernels/fused_matmul_residual.d.ts +46 -0
- package/src/gpu/kernels/fused_matmul_residual.js +152 -0
- package/src/gpu/kernels/fused_matmul_rmsnorm.d.ts +64 -0
- package/src/gpu/kernels/fused_matmul_rmsnorm.js +273 -0
- package/src/gpu/kernels/fused_matmul_rmsnorm.wgsl +324 -0
- package/src/gpu/kernels/fused_matmul_rmsnorm_f16.wgsl +303 -0
- package/src/gpu/kernels/fused_swiglu.wgsl +63 -0
- package/src/gpu/kernels/fused_swiglu_f16.wgsl +57 -0
- package/src/gpu/kernels/gather.d.ts +64 -0
- package/src/gpu/kernels/gather.js +119 -0
- package/src/gpu/kernels/gather.wgsl +61 -0
- package/src/gpu/kernels/gather_f16.wgsl +65 -0
- package/src/gpu/kernels/gather_f16_f16_out.wgsl +55 -0
- package/src/gpu/kernels/gather_f16_out.wgsl +55 -0
- package/src/gpu/kernels/gather_f16_vec4.wgsl +76 -0
- package/src/gpu/kernels/gather_f16_vec4_f16_out.wgsl +68 -0
- package/src/gpu/kernels/gather_vec4.wgsl +74 -0
- package/src/gpu/kernels/gather_vec4_f16_out.wgsl +68 -0
- package/src/gpu/kernels/gelu.d.ts +33 -0
- package/src/gpu/kernels/gelu.js +47 -0
- package/src/gpu/kernels/gelu.wgsl +64 -0
- package/src/gpu/kernels/gelu_f16.wgsl +66 -0
- package/src/gpu/kernels/gptoss_mxfp4_expert_fused.wgsl +127 -0
- package/src/gpu/kernels/gptoss_router_topk.wgsl +119 -0
- package/src/gpu/kernels/groupnorm.d.ts +31 -0
- package/src/gpu/kernels/groupnorm.js +91 -0
- package/src/gpu/kernels/groupnorm_apply.wgsl +41 -0
- package/src/gpu/kernels/groupnorm_apply_f16.wgsl +46 -0
- package/src/gpu/kernels/groupnorm_stats.wgsl +76 -0
- package/src/gpu/kernels/groupnorm_stats_f16.wgsl +79 -0
- package/src/gpu/kernels/index.d.ts +336 -0
- package/src/gpu/kernels/index.js +284 -0
- package/src/gpu/kernels/kernel-base.d.ts +33 -0
- package/src/gpu/kernels/kernel-base.js +46 -0
- package/src/gpu/kernels/kernel-configs.d.ts +65 -0
- package/src/gpu/kernels/kernel-configs.js +50 -0
- package/src/gpu/kernels/kernel-tuning.d.ts +42 -0
- package/src/gpu/kernels/kernel-tuning.js +149 -0
- package/src/gpu/kernels/kv-quantize.d.ts +37 -0
- package/src/gpu/kernels/kv-quantize.js +138 -0
- package/src/gpu/kernels/kv_quantize_int4.wgsl +119 -0
- package/src/gpu/kernels/kv_quantize_int8.wgsl +119 -0
- package/src/gpu/kernels/layernorm.d.ts +37 -0
- package/src/gpu/kernels/layernorm.js +80 -0
- package/src/gpu/kernels/layernorm.wgsl +121 -0
- package/src/gpu/kernels/layernorm_f16.wgsl +103 -0
- package/src/gpu/kernels/linear-attention-core.d.ts +39 -0
- package/src/gpu/kernels/linear-attention-core.js +535 -0
- package/src/gpu/kernels/logit-merge.d.ts +110 -0
- package/src/gpu/kernels/logit-merge.js +392 -0
- package/src/gpu/kernels/matmul-dispatch.d.ts +38 -0
- package/src/gpu/kernels/matmul-dispatch.js +155 -0
- package/src/gpu/kernels/matmul-selection.d.ts +87 -0
- package/src/gpu/kernels/matmul-selection.js +474 -0
- package/src/gpu/kernels/matmul.d.ts +109 -0
- package/src/gpu/kernels/matmul.js +271 -0
- package/src/gpu/kernels/matmul_f16.wgsl +170 -0
- package/src/gpu/kernels/matmul_f16_tiled.wgsl +165 -0
- package/src/gpu/kernels/matmul_f16w_f32a.wgsl +89 -0
- package/src/gpu/kernels/matmul_f16w_f32a_tiled.wgsl +154 -0
- package/src/gpu/kernels/matmul_f32.wgsl +100 -0
- package/src/gpu/kernels/matmul_gemv.wgsl +80 -0
- package/src/gpu/kernels/matmul_gemv_f16a.wgsl +81 -0
- package/src/gpu/kernels/matmul_gemv_residual.wgsl +119 -0
- package/src/gpu/kernels/matmul_gemv_residual_f16.wgsl +78 -0
- package/src/gpu/kernels/matmul_gemv_subgroup.wgsl +345 -0
- package/src/gpu/kernels/matmul_gemv_subgroup_f16a.wgsl +514 -0
- package/src/gpu/kernels/modulate.d.ts +29 -0
- package/src/gpu/kernels/modulate.js +49 -0
- package/src/gpu/kernels/modulate.wgsl +40 -0
- package/src/gpu/kernels/modulate_f16.wgsl +43 -0
- package/src/gpu/kernels/moe.d.ts +164 -0
- package/src/gpu/kernels/moe.js +496 -0
- package/src/gpu/kernels/moe_gather.wgsl +170 -0
- package/src/gpu/kernels/moe_gather_f16.wgsl +82 -0
- package/src/gpu/kernels/moe_gather_vec4.wgsl +74 -0
- package/src/gpu/kernels/moe_offsets.wgsl +48 -0
- package/src/gpu/kernels/pipeline-cache.d.ts +88 -0
- package/src/gpu/kernels/pipeline-cache.js +305 -0
- package/src/gpu/kernels/pixel_shuffle.d.ts +27 -0
- package/src/gpu/kernels/pixel_shuffle.js +49 -0
- package/src/gpu/kernels/pixel_shuffle.wgsl +44 -0
- package/src/gpu/kernels/pixel_shuffle_f16.wgsl +47 -0
- package/src/gpu/kernels/residual.d.ts +74 -0
- package/src/gpu/kernels/residual.js +127 -0
- package/src/gpu/kernels/residual.wgsl +53 -0
- package/src/gpu/kernels/residual_f16.wgsl +35 -0
- package/src/gpu/kernels/residual_f16_vec4.wgsl +47 -0
- package/src/gpu/kernels/residual_vec4.wgsl +46 -0
- package/src/gpu/kernels/rmsnorm.d.ts +53 -0
- package/src/gpu/kernels/rmsnorm.js +140 -0
- package/src/gpu/kernels/rmsnorm.wgsl +417 -0
- package/src/gpu/kernels/rmsnorm_f16.wgsl +164 -0
- package/src/gpu/kernels/rope.d.ts +48 -0
- package/src/gpu/kernels/rope.js +53 -0
- package/src/gpu/kernels/rope.wgsl +328 -0
- package/src/gpu/kernels/rope_f16.wgsl +271 -0
- package/src/gpu/kernels/rule-matcher.d.ts +30 -0
- package/src/gpu/kernels/rule-matcher.js +42 -0
- package/src/gpu/kernels/rule-registry.d.ts +7 -0
- package/src/gpu/kernels/rule-registry.js +41 -0
- package/src/gpu/kernels/sample.d.ts +75 -0
- package/src/gpu/kernels/sample.js +578 -0
- package/src/gpu/kernels/sample.wgsl +377 -0
- package/src/gpu/kernels/sample_f16.wgsl +331 -0
- package/src/gpu/kernels/scale.d.ts +35 -0
- package/src/gpu/kernels/scale.js +37 -0
- package/src/gpu/kernels/scale.wgsl +38 -0
- package/src/gpu/kernels/scatter_add.wgsl +88 -0
- package/src/gpu/kernels/scatter_add_dynamic.wgsl +59 -0
- package/src/gpu/kernels/scatter_add_dynamic_f16.wgsl +52 -0
- package/src/gpu/kernels/scatter_add_dynamic_f16_weights.wgsl +50 -0
- package/src/gpu/kernels/scatter_add_vec4.wgsl +70 -0
- package/src/gpu/kernels/shader-cache.d.ts +56 -0
- package/src/gpu/kernels/shader-cache.js +206 -0
- package/src/gpu/kernels/silu.d.ts +75 -0
- package/src/gpu/kernels/silu.js +340 -0
- package/src/gpu/kernels/silu.wgsl +99 -0
- package/src/gpu/kernels/silu_f16.wgsl +98 -0
- package/src/gpu/kernels/softmax.d.ts +57 -0
- package/src/gpu/kernels/softmax.js +106 -0
- package/src/gpu/kernels/softmax.wgsl +388 -0
- package/src/gpu/kernels/softmax_subgroup.wgsl +175 -0
- package/src/gpu/kernels/split_qkv.d.ts +51 -0
- package/src/gpu/kernels/split_qkv.js +41 -0
- package/src/gpu/kernels/split_qkv.wgsl +71 -0
- package/src/gpu/kernels/split_qkv_f16.wgsl +75 -0
- package/src/gpu/kernels/topk.wgsl +243 -0
- package/src/gpu/kernels/topk_f16.wgsl +108 -0
- package/src/gpu/kernels/topk_f16_weights.wgsl +101 -0
- package/src/gpu/kernels/transpose.d.ts +21 -0
- package/src/gpu/kernels/transpose.js +30 -0
- package/src/gpu/kernels/transpose.wgsl +32 -0
- package/src/gpu/kernels/types.d.ts +21 -0
- package/src/gpu/kernels/types.js +4 -0
- package/src/gpu/kernels/uniform-utils.d.ts +48 -0
- package/src/gpu/kernels/uniform-utils.js +94 -0
- package/src/gpu/kernels/upsample2d.d.ts +25 -0
- package/src/gpu/kernels/upsample2d.js +58 -0
- package/src/gpu/kernels/upsample2d.wgsl +37 -0
- package/src/gpu/kernels/upsample2d_f16.wgsl +41 -0
- package/src/gpu/kernels/utils.d.ts +106 -0
- package/src/gpu/kernels/utils.js +224 -0
- package/src/gpu/multi-model-recorder.d.ts +21 -0
- package/src/gpu/multi-model-recorder.js +31 -0
- package/src/gpu/partitioned-buffer-pool.d.ts +28 -0
- package/src/gpu/partitioned-buffer-pool.js +49 -0
- package/src/gpu/perf-guards.d.ts +25 -0
- package/src/gpu/perf-guards.js +140 -0
- package/src/gpu/profiler.d.ts +114 -0
- package/src/gpu/profiler.js +391 -0
- package/src/gpu/submit-tracker.d.ts +111 -0
- package/src/gpu/submit-tracker.js +229 -0
- package/src/gpu/tensor.d.ts +69 -0
- package/src/gpu/tensor.js +75 -0
- package/src/gpu/uniform-cache.d.ts +108 -0
- package/src/gpu/uniform-cache.js +242 -0
- package/src/gpu/weight-buffer.d.ts +115 -0
- package/src/gpu/weight-buffer.js +118 -0
- package/src/hotswap/intent-bundle.d.ts +37 -0
- package/src/hotswap/intent-bundle.js +123 -0
- package/src/hotswap/manifest.d.ts +33 -0
- package/src/hotswap/manifest.js +114 -0
- package/src/hotswap/runtime.d.ts +31 -0
- package/src/hotswap/runtime.js +128 -0
- package/src/index-browser.d.ts +47 -0
- package/src/index-browser.js +53 -0
- package/src/index-internal.d.ts +2 -0
- package/src/index-internal.js +2 -0
- package/src/index.d.ts +102 -0
- package/src/index.js +75 -0
- package/src/inference/README.md +593 -0
- package/src/inference/browser-harness.d.ts +234 -0
- package/src/inference/browser-harness.js +2665 -0
- package/src/inference/decode-buffers.d.ts +108 -0
- package/src/inference/decode-buffers.js +181 -0
- package/src/inference/decode-ring.d.ts +52 -0
- package/src/inference/decode-ring.js +273 -0
- package/src/inference/expert-router.d.ts +27 -0
- package/src/inference/expert-router.js +55 -0
- package/src/inference/functiongemma.d.ts +15 -0
- package/src/inference/functiongemma.js +1 -0
- package/src/inference/kv-cache/base.d.ts +150 -0
- package/src/inference/kv-cache/base.js +1037 -0
- package/src/inference/kv-cache/basis-decomposed-paged.d.ts +50 -0
- package/src/inference/kv-cache/basis-decomposed-paged.js +276 -0
- package/src/inference/kv-cache/index.d.ts +35 -0
- package/src/inference/kv-cache/index.js +20 -0
- package/src/inference/kv-cache/sliding-window.d.ts +72 -0
- package/src/inference/kv-cache/sliding-window.js +243 -0
- package/src/inference/kv-cache/tiered.d.ts +89 -0
- package/src/inference/kv-cache/tiered.js +574 -0
- package/src/inference/kv-cache/types.d.ts +188 -0
- package/src/inference/kv-cache/types.js +80 -0
- package/src/inference/kv-cache.d.ts +36 -0
- package/src/inference/kv-cache.js +18 -0
- package/src/inference/moe-router.d.ts +212 -0
- package/src/inference/moe-router.js +553 -0
- package/src/inference/multi-model-network.d.ts +139 -0
- package/src/inference/multi-model-network.js +769 -0
- package/src/inference/multi-pipeline-pool.d.ts +62 -0
- package/src/inference/multi-pipeline-pool.js +161 -0
- package/src/inference/network-evolution.d.ts +46 -0
- package/src/inference/network-evolution.js +80 -0
- package/src/inference/pipelines/context.d.ts +18 -0
- package/src/inference/pipelines/context.js +44 -0
- package/src/inference/pipelines/diffusion/helpers.d.ts +29 -0
- package/src/inference/pipelines/diffusion/helpers.js +112 -0
- package/src/inference/pipelines/diffusion/index.d.ts +3 -0
- package/src/inference/pipelines/diffusion/index.js +3 -0
- package/src/inference/pipelines/diffusion/init.d.ts +24 -0
- package/src/inference/pipelines/diffusion/init.js +124 -0
- package/src/inference/pipelines/diffusion/pipeline.d.ts +38 -0
- package/src/inference/pipelines/diffusion/pipeline.js +632 -0
- package/src/inference/pipelines/diffusion/scheduler.d.ts +19 -0
- package/src/inference/pipelines/diffusion/scheduler.js +65 -0
- package/src/inference/pipelines/diffusion/sd3-transformer.d.ts +20 -0
- package/src/inference/pipelines/diffusion/sd3-transformer.js +1194 -0
- package/src/inference/pipelines/diffusion/sd3-weights.d.ts +21 -0
- package/src/inference/pipelines/diffusion/sd3-weights.js +287 -0
- package/src/inference/pipelines/diffusion/text-encoder-gpu.d.ts +80 -0
- package/src/inference/pipelines/diffusion/text-encoder-gpu.js +935 -0
- package/src/inference/pipelines/diffusion/text-encoder.d.ts +29 -0
- package/src/inference/pipelines/diffusion/text-encoder.js +178 -0
- package/src/inference/pipelines/diffusion/types.d.ts +112 -0
- package/src/inference/pipelines/diffusion/types.js +1 -0
- package/src/inference/pipelines/diffusion/vae.d.ts +20 -0
- package/src/inference/pipelines/diffusion/vae.js +675 -0
- package/src/inference/pipelines/diffusion/weights.d.ts +40 -0
- package/src/inference/pipelines/diffusion/weights.js +150 -0
- package/src/inference/pipelines/dream/energy-head-pipeline.d.ts +29 -0
- package/src/inference/pipelines/dream/energy-head-pipeline.js +6 -0
- package/src/inference/pipelines/dream/pipeline.d.ts +17 -0
- package/src/inference/pipelines/dream/pipeline.js +8 -0
- package/src/inference/pipelines/energy/index.d.ts +1 -0
- package/src/inference/pipelines/energy/index.js +1 -0
- package/src/inference/pipelines/energy/pipeline.d.ts +27 -0
- package/src/inference/pipelines/energy/pipeline.js +680 -0
- package/src/inference/pipelines/energy/quintel.d.ts +87 -0
- package/src/inference/pipelines/energy/quintel.js +207 -0
- package/src/inference/pipelines/energy/types.d.ts +63 -0
- package/src/inference/pipelines/energy/types.js +1 -0
- package/src/inference/pipelines/energy-head/index.d.ts +6 -0
- package/src/inference/pipelines/energy-head/index.js +6 -0
- package/src/inference/pipelines/energy-head/row-head-pipeline.d.ts +103 -0
- package/src/inference/pipelines/energy-head/row-head-pipeline.js +487 -0
- package/src/inference/pipelines/factory.d.ts +10 -0
- package/src/inference/pipelines/factory.js +6 -0
- package/src/inference/pipelines/index.d.ts +22 -0
- package/src/inference/pipelines/index.js +19 -0
- package/src/inference/pipelines/registry.d.ts +15 -0
- package/src/inference/pipelines/registry.js +23 -0
- package/src/inference/pipelines/rng.d.ts +2 -0
- package/src/inference/pipelines/rng.js +17 -0
- package/src/inference/pipelines/structured/index.d.ts +8 -0
- package/src/inference/pipelines/structured/index.js +8 -0
- package/src/inference/pipelines/structured/json-head-pipeline.d.ts +58 -0
- package/src/inference/pipelines/structured/json-head-pipeline.js +181 -0
- package/src/inference/pipelines/text/attention/index.d.ts +24 -0
- package/src/inference/pipelines/text/attention/index.js +17 -0
- package/src/inference/pipelines/text/attention/projections.d.ts +101 -0
- package/src/inference/pipelines/text/attention/projections.js +435 -0
- package/src/inference/pipelines/text/attention/record.d.ts +36 -0
- package/src/inference/pipelines/text/attention/record.js +613 -0
- package/src/inference/pipelines/text/attention/run.d.ts +38 -0
- package/src/inference/pipelines/text/attention/run.js +826 -0
- package/src/inference/pipelines/text/attention/types.d.ts +98 -0
- package/src/inference/pipelines/text/attention/types.js +67 -0
- package/src/inference/pipelines/text/attention.d.ts +23 -0
- package/src/inference/pipelines/text/attention.js +12 -0
- package/src/inference/pipelines/text/bdpa-steamroller.d.ts +22 -0
- package/src/inference/pipelines/text/bdpa-steamroller.js +158 -0
- package/src/inference/pipelines/text/buffer-types.d.ts +7 -0
- package/src/inference/pipelines/text/buffer-types.js +4 -0
- package/src/inference/pipelines/text/chat-format.d.ts +46 -0
- package/src/inference/pipelines/text/chat-format.js +366 -0
- package/src/inference/pipelines/text/config.d.ts +235 -0
- package/src/inference/pipelines/text/config.js +623 -0
- package/src/inference/pipelines/text/debug-utils/config.d.ts +144 -0
- package/src/inference/pipelines/text/debug-utils/config.js +156 -0
- package/src/inference/pipelines/text/debug-utils/index.d.ts +53 -0
- package/src/inference/pipelines/text/debug-utils/index.js +44 -0
- package/src/inference/pipelines/text/debug-utils/logging.d.ts +106 -0
- package/src/inference/pipelines/text/debug-utils/logging.js +152 -0
- package/src/inference/pipelines/text/debug-utils/tensor.d.ts +119 -0
- package/src/inference/pipelines/text/debug-utils/tensor.js +268 -0
- package/src/inference/pipelines/text/debug-utils/utils.d.ts +77 -0
- package/src/inference/pipelines/text/debug-utils/utils.js +139 -0
- package/src/inference/pipelines/text/debug-utils.d.ts +42 -0
- package/src/inference/pipelines/text/debug-utils.js +34 -0
- package/src/inference/pipelines/text/embed.d.ts +67 -0
- package/src/inference/pipelines/text/embed.js +461 -0
- package/src/inference/pipelines/text/execution-plan.d.ts +116 -0
- package/src/inference/pipelines/text/execution-plan.js +314 -0
- package/src/inference/pipelines/text/execution-v0.d.ts +66 -0
- package/src/inference/pipelines/text/execution-v0.js +1139 -0
- package/src/inference/pipelines/text/ffn/dense.d.ts +40 -0
- package/src/inference/pipelines/text/ffn/dense.js +759 -0
- package/src/inference/pipelines/text/ffn/index.d.ts +23 -0
- package/src/inference/pipelines/text/ffn/index.js +16 -0
- package/src/inference/pipelines/text/ffn/moe.d.ts +21 -0
- package/src/inference/pipelines/text/ffn/moe.js +49 -0
- package/src/inference/pipelines/text/ffn/sandwich.d.ts +25 -0
- package/src/inference/pipelines/text/ffn/sandwich.js +196 -0
- package/src/inference/pipelines/text/ffn/standard.d.ts +23 -0
- package/src/inference/pipelines/text/ffn/standard.js +84 -0
- package/src/inference/pipelines/text/ffn/types.d.ts +30 -0
- package/src/inference/pipelines/text/ffn/types.js +25 -0
- package/src/inference/pipelines/text/ffn.d.ts +31 -0
- package/src/inference/pipelines/text/ffn.js +18 -0
- package/src/inference/pipelines/text/finiteness-guard-status.d.ts +11 -0
- package/src/inference/pipelines/text/finiteness-guard-status.js +21 -0
- package/src/inference/pipelines/text/finiteness-policy.d.ts +35 -0
- package/src/inference/pipelines/text/finiteness-policy.js +45 -0
- package/src/inference/pipelines/text/generator-helpers.d.ts +34 -0
- package/src/inference/pipelines/text/generator-helpers.js +175 -0
- package/src/inference/pipelines/text/generator-runtime.d.ts +93 -0
- package/src/inference/pipelines/text/generator-runtime.js +373 -0
- package/src/inference/pipelines/text/generator-steps.d.ts +75 -0
- package/src/inference/pipelines/text/generator-steps.js +1078 -0
- package/src/inference/pipelines/text/generator.d.ts +41 -0
- package/src/inference/pipelines/text/generator.js +1345 -0
- package/src/inference/pipelines/text/index.d.ts +5 -0
- package/src/inference/pipelines/text/index.js +6 -0
- package/src/inference/pipelines/text/init.d.ts +295 -0
- package/src/inference/pipelines/text/init.js +965 -0
- package/src/inference/pipelines/text/kernel-path-auto-select.d.ts +12 -0
- package/src/inference/pipelines/text/kernel-path-auto-select.js +90 -0
- package/src/inference/pipelines/text/kernel-trace.d.ts +150 -0
- package/src/inference/pipelines/text/kernel-trace.js +324 -0
- package/src/inference/pipelines/text/layer-plan.d.ts +65 -0
- package/src/inference/pipelines/text/layer-plan.js +249 -0
- package/src/inference/pipelines/text/layer.d.ts +56 -0
- package/src/inference/pipelines/text/layer.js +916 -0
- package/src/inference/pipelines/text/linear-attention.d.ts +94 -0
- package/src/inference/pipelines/text/linear-attention.js +803 -0
- package/src/inference/pipelines/text/logits/cpu.d.ts +81 -0
- package/src/inference/pipelines/text/logits/cpu.js +91 -0
- package/src/inference/pipelines/text/logits/gpu.d.ts +113 -0
- package/src/inference/pipelines/text/logits/gpu.js +406 -0
- package/src/inference/pipelines/text/logits/index.d.ts +57 -0
- package/src/inference/pipelines/text/logits/index.js +305 -0
- package/src/inference/pipelines/text/logits/types.d.ts +46 -0
- package/src/inference/pipelines/text/logits/types.js +4 -0
- package/src/inference/pipelines/text/logits/utils.d.ts +49 -0
- package/src/inference/pipelines/text/logits/utils.js +59 -0
- package/src/inference/pipelines/text/logits.d.ts +27 -0
- package/src/inference/pipelines/text/logits.js +16 -0
- package/src/inference/pipelines/text/lora-apply.d.ts +28 -0
- package/src/inference/pipelines/text/lora-apply.js +58 -0
- package/src/inference/pipelines/text/lora-types.d.ts +39 -0
- package/src/inference/pipelines/text/lora-types.js +18 -0
- package/src/inference/pipelines/text/lora.d.ts +18 -0
- package/src/inference/pipelines/text/lora.js +12 -0
- package/src/inference/pipelines/text/model-load.d.ts +58 -0
- package/src/inference/pipelines/text/model-load.js +561 -0
- package/src/inference/pipelines/text/moe-cache.d.ts +32 -0
- package/src/inference/pipelines/text/moe-cache.js +107 -0
- package/src/inference/pipelines/text/moe-cpu-gptoss.d.ts +9 -0
- package/src/inference/pipelines/text/moe-cpu-gptoss.js +110 -0
- package/src/inference/pipelines/text/moe-cpu.d.ts +13 -0
- package/src/inference/pipelines/text/moe-cpu.js +116 -0
- package/src/inference/pipelines/text/moe-gpu.d.ts +13 -0
- package/src/inference/pipelines/text/moe-gpu.js +611 -0
- package/src/inference/pipelines/text/moe-helpers.d.ts +12 -0
- package/src/inference/pipelines/text/moe-helpers.js +21 -0
- package/src/inference/pipelines/text/moe-impl.d.ts +117 -0
- package/src/inference/pipelines/text/moe-impl.js +9 -0
- package/src/inference/pipelines/text/moe-shape-validator.d.ts +31 -0
- package/src/inference/pipelines/text/moe-shape-validator.js +78 -0
- package/src/inference/pipelines/text/ops.d.ts +167 -0
- package/src/inference/pipelines/text/ops.js +367 -0
- package/src/inference/pipelines/text/probes.d.ts +31 -0
- package/src/inference/pipelines/text/probes.js +170 -0
- package/src/inference/pipelines/text/sampling.d.ts +54 -0
- package/src/inference/pipelines/text/sampling.js +203 -0
- package/src/inference/pipelines/text/state.d.ts +112 -0
- package/src/inference/pipelines/text/state.js +152 -0
- package/src/inference/pipelines/text/types.d.ts +627 -0
- package/src/inference/pipelines/text/types.js +4 -0
- package/src/inference/pipelines/text/weights.d.ts +110 -0
- package/src/inference/pipelines/text/weights.js +163 -0
- package/src/inference/pipelines/text.d.ts +157 -0
- package/src/inference/pipelines/text.js +586 -0
- package/src/inference/speculative.d.ts +239 -0
- package/src/inference/speculative.js +416 -0
- package/src/inference/test-harness.d.ts +178 -0
- package/src/inference/test-harness.js +349 -0
- package/src/inference/tokenizer.d.ts +77 -0
- package/src/inference/tokenizer.js +258 -0
- package/src/inference/tokenizers/base.d.ts +39 -0
- package/src/inference/tokenizers/base.js +69 -0
- package/src/inference/tokenizers/bpe.d.ts +27 -0
- package/src/inference/tokenizers/bpe.js +171 -0
- package/src/inference/tokenizers/bundled.d.ts +63 -0
- package/src/inference/tokenizers/bundled.js +866 -0
- package/src/inference/tokenizers/sentencepiece.d.ts +28 -0
- package/src/inference/tokenizers/sentencepiece.js +389 -0
- package/src/inference/tokenizers/types.d.ts +166 -0
- package/src/inference/tokenizers/types.js +7 -0
- package/src/loader/doppler-loader.d.ts +134 -0
- package/src/loader/doppler-loader.js +1036 -0
- package/src/loader/dtype-utils.d.ts +40 -0
- package/src/loader/dtype-utils.js +102 -0
- package/src/loader/embedding-loader.d.ts +56 -0
- package/src/loader/embedding-loader.js +207 -0
- package/src/loader/experts/expert-cache.d.ts +156 -0
- package/src/loader/experts/expert-cache.js +375 -0
- package/src/loader/experts/expert-loader.d.ts +108 -0
- package/src/loader/experts/expert-loader.js +384 -0
- package/src/loader/final-weights-loader.d.ts +68 -0
- package/src/loader/final-weights-loader.js +262 -0
- package/src/loader/index.d.ts +150 -0
- package/src/loader/index.js +124 -0
- package/src/loader/layer-loader.d.ts +63 -0
- package/src/loader/layer-loader.js +417 -0
- package/src/loader/loader-state.d.ts +51 -0
- package/src/loader/loader-state.js +142 -0
- package/src/loader/loader-types.d.ts +236 -0
- package/src/loader/loader-types.js +4 -0
- package/src/loader/manifest-config.d.ts +97 -0
- package/src/loader/manifest-config.js +132 -0
- package/src/loader/memory-monitor.d.ts +112 -0
- package/src/loader/memory-monitor.js +276 -0
- package/src/loader/multi-model-loader.d.ts +37 -0
- package/src/loader/multi-model-loader.js +87 -0
- package/src/loader/quantization-constants.d.ts +23 -0
- package/src/loader/quantization-constants.js +14 -0
- package/src/loader/shard-cache.d.ts +60 -0
- package/src/loader/shard-cache.js +568 -0
- package/src/loader/shard-resolver.d.ts +12 -0
- package/src/loader/shard-resolver.js +83 -0
- package/src/loader/tensors/tensor-loader.d.ts +154 -0
- package/src/loader/tensors/tensor-loader.js +427 -0
- package/src/loader/tensors/tensor-reader.d.ts +22 -0
- package/src/loader/tensors/tensor-reader.js +56 -0
- package/src/loader/tensors/tensor-role.d.ts +7 -0
- package/src/loader/tensors/tensor-role.js +12 -0
- package/src/loader/weight-downcast.d.ts +62 -0
- package/src/loader/weight-downcast.js +213 -0
- package/src/loader/weights.d.ts +22 -0
- package/src/loader/weights.js +4 -0
- package/src/memory/address-table.d.ts +104 -0
- package/src/memory/address-table.js +114 -0
- package/src/memory/buffer-pool.d.ts +196 -0
- package/src/memory/buffer-pool.js +756 -0
- package/src/memory/capability.d.ts +49 -0
- package/src/memory/capability.js +95 -0
- package/src/memory/heap-manager.d.ts +104 -0
- package/src/memory/heap-manager.js +264 -0
- package/src/memory/unified-detect.d.ts +59 -0
- package/src/memory/unified-detect.js +192 -0
- package/src/rules/converter/execution.rules.json +20 -0
- package/src/rules/converter/tensor-roles.rules.json +13 -0
- package/src/rules/converter/tokenizer.rules.json +7 -0
- package/src/rules/inference/attention.rules.json +54 -0
- package/src/rules/inference/config.rules.json +58 -0
- package/src/rules/inference/dtype.rules.json +94 -0
- package/src/rules/inference/execution.rules.json +45 -0
- package/src/rules/inference/ffn.rules.json +35 -0
- package/src/rules/inference/kernel-path.rules.json +76 -0
- package/src/rules/inference/layer-pattern.rules.json +16 -0
- package/src/rules/inference/layer.rules.json +7 -0
- package/src/rules/inference/moe.rules.json +48 -0
- package/src/rules/kernels/attention.rules.json +61 -0
- package/src/rules/kernels/conv2d.rules.json +6 -0
- package/src/rules/kernels/dequant.rules.json +58 -0
- package/src/rules/kernels/energy.rules.json +22 -0
- package/src/rules/kernels/fused-ffn.rules.json +13 -0
- package/src/rules/kernels/fused-matmul-residual.rules.json +6 -0
- package/src/rules/kernels/fused-matmul-rmsnorm.rules.json +8 -0
- package/src/rules/kernels/gather.rules.json +12 -0
- package/src/rules/kernels/gelu.rules.json +11 -0
- package/src/rules/kernels/groupnorm.rules.json +10 -0
- package/src/rules/kernels/kernel-validator.d.ts +24 -0
- package/src/rules/kernels/kernel-validator.js +160 -0
- package/src/rules/kernels/kv_quantize.rules.json +7 -0
- package/src/rules/kernels/layernorm.rules.json +6 -0
- package/src/rules/kernels/matmul.rules.json +60 -0
- package/src/rules/kernels/modulate.rules.json +6 -0
- package/src/rules/kernels/moe.rules.gptoss.json +105 -0
- package/src/rules/kernels/moe.rules.json +11 -0
- package/src/rules/kernels/pixel_shuffle.rules.json +6 -0
- package/src/rules/kernels/residual.rules.json +12 -0
- package/src/rules/kernels/rmsnorm.rules.json +11 -0
- package/src/rules/kernels/rope.rules.json +6 -0
- package/src/rules/kernels/sample.rules.json +6 -0
- package/src/rules/kernels/scale.rules.json +6 -0
- package/src/rules/kernels/silu.rules.json +21 -0
- package/src/rules/kernels/softmax.rules.json +23 -0
- package/src/rules/kernels/split-qkv.rules.json +6 -0
- package/src/rules/kernels/upsample2d.rules.json +6 -0
- package/src/rules/loader/tensor-loader.rules.json +15 -0
- package/src/rules/loader/weights.rules.json +41 -0
- package/src/rules/rule-registry.d.ts +48 -0
- package/src/rules/rule-registry.js +177 -0
- package/src/rules/tooling/command-runtime.rules.json +38 -0
- package/src/storage/backends/idb-store.d.ts +52 -0
- package/src/storage/backends/idb-store.js +590 -0
- package/src/storage/backends/memory-store.d.ts +36 -0
- package/src/storage/backends/memory-store.js +242 -0
- package/src/storage/backends/opfs-store.d.ts +41 -0
- package/src/storage/backends/opfs-store.js +429 -0
- package/src/storage/blake3.d.ts +17 -0
- package/src/storage/blake3.js +269 -0
- package/src/storage/download-types.d.ts +157 -0
- package/src/storage/download-types.js +48 -0
- package/src/storage/downloader.d.ts +103 -0
- package/src/storage/downloader.js +839 -0
- package/src/storage/emulated-vram.d.ts +264 -0
- package/src/storage/emulated-vram.js +576 -0
- package/src/storage/export.d.ts +20 -0
- package/src/storage/export.js +159 -0
- package/src/storage/index.d.ts +253 -0
- package/src/storage/index.js +185 -0
- package/src/storage/inventory.d.ts +26 -0
- package/src/storage/inventory.js +218 -0
- package/src/storage/preflight.d.ts +144 -0
- package/src/storage/preflight.js +294 -0
- package/src/storage/quickstart-downloader.d.ts +154 -0
- package/src/storage/quickstart-downloader.js +265 -0
- package/src/storage/quota.d.ts +150 -0
- package/src/storage/quota.js +304 -0
- package/src/storage/registry.d.ts +28 -0
- package/src/storage/registry.js +125 -0
- package/src/storage/reports.d.ts +20 -0
- package/src/storage/reports.js +94 -0
- package/src/storage/shard-manager.d.ts +137 -0
- package/src/storage/shard-manager.js +801 -0
- package/src/sw.d.ts +1 -0
- package/src/sw.js +187 -0
- package/src/tooling/browser-command-runner.d.ts +28 -0
- package/src/tooling/browser-command-runner.js +82 -0
- package/src/tooling/command-api.d.ts +147 -0
- package/src/tooling/command-api.js +523 -0
- package/src/tooling/command-envelope.d.ts +81 -0
- package/src/tooling/command-envelope.js +195 -0
- package/src/tooling/command-runner-shared.d.ts +73 -0
- package/src/tooling/command-runner-shared.js +146 -0
- package/src/tooling/command-runner.html +45 -0
- package/src/tooling/node-browser-command-runner.d.ts +30 -0
- package/src/tooling/node-browser-command-runner.js +868 -0
- package/src/tooling/node-command-runner.d.ts +36 -0
- package/src/tooling/node-command-runner.js +127 -0
- package/src/tooling/node-convert-worker-pool.d.ts +16 -0
- package/src/tooling/node-convert-worker-pool.js +186 -0
- package/src/tooling/node-convert-worker.d.ts +1 -0
- package/src/tooling/node-convert-worker.js +60 -0
- package/src/tooling/node-convert.d.ts +44 -0
- package/src/tooling/node-converter.d.ts +1 -0
- package/src/tooling/node-converter.js +1227 -0
- package/src/tooling/node-file-fetch.d.ts +1 -0
- package/src/tooling/node-file-fetch.js +38 -0
- package/src/tooling/node-source-runtime.d.ts +19 -0
- package/src/tooling/node-source-runtime.js +469 -0
- package/src/tooling/node-webgpu.d.ts +6 -0
- package/src/tooling/node-webgpu.js +321 -0
- package/src/tooling/opfs-cache.d.ts +11 -0
- package/src/tooling/opfs-cache.js +174 -0
- package/src/tooling/source-runtime-bundle.d.ts +102 -0
- package/src/tooling/source-runtime-bundle.js +484 -0
- package/src/tooling-exports.browser.d.ts +7 -0
- package/src/tooling-exports.browser.js +2 -0
- package/src/tooling-exports.d.ts +22 -0
- package/src/tooling-exports.js +7 -0
- package/src/tooling-exports.shared.d.ts +105 -0
- package/src/tooling-exports.shared.js +92 -0
- package/src/training/README.md +153 -0
- package/src/training/artifacts.d.ts +160 -0
- package/src/training/artifacts.js +896 -0
- package/src/training/attention-backward.d.ts +30 -0
- package/src/training/attention-backward.js +217 -0
- package/src/training/attention-forward.d.ts +22 -0
- package/src/training/attention-forward.js +82 -0
- package/src/training/autograd.d.ts +51 -0
- package/src/training/autograd.js +380 -0
- package/src/training/checkpoint.d.ts +31 -0
- package/src/training/checkpoint.js +238 -0
- package/src/training/clip.d.ts +9 -0
- package/src/training/clip.js +54 -0
- package/src/training/dataloader.d.ts +8 -0
- package/src/training/dataloader.js +44 -0
- package/src/training/datasets/index.d.ts +12 -0
- package/src/training/datasets/index.js +6 -0
- package/src/training/datasets/jsonl.d.ts +11 -0
- package/src/training/datasets/jsonl.js +50 -0
- package/src/training/datasets/reploid.d.ts +3 -0
- package/src/training/datasets/reploid.js +36 -0
- package/src/training/datasets/text-pairs.d.ts +21 -0
- package/src/training/datasets/text-pairs.js +42 -0
- package/src/training/datasets/token-batch.d.ts +21 -0
- package/src/training/datasets/token-batch.js +40 -0
- package/src/training/datasets/translation-pairs.d.ts +34 -0
- package/src/training/datasets/translation-pairs.js +49 -0
- package/src/training/export.d.ts +32 -0
- package/src/training/export.js +112 -0
- package/src/training/index.d.ts +52 -0
- package/src/training/index.js +41 -0
- package/src/training/lora.d.ts +19 -0
- package/src/training/lora.js +57 -0
- package/src/training/loss-scaling.d.ts +21 -0
- package/src/training/loss-scaling.js +80 -0
- package/src/training/loss.d.ts +10 -0
- package/src/training/loss.js +41 -0
- package/src/training/objectives/base.d.ts +58 -0
- package/src/training/objectives/base.js +38 -0
- package/src/training/objectives/cross_entropy.d.ts +18 -0
- package/src/training/objectives/cross_entropy.js +37 -0
- package/src/training/objectives/distill_kd.d.ts +16 -0
- package/src/training/objectives/distill_kd.js +369 -0
- package/src/training/objectives/distill_triplet.d.ts +16 -0
- package/src/training/objectives/distill_triplet.js +412 -0
- package/src/training/objectives/index.d.ts +12 -0
- package/src/training/objectives/index.js +6 -0
- package/src/training/objectives/ul_stage1_joint.d.ts +16 -0
- package/src/training/objectives/ul_stage1_joint.js +188 -0
- package/src/training/objectives/ul_stage2_base.d.ts +16 -0
- package/src/training/objectives/ul_stage2_base.js +222 -0
- package/src/training/optimizer.d.ts +22 -0
- package/src/training/optimizer.js +115 -0
- package/src/training/runner.d.ts +196 -0
- package/src/training/runner.js +1194 -0
- package/src/training/suite.d.ts +187 -0
- package/src/training/suite.js +3156 -0
- package/src/training/trainer.d.ts +89 -0
- package/src/training/trainer.js +301 -0
- package/src/training/ul_dataset.d.ts +47 -0
- package/src/training/ul_dataset.js +153 -0
- package/src/training/ul_schedule.d.ts +6 -0
- package/src/training/ul_schedule.js +29 -0
- package/src/types/chrome.d.ts +36 -0
- package/src/types/chrome.js +1 -0
- package/src/types/gpu.d.ts +185 -0
- package/src/types/gpu.js +5 -0
- package/src/types/index.d.ts +3 -0
- package/src/types/index.js +3 -0
- package/src/types/inference.d.ts +197 -0
- package/src/types/inference.js +5 -0
- package/src/types/model.d.ts +125 -0
- package/src/types/model.js +5 -0
- package/src/utils/index.d.ts +7 -0
- package/src/utils/index.js +7 -0
- package/src/utils/load-json.d.ts +5 -0
- package/src/utils/load-json.js +23 -0
- package/src/utils/plain-object.d.ts +1 -0
- package/src/utils/plain-object.js +3 -0
- package/src/utils/sha256.d.ts +4 -0
- package/src/utils/sha256.js +135 -0
- package/tools/convert-safetensors-node.js +180 -0
- package/tools/doppler-cli.js +1170 -0
|
@@ -0,0 +1,214 @@
|
|
|
1
|
+
// Auto-generated from src/config/kernels/registry.json + src/gpu/kernels/*.wgsl
|
|
2
|
+
// Content hash input: normalized WGSL source + "\n@@entry:<entry>"
|
|
3
|
+
export const KERNEL_REF_CONTENT_DIGESTS = Object.freeze({
|
|
4
|
+
"attention_bdpa_decode_f16.wgsl#main": "e2d73e455ea26a1cddb27eef2835e9e404db731255c7d2c6a5812245375562f2",
|
|
5
|
+
"attention_decode_chunked_f16.wgsl#main": "9733ffdcf79d064322ef400635104f2b4e3f9e34ca08a6e2638bab72769bac0a",
|
|
6
|
+
"attention_decode_chunked_f16kv.wgsl#main": "21ea87fa5a0267ee2a960fa24d7a6582fe9b3d8276695289f924601236a29d62",
|
|
7
|
+
"attention_decode_f16.wgsl#attention_decode": "cf38d065b282986456d7f8305b84f4909910c0bda845c4c3df39d1e9c6a1e675",
|
|
8
|
+
"attention_decode_f16kv.wgsl#attention_decode": "6b9453d2cb09c8f409d75c38f055cffd58093213ff5af4c5d6a0b49181412491",
|
|
9
|
+
"attention_decode_online_f16.wgsl#main": "953acecd07bd1e50829f26a7e6abb6c653331504d23914b77538da99ee1474dc",
|
|
10
|
+
"attention_decode_online_f16kv.wgsl#main": "4c5d8c92a0a111af716d6b46b9559446807c086027445c7fefa150202f43dae4",
|
|
11
|
+
"attention_decode_optimized.wgsl#main": "4d52fc85b1ecc4ae3d098afb56476e0961cec195c7247672c40b0c6d9ad215de",
|
|
12
|
+
"attention_decode_optimized.wgsl#main_f16kv": "5e86e9cd0d355563ecbccddd2c858551c73d8154131d92d06a9537a4d10246a8",
|
|
13
|
+
"attention_decode_optimized.wgsl#main_multihead": "78a07073d59366cf685aec7603ca40c15f833c863fc57f4fa55f73225880cf7f",
|
|
14
|
+
"attention_decode_paged_f16.wgsl#main": "773c9b37398ec7aea07bde0f53b80ca942eaa2dae76d906a0ad5d7a28a2deacd",
|
|
15
|
+
"attention_decode_paged_f16kv.wgsl#main": "4b4eba2a40a0b75c3ba25270f0ec4874ce0328852df6f678e8feedc7defa6718",
|
|
16
|
+
"attention_decode_subgroup.wgsl#main": "a23ad4ccd5ad438bcdbd60a396979e28d846ec0fc10034cc80f841b2b08bd89a",
|
|
17
|
+
"attention_decode_tiered_f16.wgsl#main": "fac78173568bf5c966cf5e1fb198d5b277cc7e3c7c56aa7475da889f45197f3e",
|
|
18
|
+
"attention_decode_tiered_f16kv.wgsl#main": "0fd8866d5bc4e9d994ae2d1f273600dda19963f99f44dc93bf57d7a170fa927e",
|
|
19
|
+
"attention_decode_tiered_int4_f16kv.wgsl#main": "3574d75f7211c47b261e217d25e9dbbea17b4d1a2a36ca117051038bfde6be51",
|
|
20
|
+
"attention_decode_tiered_int8_f16kv.wgsl#main": "89cf88af432c537b82a8d1e2e639b6867086dc61f705711b3ab7fb547e9cd491",
|
|
21
|
+
"attention_decode.wgsl#attention_decode": "14cbf16eebe643d2cb0c957d5f125f6f69ea34957f2daa19488ee8985f81db1b",
|
|
22
|
+
"attention_f16.wgsl#main": "693e9745454638ae328d25e99cd7cc76c51e47666bf5d9e228b78ed6bc5df876",
|
|
23
|
+
"attention_f16kv.wgsl#main": "a86b3a671eccec8db9f76ffa9da59ab7d6989b1647cd4e73b7abea13938889b8",
|
|
24
|
+
"attention_small_f16.wgsl#main": "fa51616d4a6b2322b2bfbc7cd39ab3b60e76ea49790cbda99b64c564db8d63fd",
|
|
25
|
+
"attention_small_f16kv.wgsl#main": "7e3acc24b8b45294d18052c759319fe8a202d4ebfc4b653d4b904b245ae7e5c9",
|
|
26
|
+
"attention_small.wgsl#main": "6752ddd7ab53e6235c9b5b1a9515141c0d111df7fac9f4c0d7a38f9943490ed4",
|
|
27
|
+
"attention_streaming_f16.wgsl#main": "2a1591412e581fb855bc54003c616819008c9ee21b56adee81afbe672eba32c9",
|
|
28
|
+
"attention_streaming_f16kv.wgsl#main": "b337a2dcf7b40431a733d1726eee8bf23504136fd5f915bec057fb59f3ea7480",
|
|
29
|
+
"attention_streaming.wgsl#main": "8dfc44cd633f0fd22d3a30b5699f2877c75ef3244bed04f71012245fe53a9e0a",
|
|
30
|
+
"attention.wgsl#main": "05eeaecb208128a415df3ab3acfc0642272a934d97bc919d00b59fdbaa35687b",
|
|
31
|
+
"backward/adam.wgsl#main": "5520ab52aed793a665c16f25db2d0eecb69b34f953ce870691ab20f7a2e3acb7",
|
|
32
|
+
"backward/attention_backward.wgsl#main": "482e144cc1c76f371e5b39f439e9cab591db03e469045424c7de107adbd498c4",
|
|
33
|
+
"backward/bias_add_backward.wgsl#main": "413b1fe8682c09075e2dccc40caf3ebdae13957133224fd4e4497ca9c5109c47",
|
|
34
|
+
"backward/conv2d_backward_input.wgsl#main": "34066c2fe6c9133206a26d487475fa880431d437f805ba9e6fc33f7a881c5d26",
|
|
35
|
+
"backward/conv2d_backward_weight.wgsl#main": "c7a66234f7c4c6d46413e5317f81bc70a2fcd93f455e43a21b4c557e3111d9b1",
|
|
36
|
+
"backward/cross_entropy_backward.wgsl#main": "9082afb09968061fc71af354a07d47d51b475432822a8d1094fdb42678c1d2db",
|
|
37
|
+
"backward/embed_backward.wgsl#main": "c6b476cde4e61dacdde7012c33b3c546ae6439824ff299e01607564eb393003f",
|
|
38
|
+
"backward/gelu_backward.wgsl#main": "6ea546fd461b5f1d88efdda75a362e01fc900996330aed9250f53364696c2b4a",
|
|
39
|
+
"backward/groupnorm_backward.wgsl#main": "2962192fd365b0d4e95a02273d07724d0f145137bf0e2545a9da85b21ed30ec3",
|
|
40
|
+
"backward/layernorm_backward.wgsl#main": "459ca43b663ce1fdaad6ba578e1df6719cc97c9d2c25b4fe8ea539028d4ab582",
|
|
41
|
+
"backward/matmul_backward.wgsl#main": "fd6569fe4649d0cc9154a2c245814f219c523e0f1f6b24d3de53121925d25983",
|
|
42
|
+
"backward/matmul_transpose_a.wgsl#main": "48f9efa999af7614cf575026afd9ea96f1fc7164843e9c46336ebad838e0cea0",
|
|
43
|
+
"backward/pixel_shuffle_backward.wgsl#main": "d56095ace4c14e267ae590f823a0c6d3da07c4dac35071ab17a831418f5bd27c",
|
|
44
|
+
"backward/rmsnorm_backward.wgsl#main": "cee88dc03432fdd4a8817997ba10a26738b9edd2f8f4b7c9cc14be499353bf49",
|
|
45
|
+
"backward/rope_backward.wgsl#main": "fa8bf2d6d20e543d33436b129bd63479c65a516bcf8b30b1cfd64e9b6bc9ca3b",
|
|
46
|
+
"backward/scale_backward.wgsl#main": "cb2e639b1049a8662c0694fe9e9757327bb55dbb4546ce1c298c5927be33a896",
|
|
47
|
+
"backward/silu_backward.wgsl#main": "86275df7afce1554a6b4203198eb592d32fe4469aaa1e2d35110106decb9ae6b",
|
|
48
|
+
"backward/softmax_backward.wgsl#main": "61c39e76841e33fdc8e91cac64202e15da4f867c015dba7e444b8ecaf3d66669",
|
|
49
|
+
"backward/upsample2d_backward.wgsl#main": "3f3a27fc880d3c4cba49889cafb4be66d30132cfeec9c2105751cc12a5a44ae3",
|
|
50
|
+
"bf16_to_f16.wgsl#main": "91c009d291a205fc42e7b45942e58630442eb67f8cb4bacc2e54160fa5f25c35",
|
|
51
|
+
"bf16_to_f32.wgsl#main": "2c5c08476d40aeb64287b8b31d1ef088c63e8bb4568bf7b7c5faf4a4ed493576",
|
|
52
|
+
"bias_add_f16.wgsl#main": "8ec47faa1a52190616b431ca8bbd7b1e657ef13720e35e787e4df16b65817b03",
|
|
53
|
+
"bias_add.wgsl#main": "d15e3f38379e5a7b393ab88c8a1a6eae13bbc8fec4348267d891ce5b7e8bb963",
|
|
54
|
+
"cast_f16_to_f32.wgsl#main": "98a0b31fcba2bad945e9f8522d643ae75135bfdf6b39a824565f57d5d4fd965d",
|
|
55
|
+
"cast_f32_to_f16.wgsl#main": "0b669e7812c09a2f44e219548e11ca0dfd8af921d30017e2a5c79d54f792542f",
|
|
56
|
+
"clamp.wgsl#main": "9d8039a590d102133636d67de8ae76ca8e9129bf220c48f7a08e9d82fcc48bad",
|
|
57
|
+
"conv2d_f16.wgsl#main": "aa139e9f0270873acbc1c4b3cbacff4d224cae7247b520ec129a4f068eb6ed59",
|
|
58
|
+
"conv2d.wgsl#main": "484a676692d2b8097daeefe42e2296a1f8b3ef11abfd7b41df6cdcdf16b7a8fd",
|
|
59
|
+
"cross_entropy_loss.wgsl#main": "5a48087bdec94184432c90ce5b345e1eadbdfcb13b9793ecee8052bc7392239c",
|
|
60
|
+
"dequant_f16_out_vec4.wgsl#main_vec4": "61c20e6c71c1c8421b4ec202dbd26292a6300587bd44c314f2a6c6d9d9442c3a",
|
|
61
|
+
"dequant_f16_out.wgsl#main": "94d61843d56f9a3bbc6b7c2b95dc6ecbba3f6a262b2c4086a076f69a8c38ccae",
|
|
62
|
+
"dequant_f16_rowwise.wgsl#main": "f5bf7cef950b52d65cee6121dbaa176244d3221045b3b6386b3be47f23ce17dc",
|
|
63
|
+
"dequant_f32_rowwise.wgsl#main": "e73606e1b47e1191203a210bececa8a597bcab8bcc535146718afa6a021cab0d",
|
|
64
|
+
"dequant_mxfp4_expert_f16.wgsl#main_expert": "96af52551ac40e1b86121a528a3ffaba835c5d0419e06407fed80353d46b17e1",
|
|
65
|
+
"dequant_mxfp4_expert.wgsl#main_expert": "5c7ab5519e222d0308a8b590e1d6449699588ae9b8208b12ed64bc2b2fb24136",
|
|
66
|
+
"dequant_mxfp4_vec4.wgsl#main_vec4": "947d4b3d790288878c7a614b2bedeb7ee92c67365bdf04c85eb10b1e65a435f8",
|
|
67
|
+
"dequant_mxfp4.wgsl#main": "885a5f752b684c6ca0bb10e3a1846a396eef14d2158e8c8ad31bd1dd4c74b9ef",
|
|
68
|
+
"dequant_q6k.wgsl#main": "be0aed027932d8b7dd1e92d0090ced39e4df8be724acf290f52db0004be9a35e",
|
|
69
|
+
"dequant_q8_0.wgsl#main": "ff5f800da963b0502a9ffab723cbcac0bbb5eb9a02898afc2aba2db215a58da7",
|
|
70
|
+
"dequant_shared_vec4.wgsl#main_vec4": "aaf330e814cbd4c2838f07639025e414542ec2d0ddbb34913d7f10d28253baf2",
|
|
71
|
+
"dequant_shared.wgsl#main": "d83d83a5d2bbcd29a06f67fb41496edaa45216438ff0dd177e7ec33af08104a4",
|
|
72
|
+
"dequant_subgroup.wgsl#main": "6b6296d8b060a9b16e7b3c2bee99dbbcbbf9cc3d557f1e2b000f7f9bfc020309",
|
|
73
|
+
"dequant_subgroup.wgsl#main_vec4": "d9de6ea645f558bcdf6db622fe146710e143fc4c2bb56f130342d62858eb456c",
|
|
74
|
+
"energy_eval_f16.wgsl#main": "09223ae193593f3555866a3acfe76ca35442ef4f3967cae376bdcc211f3054b3",
|
|
75
|
+
"energy_eval.wgsl#main": "e10d9572397ebece5275aecd907cba5970f6a5c3744dd8b982677efb8982bdd2",
|
|
76
|
+
"energy_quintel_grad_f16.wgsl#main": "eb87ed8592b46b0a4d866c245b664cadb2bca016f72419e763402a6a721c4951",
|
|
77
|
+
"energy_quintel_grad.wgsl#main": "2b64d4b2cd637f38f82d7aa7dd03014d387eae7c98cff42654f4c669f95e2bd4",
|
|
78
|
+
"energy_quintel_reduce_f16.wgsl#main": "d02bc77975ba540ea78a5023771c44a9dc100dde41acc00a6c80960c339189b3",
|
|
79
|
+
"energy_quintel_reduce.wgsl#main": "1ef0033f55e63402f9046b02bbb8bf6f580de3f552db2d02343212bf45a51408",
|
|
80
|
+
"energy_quintel_update_f16.wgsl#main": "c10c1d15cb09087d039141466caaf8acec7e0d0eb8bf103563544846705448d3",
|
|
81
|
+
"energy_quintel_update.wgsl#main": "34ef811f38b8b63e209f2aefa37b5c0241750fca2ca3f3ed6811bd75f16b4ffe",
|
|
82
|
+
"energy_update_f16.wgsl#main": "4badaf73bef8a860d8cb5bd3dff914ab2767d7b836e00c84ace3f46e90e186fd",
|
|
83
|
+
"energy_update.wgsl#main": "e2e01530f4b09903b847b3943e042dc061c0fb9994577bc13b1cadf775bb633f",
|
|
84
|
+
"fused_ffn_f16.wgsl#main": "00a835aa5bf3e9dacbf5ef0935fed04a98fac1d9ef369d1911bd7df29ed136d4",
|
|
85
|
+
"fused_ffn_f16.wgsl#main_batched": "9b8303071fccb88a9bbe309d1240a6e117e04208f846a6e39e5a68aa68f0c31e",
|
|
86
|
+
"fused_ffn_q4k.wgsl#main": "b19047a2b3f7c06db05d652483210ac2e26181f44bb8fb69917782da6a8af144",
|
|
87
|
+
"fused_ffn_q4k.wgsl#main_batched": "6bcaef0c18b0d74a2be880b6a8f6cf76da0033c77661947e2591a6cf13470321",
|
|
88
|
+
"fused_ffn.wgsl#main": "9f6592f201a428815598fc43192145aa0806e7b288a8fe58e9b8eb54908237b5",
|
|
89
|
+
"fused_ffn.wgsl#main_batched": "379fcc43f0341a8854e4ba48ce30b7d89f8de3deb07dd5bb37c1ea09239e6cc0",
|
|
90
|
+
"fused_ffn.wgsl#main_f16": "d2abaaca1ff61eacfa1bf3843649b8a86ac69f4c9dd3da33f59584766cae977b",
|
|
91
|
+
"fused_ffn.wgsl#main_multi": "a7bf0f9fd78399c0b4c52e6a07d36fd98b8155ba992382716a486aa8042e0147",
|
|
92
|
+
"fused_matmul_q4_batched_f16.wgsl#main_batched_f16": "6d54a54a91dab2218ca6cb086ae7cf2976fffa943c3440c972598c6a91561525",
|
|
93
|
+
"fused_matmul_q4_batched_f16a.wgsl#main_batched_f16a": "54c6a50454dfe59aa7a037340c1d6b8f57cf0a7c6eac1d0a338b7d708fd81dc7",
|
|
94
|
+
"fused_matmul_q4_batched.wgsl#main_batched": "a513e4170792aaf2f1c22b73f34288721541acf5954756993a0001a412c91209",
|
|
95
|
+
"fused_matmul_q4_f16a.wgsl#main_f16a": "f0373d3cacf182784382efa2ff3cd2470f7db000e602ceaae402a86c11008cfc",
|
|
96
|
+
"fused_matmul_q4_multicol_f16.wgsl#main_multicol_f16": "f63d3f7ca356bd38dc8e37877a1e9608b90271d053db2e340ab8ee37d7b660a8",
|
|
97
|
+
"fused_matmul_q4_multicol_f16a.wgsl#main_multicol_f16a": "b381dbc4de5792d34e5f0827502590e8730d8797f6db817b7ce72a0efbbfa308",
|
|
98
|
+
"fused_matmul_q4.wgsl#main": "3b1a92709281cfdcfef4dab1acbdba03d5d886058133235791f7748c085bc76c",
|
|
99
|
+
"fused_matmul_q4.wgsl#main_multicol": "c0262226752914e863e8b5dc34479b523f9c93e2cef69c9057dcbde2b82854ef",
|
|
100
|
+
"fused_matmul_rmsnorm_f16.wgsl#gemv_rmsnorm_medium": "3a72506f4565f8be0b358f7ed3e57d59c15913b6f053e2335d3d62a5745516f9",
|
|
101
|
+
"fused_matmul_rmsnorm_f16.wgsl#gemv_rmsnorm_small": "10e06578e0fd1245b84243e5e0147bca0f4641aee4c145315976aa82027d497b",
|
|
102
|
+
"fused_matmul_rmsnorm.wgsl#gemv_rmsnorm_medium": "ea26499897013c1ba83d8681e6906a26b1206ab17c591d47e54dd1d804d92cf3",
|
|
103
|
+
"fused_matmul_rmsnorm.wgsl#gemv_rmsnorm_phase1": "690aec36750c72452ff7390e6c345b0c729a41785f91a459f969b9e570de5ea0",
|
|
104
|
+
"fused_matmul_rmsnorm.wgsl#gemv_rmsnorm_small": "7de2c1e3591b71d38c98363b2496d525c9e44c811a95ee8ffc33e3dac304b48e",
|
|
105
|
+
"fused_swiglu_f16.wgsl#main": "d3d3c7b7aff3ab9b4573ca9e488d8f70b74b455a3c65cc94f5c132052212253d",
|
|
106
|
+
"fused_swiglu.wgsl#main": "1cd30d157f33273f41bdb503ffa044b95fb57ad6acc218edef363cb212dff8bc",
|
|
107
|
+
"gather_f16_f16_out.wgsl#gather_f16_out": "a190612e26675c5d5b8da39225de0ad722d212be3258d61b89014551a782fc0e",
|
|
108
|
+
"gather_f16_out.wgsl#gather_f16_out": "b7ee5469f4ed670201fd3748de3c2f18bdf3446ff54e87db9a8ac1b80fb002ca",
|
|
109
|
+
"gather_f16_vec4_f16_out.wgsl#gather_vec4_f16_out": "a9cd29445410c019262ca78ab27a0cd3b167fd61542488815c64e642b4c31dd5",
|
|
110
|
+
"gather_f16_vec4.wgsl#gather_vec4": "dc87e5e03cdbff74347ef8e42774570b1807d239ad64501a7e90ec8f600e912c",
|
|
111
|
+
"gather_f16.wgsl#main": "a4829f4067091c98ad6ebbc9b0744cdd5bbcd4fbf6092b2f7cc7f1098695860f",
|
|
112
|
+
"gather_vec4_f16_out.wgsl#gather_vec4_f16_out": "dd631155d4df58d29a1c444f521a8ebbf28532df3eb37d1b27f8015d3a42d9e6",
|
|
113
|
+
"gather_vec4.wgsl#gather_vec4": "e8d90b826350a9feb96476fc4a5e568aebfc29948cf07a2d7db0dee4a4216b6a",
|
|
114
|
+
"gather.wgsl#main": "777991fb6e4b3b506e4493b47ee998afe541924ddd7c04e1eadf4cb7fd719ef8",
|
|
115
|
+
"gelu_f16.wgsl#main": "34b2f981e8b96bd624748b84bad0512940a01ada1721e0ffb8e149259b82b22a",
|
|
116
|
+
"gelu.wgsl#main": "a9007ea08aaff98f9be08f1e0490a6bcf252883eac5513de876ab9ce918865e6",
|
|
117
|
+
"gptoss_mxfp4_expert_fused.wgsl#main_expert": "3159e8cd81da13f909cf905e6d35307fefe1dcbbdf1b2b8e8ff0ce923bd71180",
|
|
118
|
+
"gptoss_router_topk.wgsl#softmax_topk": "86e4ea709c0c0084d09c6a4cd07710dc14f380e03f91b8ed9ec871b310be49f1",
|
|
119
|
+
"groupnorm_apply_f16.wgsl#main": "cfd850b87944ac1c03ba7bd98136db556dadd8a70611e351d82d297299a7cd02",
|
|
120
|
+
"groupnorm_apply.wgsl#main": "b09b8f2f57dcdfa1a0366daa30d3910feb134204652c711d2ba564e566b5a334",
|
|
121
|
+
"groupnorm_stats_f16.wgsl#main": "fb76f78ce668ea8459110335698fe4b09a2425fc71deed3bab67efd7641c3199",
|
|
122
|
+
"groupnorm_stats.wgsl#main": "caae6dbb707b06ebb40605c74c108fc93195da3a08fec4013263e56e0af5f5de",
|
|
123
|
+
"kv_quantize_int4.wgsl#main": "e52cd0d70512e8e7a308259326c2e912b35643242f2ae6ea6cfc8580a8915c0b",
|
|
124
|
+
"kv_quantize_int8.wgsl#main": "7b0db690816603555478165c39216fd774a7ace9aac627f2151e5842350d4a0c",
|
|
125
|
+
"layernorm_f16.wgsl#main": "f0b3858fabbc5a9b28e80b74e5b8d5a778bed9372e675e961be95a57d757622c",
|
|
126
|
+
"layernorm.wgsl#main": "8ff885017bfcfc8fdb8d16e09c358c000830bfd5fbf3d87187694edb0cb44f7c",
|
|
127
|
+
"matmul_f16_tiled.wgsl#main": "30b6cb720df1565aa60a91d144bd2edacf17473b6e962f41768700f1d945d012",
|
|
128
|
+
"matmul_f16.wgsl#main": "b2a7e4a4db27d008da957225b361ef688f299ece792c844e2d2542b21d3fdfbe",
|
|
129
|
+
"matmul_f16.wgsl#main_vec4": "b398bdcd0f4900fdfd743e47d6c31e5f359409976e6b860b838adc98647ba25f",
|
|
130
|
+
"matmul_f16w_f32a_tiled.wgsl#main": "e94ae5374e8b43dd48b663eff59a45c822c3784d5702a1145266b6ffd15ba78c",
|
|
131
|
+
"matmul_f16w_f32a.wgsl#main": "027a8f1cd9713cbe0b0ada160bd175e0542bb90896ad85441b023522d9a1befc",
|
|
132
|
+
"matmul_f32.wgsl#main": "b5bb8e3d8014136e33de7935dd2a1f074c988044fe05cf5b559718c6f061eaa8",
|
|
133
|
+
"matmul_gemv_f16a.wgsl#main": "7d369dc00658f771628ed52b690f2764e054216bce54e18d86e4d946c218baa8",
|
|
134
|
+
"matmul_gemv_residual_f16.wgsl#main": "c32de0d29709b08462a83c890522aa3ec0583decbfaf28f37be4e90986cd8d06",
|
|
135
|
+
"matmul_gemv_residual.wgsl#main": "20e600aba7dac1ece7f60fdae339c85f571584d1e305426702f3519851381b71",
|
|
136
|
+
"matmul_gemv_subgroup_f16a.wgsl#main": "a064a8ba5a0db58cba769e1f95c2183dac0658fee1e14c6be2b2160da6982bdb",
|
|
137
|
+
"matmul_gemv_subgroup_f16a.wgsl#main_cols8": "506cfa19640afe4eab87f97ebf8b313a927623be5c74b09c3dda0461e5146dff",
|
|
138
|
+
"matmul_gemv_subgroup_f16a.wgsl#main_multicol": "c8e86ecbbefa27a3b7366af676d89a992c2e951329cdf19abb57b9c90144379e",
|
|
139
|
+
"matmul_gemv_subgroup_f16a.wgsl#main_vec4": "f227a403cdf9717dd68224c9ea55708ffe14c618d8146f5d48b42af0f253df29",
|
|
140
|
+
"matmul_gemv_subgroup_f16a.wgsl#main_vec4_cols8": "9e7aba97a6cf199b3f574166e295ea051ebd59e308b5f6f2ce5a4de2d04963ce",
|
|
141
|
+
"matmul_gemv_subgroup.wgsl#main": "ff526a7e70d6933f5c7a59aee476cc9c5cf123903222ed2e74f555e86e1ac757",
|
|
142
|
+
"matmul_gemv_subgroup.wgsl#main_multicol": "96c38c15e6fed0d7efdc5cd094db5843a8e8ddfe01eee3bc7322fa555dacf3d0",
|
|
143
|
+
"matmul_gemv_subgroup.wgsl#main_vec4": "3cee3bed453b40c5564a751d2a917649e10ad52f5268e77cbfecfcee34780457",
|
|
144
|
+
"matmul_gemv.wgsl#main": "dc892efc87edc6d5ddaf191b86c1cc41a603352a332023aa0b1fe55d166673d0",
|
|
145
|
+
"modulate_f16.wgsl#main": "44a98cda1cc7a3575788f865173b9890be792c94e852ac8311b6b8ffbdc1438d",
|
|
146
|
+
"modulate.wgsl#main": "dfe88a35b94752573199c16b3d8aecd4e8e7da57dc88d7b342aa61e0122e71ec",
|
|
147
|
+
"moe_gather_f16.wgsl#count_and_map": "300342dd05555b85fcb5b0da468b5879fc67a4ce7c295d3e531f13aabb7cfbdc",
|
|
148
|
+
"moe_gather_f16.wgsl#gather_tokens": "49cb7cfb105cb74cb3702c210f2425683d5e568d3084848d5c2ac2521663619d",
|
|
149
|
+
"moe_gather_vec4.wgsl#gather_tokens_vec4": "0ab4b258937ac226eb8bc88271a2aeafc11b8f41e41091efc903e8078333da2a",
|
|
150
|
+
"moe_gather.wgsl#count_and_map": "38b73b67d8f30defb86e19bc4b2f7530a478092f027ccd53d29e48a6911ef926",
|
|
151
|
+
"moe_gather.wgsl#gather_single_pass": "8dbd0c38a323ba6c73af3ecaad297d79d861b817152e6e2c12fcd9db6f767f1e",
|
|
152
|
+
"moe_gather.wgsl#gather_tokens": "58e40a63a39c2f99c47c75dee71767f98482217afd74607cee1b7efc2c85738b",
|
|
153
|
+
"moe_offsets.wgsl#build_offsets": "3ea004145fa234659408cdeb0d4d802adff1037c9c5c03af146b3734cc69dd27",
|
|
154
|
+
"pixel_shuffle_f16.wgsl#main": "57903a9c19cecc56371b2198402745127115680d266c3ce609201be9119aa359",
|
|
155
|
+
"pixel_shuffle.wgsl#main": "845b88700b1b46d18cde6f2ec11bb89512c90d7e148763e74ce2a4173fd99b21",
|
|
156
|
+
"residual_f16_vec4.wgsl#add_vec4": "30e9226fb6636e2f01e65b1dc8e93c8e849a87acec6215342fc114996da1ed41",
|
|
157
|
+
"residual_f16.wgsl#main": "d392433f3065d1caf68b033219f4ffacf022dc1f90fc3cf3fd620e4ba49f3219",
|
|
158
|
+
"residual_vec4.wgsl#add_vec4": "ef011d1683e62887db712da563e783d12fdc80c152955661137d2dca612d7d6a",
|
|
159
|
+
"residual.wgsl#main": "1fc456b14e2fb2bc9627107b4e51e7a2098f723b5ba6ab5542cd9455af99f423",
|
|
160
|
+
"rmsnorm_f16.wgsl#main": "21d6d702cfcb8d653d8e105bbbc522e85704fa5628b865a7d36cde338ca5c779",
|
|
161
|
+
"rmsnorm_f16.wgsl#rmsnorm_small_f16": "10803373bce7d5d5eb0939821e629e0f3fc4e8d27cdddab89f14a9db02e83603",
|
|
162
|
+
"rmsnorm.wgsl#main": "c529986befb29a04b94d89744585923a7cef82baf4b2b0a243aa2431618622cc",
|
|
163
|
+
"rmsnorm.wgsl#main_cached": "828d30dd8b6137457375b9f7b446a6bbdc1df826ce4d9b1818f6d1abe7adc9c5",
|
|
164
|
+
"rmsnorm.wgsl#main_small": "2baf3a1a1a1e982cf9ffcddd3646655503536148c4a3ce9938646bc05cf701b5",
|
|
165
|
+
"rmsnorm.wgsl#main_small_subgroup": "909b52d4dfcdd780f531afeb6964885adc90552487f36f131aa349bcc54fb495",
|
|
166
|
+
"rmsnorm.wgsl#main_subgroup": "545ebdd1e6e3ab9790a17a4dd5501b12dbb7bc0073787f9a4f5821752a75991c",
|
|
167
|
+
"rope_f16.wgsl#main": "a87f2964b77e851a2fbcc88305adeecaf8eba372291d83a71b817c8ef3da5c58",
|
|
168
|
+
"rope_f16.wgsl#rope_compute_freqs": "c7aa2cb50420ab2709b20e0a33e93ab6aa4f50d2fa8d9f79b0bfcfefb2f7abab",
|
|
169
|
+
"rope_f16.wgsl#rope_ntk_scaled": "46d2574b46539b289371c0c37a0b4e2fb21279134126b36a1fc30b98523905dd",
|
|
170
|
+
"rope_f16.wgsl#rope_qk": "3bc01e167dc3cc5397bd7751e493311b4d3f5c6c0f6fad30234a740bb4c4507b",
|
|
171
|
+
"rope_f16.wgsl#rope_yarn": "9b788dd05a1598aadcba5b0218d1666ce75faadbe32b71ee1def65ec23fb7dfe",
|
|
172
|
+
"rope.wgsl#main": "b639fe8a54508115c82c13c923bfea89f59c6e15a5bef66bfc34e12f0ab4e32f",
|
|
173
|
+
"rope.wgsl#rope_compute_freqs": "3d3a25df6036c9e87f6270bd6e1f67ffcd5cf3a1aac7cea19b1935fdb6ea6046",
|
|
174
|
+
"rope.wgsl#rope_ntk_scaled": "14b044c70e8f7bc775606a88298fefbf1181024f45775497cf2350682879fcd4",
|
|
175
|
+
"rope.wgsl#rope_qk": "d84c973564235347fa16b13552b5a6d7ab524f4ac9b65f0504fec07568b2f032",
|
|
176
|
+
"rope.wgsl#rope_yarn": "00b473ce1d86f16444d948739da39e834e21d4db30ab09b0df7502d2e4c511b2",
|
|
177
|
+
"sample_f16.wgsl#argmax": "30b9f199b49352e5aff91b7aa8016edb423ce33f77481c3a7bc184251856fb27",
|
|
178
|
+
"sample_f16.wgsl#argmax_reduce": "a3ca27fc50b10c36c1676bdd5dbfe5edc67850cdd5c1af7a1d3ad70f830dd8a7",
|
|
179
|
+
"sample_f16.wgsl#find_topk_phase1": "24e47e5ced28af802959e350ff0a6eec6b9a26f89fb38e222990eeaffb16bd36",
|
|
180
|
+
"sample_f16.wgsl#find_topk_phase2": "f40b76d7e1e5606ff6a1e369cf285cb3e802047cfd81377458d4ed0248fcb6a5",
|
|
181
|
+
"sample_f16.wgsl#sample_single_pass": "f3551288786f0a47f5222e07ab7f54a646e422c2493a62b8074202601eb06ddb",
|
|
182
|
+
"sample_f16.wgsl#softmax_and_sample": "f1201f01af94c5536463be6439fbe7ed166fcb9494ea081f7d393be3125a7efd",
|
|
183
|
+
"sample.wgsl#argmax": "bde072faa1ce26e19b6fa1f4ba24c4cd2465909c77a60b175bde1d250113a3b4",
|
|
184
|
+
"sample.wgsl#argmax_reduce": "b7fa3aea69d9888b9385b380dc70416350d375564b95688592768cd81560b214",
|
|
185
|
+
"sample.wgsl#find_topk_phase1": "d5fbe00d9ad0c89bd3af72a06f7c964c1b6c2634974ed9a76397fd441422827c",
|
|
186
|
+
"sample.wgsl#find_topk_phase2": "940b216e605d22096da5aca65950a8030866fc5a39e7fdf484d69a832de1b63a",
|
|
187
|
+
"sample.wgsl#sample_single_pass": "4412357e84113ee2f1bc0dc8bf89e314c2ab482c89c14ca016ea9949d16a9d0c",
|
|
188
|
+
"sample.wgsl#softmax_and_sample": "7172c60e76430fbe130e530e3564b569b45eccf193987b32d6f52bd6bbcc9f08",
|
|
189
|
+
"scale.wgsl#main": "44ec481452b586307957163e3d65c9d02561d3f2f3db633f906f5488b1ea1ca4",
|
|
190
|
+
"scale.wgsl#main_inplace": "020824c7118a59c461ce81f1c2cd01b7c2a3f1aab326392b7d48d4448a0c2ed1",
|
|
191
|
+
"scatter_add_dynamic_f16_weights.wgsl#scatter_add_dynamic": "42799e745bc445b199b1cbc384bc12bb9372ed1599af3260a803cefc8dd35497",
|
|
192
|
+
"scatter_add_dynamic_f16.wgsl#scatter_add_dynamic": "d341c5d6e943fda09c8df9875019c295da734599a671fdfe86fea2351ea4ecf2",
|
|
193
|
+
"scatter_add_dynamic.wgsl#scatter_add_dynamic": "fe805dc11dbe358aa1cc733152a071ee4947060a1ba6fb2b006b3e8e5d9c94fe",
|
|
194
|
+
"scatter_add_vec4.wgsl#scatter_add_vec4": "247c4f23129cdfbb19593b17c5833d85048da117d77141c74bc4e16e691d94e1",
|
|
195
|
+
"scatter_add.wgsl#main": "dea947b8014e9b674e4fec8f15fac6c926e8a3a4d8eff104b953d77f35a1ac35",
|
|
196
|
+
"scatter_add.wgsl#scatter_add_accumulate": "561800af22dedae63f1abe69b757b0ef6c7832a2bff228c2262e0b7111d89247",
|
|
197
|
+
"silu_f16.wgsl#main": "66a87710c1a680f9bf8d01f804709ecd2ce6a99bf2ee82a5f7e199e48eca6e29",
|
|
198
|
+
"silu.wgsl#main": "e833f98b090845ada7b9ae8edc68cd6cfe823bd1d5ae7a6401f32b6e3bb6ee75",
|
|
199
|
+
"softmax_subgroup.wgsl#main_subgroup": "88472c0dab5f81c5f045f0ee79c4c3bb484791a4a2b84af398c019851438f091",
|
|
200
|
+
"softmax_subgroup.wgsl#softmax_small_subgroup": "5d7bd1b698910a437197bf6c8b7f8b259036dd006ad5470f767b539dba8538f8",
|
|
201
|
+
"softmax.wgsl#main": "45c5876806b442222d7e190e595f55a0079bae82e07d37586996c1a63790bb7a",
|
|
202
|
+
"softmax.wgsl#softmax_online": "6c62601ba2f88f7de9dacf026cc2357168df47d009fd108736655b645217cd0a",
|
|
203
|
+
"softmax.wgsl#softmax_small": "ad75f10e0a3caadd278130504e7d0e4e1b2f0621f8bd390abae5d973e301e47b",
|
|
204
|
+
"split_qkv_f16.wgsl#main": "bd1a92fcd9382bdcdf00bcc59248a12489444904a4f10845a381f177b6ad649f",
|
|
205
|
+
"split_qkv.wgsl#main": "bc7c95a47322edc11fec19105efd3774c2adfed151530c849909d03af7503e4c",
|
|
206
|
+
"topk_f16_weights.wgsl#softmax_topk": "863559c28eb46a2b4dc16f21a19aca2424a5d68fc3430b29461bebdd7ec8f625",
|
|
207
|
+
"topk_f16.wgsl#softmax_topk": "2e3815ec3cab6cdc3072c1fcac0bc579a09ad7ad1114b903e8d3f98a315a3ad3",
|
|
208
|
+
"topk.wgsl#main": "a18763303cd18e8a020e647f8a52f65403526849faf835d9f9394f634c3c97eb",
|
|
209
|
+
"topk.wgsl#softmax_topk": "95ff3517da909e4bd4d0ff8d85b619bd250522943aeb9276375edc59f67e9604",
|
|
210
|
+
"topk.wgsl#topk_2_small": "289eaa5c4f005e0aaf37dfe5343aeda30d9ab3929979dbf0cc3553f23e136807",
|
|
211
|
+
"transpose.wgsl#main": "3f80f9db1b150f453a9abb62300bfef35e744ec6a8c774c7b64f2ad6088204e5",
|
|
212
|
+
"upsample2d_f16.wgsl#main": "961f876852d9f4f1d3b224f39cdae3fa434378db12f7903d23a8dab6824644d4",
|
|
213
|
+
"upsample2d.wgsl#main": "9ff2d864c58a6776b7e03221bc6a7409e9e646fbabe7d9a33f4e06ed62120e43",
|
|
214
|
+
});
|
|
@@ -0,0 +1,17 @@
|
|
|
1
|
+
export interface KernelRefSchema {
|
|
2
|
+
id: string;
|
|
3
|
+
version: string;
|
|
4
|
+
digest: string;
|
|
5
|
+
}
|
|
6
|
+
|
|
7
|
+
export declare const KERNEL_REF_VERSION: string;
|
|
8
|
+
|
|
9
|
+
export declare function getKernelRefContentDigest(kernel: string, entry?: string): string;
|
|
10
|
+
export declare function buildKernelRefFromKernelEntry(kernel: string, entry?: string): KernelRefSchema;
|
|
11
|
+
export declare function buildLegacyKernelRefFromKernelEntry(kernel: string, entry?: string): KernelRefSchema;
|
|
12
|
+
export declare function isKernelRefBoundToKernel(
|
|
13
|
+
kernelRef: KernelRefSchema | Record<string, unknown> | null | undefined,
|
|
14
|
+
kernel: string,
|
|
15
|
+
entry?: string
|
|
16
|
+
): boolean;
|
|
17
|
+
|
|
@@ -0,0 +1,75 @@
|
|
|
1
|
+
import { sha256Hex } from '../../utils/sha256.js';
|
|
2
|
+
import { KERNEL_REF_CONTENT_DIGESTS } from './kernel-ref-digests.js';
|
|
3
|
+
|
|
4
|
+
export const KERNEL_REF_VERSION = '1.0.0';
|
|
5
|
+
|
|
6
|
+
function normalizeKernelRefIdToken(value) {
|
|
7
|
+
return String(value ?? '')
|
|
8
|
+
.trim()
|
|
9
|
+
.toLowerCase()
|
|
10
|
+
.replace(/[^a-z0-9]+/g, '.')
|
|
11
|
+
.replace(/^\.+|\.+$/g, '') || 'kernel';
|
|
12
|
+
}
|
|
13
|
+
|
|
14
|
+
function normalizeKernel(value) {
|
|
15
|
+
const kernel = String(value ?? '').trim();
|
|
16
|
+
if (kernel.length === 0) {
|
|
17
|
+
throw new Error('kernel must be a non-empty string');
|
|
18
|
+
}
|
|
19
|
+
return kernel;
|
|
20
|
+
}
|
|
21
|
+
|
|
22
|
+
function normalizeEntry(value) {
|
|
23
|
+
const entry = String(value ?? 'main').trim() || 'main';
|
|
24
|
+
return entry;
|
|
25
|
+
}
|
|
26
|
+
|
|
27
|
+
export function getKernelRefContentDigest(kernel, entry = 'main') {
|
|
28
|
+
const normalizedKernel = normalizeKernel(kernel);
|
|
29
|
+
const normalizedEntry = normalizeEntry(entry);
|
|
30
|
+
const key = `${normalizedKernel}#${normalizedEntry}`;
|
|
31
|
+
const digest = KERNEL_REF_CONTENT_DIGESTS[key];
|
|
32
|
+
if (typeof digest !== 'string' || digest.length !== 64) {
|
|
33
|
+
throw new Error(`No kernel content digest registered for "${key}"`);
|
|
34
|
+
}
|
|
35
|
+
return digest;
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
export function buildKernelRefFromKernelEntry(kernel, entry = 'main') {
|
|
39
|
+
const normalizedKernel = normalizeKernel(kernel);
|
|
40
|
+
const normalizedEntry = normalizeEntry(entry);
|
|
41
|
+
const kernelToken = normalizeKernelRefIdToken(normalizedKernel.replace(/\.wgsl$/i, ''));
|
|
42
|
+
const entryToken = normalizeKernelRefIdToken(normalizedEntry);
|
|
43
|
+
const digest = getKernelRefContentDigest(normalizedKernel, normalizedEntry);
|
|
44
|
+
return {
|
|
45
|
+
id: `${kernelToken}.${entryToken}`,
|
|
46
|
+
version: KERNEL_REF_VERSION,
|
|
47
|
+
digest: `sha256:${digest}`,
|
|
48
|
+
};
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
export function buildLegacyKernelRefFromKernelEntry(kernel, entry = 'main') {
|
|
52
|
+
const normalizedKernel = normalizeKernel(kernel);
|
|
53
|
+
const normalizedEntry = normalizeEntry(entry);
|
|
54
|
+
const kernelToken = normalizeKernelRefIdToken(normalizedKernel.replace(/\.wgsl$/i, ''));
|
|
55
|
+
const entryToken = normalizeKernelRefIdToken(normalizedEntry);
|
|
56
|
+
const digest = sha256Hex(`${normalizedKernel}#${normalizedEntry}`);
|
|
57
|
+
return {
|
|
58
|
+
id: `${kernelToken}.${entryToken}`,
|
|
59
|
+
version: KERNEL_REF_VERSION,
|
|
60
|
+
digest: `sha256:${digest}`,
|
|
61
|
+
};
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
export function isKernelRefBoundToKernel(kernelRef, kernel, entry = 'main') {
|
|
65
|
+
if (!kernelRef || typeof kernelRef !== 'object' || Array.isArray(kernelRef)) {
|
|
66
|
+
return false;
|
|
67
|
+
}
|
|
68
|
+
const expected = buildKernelRefFromKernelEntry(kernel, entry);
|
|
69
|
+
return (
|
|
70
|
+
kernelRef.id === expected.id
|
|
71
|
+
&& kernelRef.version === expected.version
|
|
72
|
+
&& kernelRef.digest === expected.digest
|
|
73
|
+
);
|
|
74
|
+
}
|
|
75
|
+
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
{
|
|
2
|
+
"id": "gpt-oss-moe-v1",
|
|
3
|
+
"description": "Deterministic GPT-OSS MoE kernel routing profile.",
|
|
4
|
+
"router": {
|
|
5
|
+
"topk": [
|
|
6
|
+
{
|
|
7
|
+
"match": { "hasF16": true, "hasSubgroups": true, "routerDtype": "f16", "weightsDtype": "f16" },
|
|
8
|
+
"value": "gptoss_router_topk"
|
|
9
|
+
},
|
|
10
|
+
{
|
|
11
|
+
"match": { "hasF16": true, "routerDtype": "f16", "weightsDtype": "f16" },
|
|
12
|
+
"value": "fused_f16_w16"
|
|
13
|
+
},
|
|
14
|
+
{
|
|
15
|
+
"match": { "routerDtype": "f16" },
|
|
16
|
+
"value": "fused_f16"
|
|
17
|
+
},
|
|
18
|
+
{
|
|
19
|
+
"match": {},
|
|
20
|
+
"value": "fused"
|
|
21
|
+
}
|
|
22
|
+
]
|
|
23
|
+
},
|
|
24
|
+
"dequant": {
|
|
25
|
+
"mxfp4Expert": [
|
|
26
|
+
{
|
|
27
|
+
"match": {
|
|
28
|
+
"hasF16": true,
|
|
29
|
+
"hasSubgroups": true,
|
|
30
|
+
"outputDtype": "f16",
|
|
31
|
+
"groupSize": 32,
|
|
32
|
+
"tileShape": "vec4"
|
|
33
|
+
},
|
|
34
|
+
"value": "mxfp4_expert_fused_f16"
|
|
35
|
+
},
|
|
36
|
+
{
|
|
37
|
+
"match": {
|
|
38
|
+
"hasF16": true,
|
|
39
|
+
"outputDtype": "f16"
|
|
40
|
+
},
|
|
41
|
+
"value": "mxfp4_expert_f16"
|
|
42
|
+
},
|
|
43
|
+
{
|
|
44
|
+
"match": {},
|
|
45
|
+
"value": "mxfp4_expert"
|
|
46
|
+
}
|
|
47
|
+
]
|
|
48
|
+
}
|
|
49
|
+
}
|
|
@@ -0,0 +1,86 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Kernel Registry Loader
|
|
3
|
+
*
|
|
4
|
+
* Loads and caches the kernel registry from JSON.
|
|
5
|
+
* Provides resolved kernel configs with base + variant merged.
|
|
6
|
+
*
|
|
7
|
+
* @module config/kernels/registry
|
|
8
|
+
*/
|
|
9
|
+
|
|
10
|
+
import type {
|
|
11
|
+
KernelRegistrySchema,
|
|
12
|
+
OperationSchema,
|
|
13
|
+
KernelVariantSchema,
|
|
14
|
+
BindingSchema,
|
|
15
|
+
ResolvedKernelConfig,
|
|
16
|
+
} from '../schema/kernel-registry.schema.js';
|
|
17
|
+
import type { RuntimeCapabilities } from '../schema/platform.schema.js';
|
|
18
|
+
|
|
19
|
+
/**
|
|
20
|
+
* Set the URL for loading the registry.
|
|
21
|
+
* Must be called before getRegistry() if not using default.
|
|
22
|
+
*/
|
|
23
|
+
export function setRegistryUrl(url: string): void;
|
|
24
|
+
|
|
25
|
+
/**
|
|
26
|
+
* Get the kernel registry, loading it if needed.
|
|
27
|
+
*/
|
|
28
|
+
export function getRegistry(): Promise<KernelRegistrySchema>;
|
|
29
|
+
|
|
30
|
+
/**
|
|
31
|
+
* Get registry synchronously (throws if not loaded).
|
|
32
|
+
* Use after awaiting getRegistry() at startup.
|
|
33
|
+
*/
|
|
34
|
+
export function getRegistrySync(): KernelRegistrySchema;
|
|
35
|
+
|
|
36
|
+
/**
|
|
37
|
+
* Clear the cached registry. Useful for hot-reloading.
|
|
38
|
+
*/
|
|
39
|
+
export function clearRegistryCache(): void;
|
|
40
|
+
|
|
41
|
+
/**
|
|
42
|
+
* Get an operation schema by name.
|
|
43
|
+
*/
|
|
44
|
+
export function getOperation(operation: string): OperationSchema | undefined;
|
|
45
|
+
|
|
46
|
+
/**
|
|
47
|
+
* Get a variant schema by operation and variant name.
|
|
48
|
+
*/
|
|
49
|
+
export function getVariant(operation: string, variant: string): KernelVariantSchema | undefined;
|
|
50
|
+
|
|
51
|
+
/**
|
|
52
|
+
* Get all variant names for an operation.
|
|
53
|
+
*/
|
|
54
|
+
export function getVariantNames(operation: string): string[];
|
|
55
|
+
|
|
56
|
+
/**
|
|
57
|
+
* Check if a variant's requirements are met by capabilities.
|
|
58
|
+
*/
|
|
59
|
+
export function isVariantAvailable(
|
|
60
|
+
operation: string,
|
|
61
|
+
variant: string,
|
|
62
|
+
capabilities: RuntimeCapabilities
|
|
63
|
+
): boolean;
|
|
64
|
+
|
|
65
|
+
/**
|
|
66
|
+
* Get all available variants for an operation given capabilities.
|
|
67
|
+
*/
|
|
68
|
+
export function getAvailableVariants(operation: string, capabilities: RuntimeCapabilities): string[];
|
|
69
|
+
|
|
70
|
+
/**
|
|
71
|
+
* Merge base and variant bindings.
|
|
72
|
+
* Variant bindings with matching indices override base bindings.
|
|
73
|
+
*/
|
|
74
|
+
export function mergeBindings(
|
|
75
|
+
base: BindingSchema[],
|
|
76
|
+
override: BindingSchema[] | undefined
|
|
77
|
+
): BindingSchema[];
|
|
78
|
+
|
|
79
|
+
/**
|
|
80
|
+
* Resolve a kernel variant to a complete configuration.
|
|
81
|
+
* Merges base operation config with variant-specific overrides.
|
|
82
|
+
*/
|
|
83
|
+
export function resolveKernelConfig(
|
|
84
|
+
operation: string,
|
|
85
|
+
variant: string
|
|
86
|
+
): ResolvedKernelConfig | null;
|
|
@@ -0,0 +1,103 @@
|
|
|
1
|
+
import { loadJson } from '../../utils/load-json.js';
|
|
2
|
+
|
|
3
|
+
let cachedRegistry = null;
|
|
4
|
+
|
|
5
|
+
let registryUrl = null;
|
|
6
|
+
|
|
7
|
+
export function setRegistryUrl(url) {
|
|
8
|
+
registryUrl = url;
|
|
9
|
+
cachedRegistry = null;
|
|
10
|
+
}
|
|
11
|
+
|
|
12
|
+
export async function getRegistry() {
|
|
13
|
+
if (cachedRegistry) {
|
|
14
|
+
return cachedRegistry;
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
const source = registryUrl || './registry.json';
|
|
18
|
+
cachedRegistry = await loadJson(source, import.meta.url, 'Failed to load kernel registry');
|
|
19
|
+
return cachedRegistry;
|
|
20
|
+
}
|
|
21
|
+
|
|
22
|
+
export function getRegistrySync() {
|
|
23
|
+
if (!cachedRegistry) {
|
|
24
|
+
throw new Error('Kernel registry not loaded. Call await getRegistry() first.');
|
|
25
|
+
}
|
|
26
|
+
return cachedRegistry;
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
export function clearRegistryCache() {
|
|
30
|
+
cachedRegistry = null;
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
export function getOperation(operation) {
|
|
34
|
+
const registry = getRegistrySync();
|
|
35
|
+
return registry.operations[operation];
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
export function getVariant(operation, variant) {
|
|
39
|
+
const op = getOperation(operation);
|
|
40
|
+
return op?.variants[variant];
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
export function getVariantNames(operation) {
|
|
44
|
+
const op = getOperation(operation);
|
|
45
|
+
return op ? Object.keys(op.variants) : [];
|
|
46
|
+
}
|
|
47
|
+
|
|
48
|
+
export function isVariantAvailable(operation, variant, capabilities) {
|
|
49
|
+
const variantSchema = getVariant(operation, variant);
|
|
50
|
+
if (!variantSchema) return false;
|
|
51
|
+
|
|
52
|
+
const requires = variantSchema.requires || [];
|
|
53
|
+
for (const req of requires) {
|
|
54
|
+
if (req === 'shader-f16' && !capabilities.hasF16) return false;
|
|
55
|
+
if (req === 'subgroups' && !capabilities.hasSubgroups) return false;
|
|
56
|
+
if (req === 'subgroups-f16' && (!capabilities.hasSubgroups || !capabilities.hasF16)) return false;
|
|
57
|
+
}
|
|
58
|
+
return true;
|
|
59
|
+
}
|
|
60
|
+
|
|
61
|
+
export function getAvailableVariants(operation, capabilities) {
|
|
62
|
+
return getVariantNames(operation).filter(v => isVariantAvailable(operation, v, capabilities));
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
export function mergeBindings(base, override) {
|
|
66
|
+
if (!override || override.length === 0) {
|
|
67
|
+
return [...base];
|
|
68
|
+
}
|
|
69
|
+
|
|
70
|
+
const result = [...base];
|
|
71
|
+
for (const binding of override) {
|
|
72
|
+
const existingIdx = result.findIndex(b => b.index === binding.index);
|
|
73
|
+
if (existingIdx >= 0) {
|
|
74
|
+
result[existingIdx] = binding;
|
|
75
|
+
} else {
|
|
76
|
+
result.push(binding);
|
|
77
|
+
}
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
return result.sort((a, b) => a.index - b.index);
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
export function resolveKernelConfig(operation, variant) {
|
|
84
|
+
const opSchema = getOperation(operation);
|
|
85
|
+
const variantSchema = getVariant(operation, variant);
|
|
86
|
+
|
|
87
|
+
if (!opSchema || !variantSchema) {
|
|
88
|
+
return null;
|
|
89
|
+
}
|
|
90
|
+
|
|
91
|
+
return {
|
|
92
|
+
operation,
|
|
93
|
+
variant,
|
|
94
|
+
wgsl: variantSchema.wgsl,
|
|
95
|
+
entryPoint: variantSchema.entryPoint,
|
|
96
|
+
workgroup: variantSchema.workgroup,
|
|
97
|
+
requires: variantSchema.requires ?? [],
|
|
98
|
+
bindings: mergeBindings(opSchema.baseBindings, variantSchema.bindingsOverride),
|
|
99
|
+
uniforms: variantSchema.uniformsOverride ?? opSchema.baseUniforms,
|
|
100
|
+
wgslOverrides: variantSchema.wgslOverrides ?? {},
|
|
101
|
+
sharedMemory: variantSchema.sharedMemory ?? 0,
|
|
102
|
+
};
|
|
103
|
+
}
|