@simulatte/doppler 0.1.8 → 0.2.0
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/README.md +3 -97
- package/package.json +5 -160
- package/BRANDING.md +0 -14
- package/CHANGELOG.md +0 -145
- package/LICENSE +0 -201
- package/NOTICE +0 -5
- package/SECURITY.md +0 -19
- package/src/adapters/adapter-manager.d.ts +0 -200
- package/src/adapters/adapter-manager.js +0 -509
- package/src/adapters/adapter-manifest.d.ts +0 -290
- package/src/adapters/adapter-manifest.js +0 -320
- package/src/adapters/adapter-registry.d.ts +0 -192
- package/src/adapters/adapter-registry.js +0 -477
- package/src/adapters/index.d.ts +0 -89
- package/src/adapters/index.js +0 -42
- package/src/adapters/lora-loader.d.ts +0 -105
- package/src/adapters/lora-loader.js +0 -414
- package/src/bootstrap.d.ts +0 -1
- package/src/bootstrap.js +0 -30
- package/src/bridge/extension/background.d.ts +0 -14
- package/src/bridge/extension/background.js +0 -168
- package/src/bridge/extension/manifest.json +0 -34
- package/src/bridge/extension-client.d.ts +0 -114
- package/src/bridge/extension-client.js +0 -409
- package/src/bridge/index.d.ts +0 -69
- package/src/bridge/index.js +0 -53
- package/src/bridge/protocol.d.ts +0 -96
- package/src/bridge/protocol.js +0 -130
- package/src/browser/browser-converter.d.ts +0 -71
- package/src/browser/browser-converter.js +0 -977
- package/src/browser/file-picker.d.ts +0 -63
- package/src/browser/file-picker.js +0 -281
- package/src/browser/gguf-importer.d.ts +0 -136
- package/src/browser/gguf-importer.js +0 -532
- package/src/browser/gguf-parser-browser.d.ts +0 -14
- package/src/browser/gguf-parser-browser.js +0 -17
- package/src/browser/quantization.d.ts +0 -69
- package/src/browser/quantization.js +0 -328
- package/src/browser/safetensors-parser-browser.d.ts +0 -193
- package/src/browser/safetensors-parser-browser.js +0 -347
- package/src/browser/shard-io-browser.d.ts +0 -57
- package/src/browser/shard-io-browser.js +0 -89
- package/src/browser/tensor-source-download.d.ts +0 -27
- package/src/browser/tensor-source-download.js +0 -245
- package/src/browser/tensor-source-file.d.ts +0 -26
- package/src/browser/tensor-source-file.js +0 -53
- package/src/browser/tensor-source-http.d.ts +0 -29
- package/src/browser/tensor-source-http.js +0 -130
- package/src/client/doppler-api.browser.d.ts +0 -1
- package/src/client/doppler-api.browser.js +0 -304
- package/src/client/doppler-api.d.ts +0 -80
- package/src/client/doppler-api.js +0 -314
- package/src/client/doppler-provider/generation.d.ts +0 -25
- package/src/client/doppler-provider/generation.js +0 -126
- package/src/client/doppler-provider/index.d.ts +0 -2
- package/src/client/doppler-provider/index.js +0 -3
- package/src/client/doppler-provider/model-manager.d.ts +0 -71
- package/src/client/doppler-provider/model-manager.js +0 -739
- package/src/client/doppler-provider/provider.d.ts +0 -5
- package/src/client/doppler-provider/provider.js +0 -102
- package/src/client/doppler-provider/source-runtime.d.ts +0 -23
- package/src/client/doppler-provider/source-runtime.js +0 -641
- package/src/client/doppler-provider/types.d.ts +0 -127
- package/src/client/doppler-provider/types.js +0 -17
- package/src/client/doppler-provider.d.ts +0 -46
- package/src/client/doppler-provider.js +0 -36
- package/src/client/doppler-registry.d.ts +0 -23
- package/src/client/doppler-registry.js +0 -88
- package/src/client/doppler-registry.json +0 -24
- package/src/config/README.md +0 -69
- package/src/config/backward-registry-loader.d.ts +0 -3
- package/src/config/backward-registry-loader.js +0 -23
- package/src/config/execution-contract-check.d.ts +0 -82
- package/src/config/execution-contract-check.js +0 -317
- package/src/config/execution-v0-contract-check.d.ts +0 -94
- package/src/config/execution-v0-contract-check.js +0 -349
- package/src/config/execution-v0-graph-contract-check.d.ts +0 -20
- package/src/config/execution-v0-graph-contract-check.js +0 -64
- package/src/config/index.d.ts +0 -63
- package/src/config/index.js +0 -31
- package/src/config/kernel-path-contract-check.d.ts +0 -76
- package/src/config/kernel-path-contract-check.js +0 -507
- package/src/config/kernel-path-loader.d.ts +0 -170
- package/src/config/kernel-path-loader.js +0 -570
- package/src/config/kernels/backward-registry.json +0 -99
- package/src/config/kernels/kernel-ref-digests.d.ts +0 -1
- package/src/config/kernels/kernel-ref-digests.js +0 -226
- package/src/config/kernels/kernel-ref.d.ts +0 -17
- package/src/config/kernels/kernel-ref.js +0 -75
- package/src/config/kernels/moe/gpt-oss.paths.json +0 -49
- package/src/config/kernels/registry.d.ts +0 -86
- package/src/config/kernels/registry.js +0 -116
- package/src/config/kernels/registry.json +0 -7443
- package/src/config/loader.d.ts +0 -57
- package/src/config/loader.js +0 -578
- package/src/config/merge-contract-check.d.ts +0 -16
- package/src/config/merge-contract-check.js +0 -383
- package/src/config/merge-helpers.d.ts +0 -58
- package/src/config/merge-helpers.js +0 -175
- package/src/config/merge.d.ts +0 -143
- package/src/config/merge.js +0 -414
- package/src/config/param-categories.d.ts +0 -17
- package/src/config/param-categories.js +0 -72
- package/src/config/param-validator.d.ts +0 -26
- package/src/config/param-validator.js +0 -280
- package/src/config/platforms/amd-rdna3.json +0 -16
- package/src/config/platforms/apple-m1.json +0 -16
- package/src/config/platforms/apple-m2.json +0 -16
- package/src/config/platforms/apple-m3.json +0 -16
- package/src/config/platforms/generic.json +0 -14
- package/src/config/platforms/loader.d.ts +0 -65
- package/src/config/platforms/loader.js +0 -153
- package/src/config/platforms/nvidia-rtx30.json +0 -16
- package/src/config/platforms/nvidia-rtx40.json +0 -16
- package/src/config/presets/kernel-paths/embeddinggemma-f16-f32a.json +0 -60
- package/src/config/presets/kernel-paths/embeddinggemma-f32-f32a.json +0 -60
- package/src/config/presets/kernel-paths/embeddinggemma-q4k-dequant-f32a.json +0 -60
- package/src/config/presets/kernel-paths/gemma2-f16-f16a.json +0 -61
- package/src/config/presets/kernel-paths/gemma2-f16-f32a.json +0 -60
- package/src/config/presets/kernel-paths/gemma2-q4k-dequant-f16a.json +0 -61
- package/src/config/presets/kernel-paths/gemma2-q4k-dequant-f32a-nosubgroups.json +0 -60
- package/src/config/presets/kernel-paths/gemma2-q4k-fused-f32a.json +0 -57
- package/src/config/presets/kernel-paths/gemma3-f16-fused-f16a-online.json +0 -200
- package/src/config/presets/kernel-paths/gemma3-f16-fused-f32a-online-streamingprefill.json +0 -223
- package/src/config/presets/kernel-paths/gemma3-f16-fused-f32a-online.json +0 -223
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f16a-online.json +0 -60
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f32a-nosubgroups.json +0 -61
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f32a-online.json +0 -61
- package/src/config/presets/kernel-paths/gemma3-q4k-dequant-f32w-f32a-online.json +0 -56
- package/src/config/presets/kernel-paths/lfm2-q4k-dequant-f32a-nosubgroups.json +0 -61
- package/src/config/presets/kernel-paths/lfm2-q4k-dequant-f32a-online.json +0 -61
- package/src/config/presets/kernel-paths/registry.json +0 -138
- package/src/config/presets/models/deepseek.json +0 -20
- package/src/config/presets/models/diffusion.json +0 -10
- package/src/config/presets/models/embeddinggemma.json +0 -74
- package/src/config/presets/models/functiongemma.json +0 -31
- package/src/config/presets/models/gemma2.json +0 -60
- package/src/config/presets/models/gemma3.json +0 -77
- package/src/config/presets/models/gpt-oss.json +0 -68
- package/src/config/presets/models/janus-text.json +0 -27
- package/src/config/presets/models/kimi-k2.json +0 -25
- package/src/config/presets/models/lfm2.json +0 -83
- package/src/config/presets/models/llama3.json +0 -40
- package/src/config/presets/models/mamba.json +0 -34
- package/src/config/presets/models/mixtral.json +0 -37
- package/src/config/presets/models/modernbert.json +0 -32
- package/src/config/presets/models/qwen3.json +0 -49
- package/src/config/presets/models/qwen3_5.json +0 -16
- package/src/config/presets/models/transformer.json +0 -78
- package/src/config/presets/models/translategemma.json +0 -30
- package/src/config/presets/platforms/nvidia-gb200-8gpu.json +0 -45
- package/src/config/presets/platforms/nvidia-gb200-nvl72.json +0 -45
- package/src/config/presets/platforms/nvidia-gh200-nvl2.json +0 -44
- package/src/config/presets/platforms/nvidia-gh200.json +0 -44
- package/src/config/presets/runtime/compute/f16-activations.json +0 -30
- package/src/config/presets/runtime/compute/f16-batched.json +0 -32
- package/src/config/presets/runtime/default.json +0 -101
- package/src/config/presets/runtime/diagnostics/debug-logits.json +0 -53
- package/src/config/presets/runtime/experiments/bench/gemma3-bench-q4k.json +0 -53
- package/src/config/presets/runtime/experiments/debug/gemma3-debug-q4k.json +0 -210
- package/src/config/presets/runtime/experiments/verify/gemma3-verify.json +0 -39
- package/src/config/presets/runtime/kernels/dequant-f16-q4k.json +0 -13
- package/src/config/presets/runtime/kernels/dequant-f32-q4k.json +0 -13
- package/src/config/presets/runtime/kernels/embeddinggemma-q4k-dequant-f32a.json +0 -37
- package/src/config/presets/runtime/kernels/fused-q4k.json +0 -13
- package/src/config/presets/runtime/kernels/gemma2-q4k-dequant-f16a.json +0 -33
- package/src/config/presets/runtime/kernels/gemma2-q4k-dequant-f32a-nosubgroups.json +0 -33
- package/src/config/presets/runtime/kernels/gemma2-q4k-fused-f32a.json +0 -33
- package/src/config/presets/runtime/kernels/safe-q4k.json +0 -13
- package/src/config/presets/runtime/model/gemma2-debug.json +0 -77
- package/src/config/presets/runtime/model/gemma2-pipeline-debug.json +0 -66
- package/src/config/presets/runtime/model/gemma2-pipeline.json +0 -75
- package/src/config/presets/runtime/model/gemma3-layer-probe.json +0 -85
- package/src/config/presets/runtime/model/qwen3-5-layer-probe.json +0 -52
- package/src/config/presets/runtime/model/qwen3-5-linear-attn-debug.json +0 -90
- package/src/config/presets/runtime/modes/bench.json +0 -37
- package/src/config/presets/runtime/modes/debug.json +0 -39
- package/src/config/presets/runtime/modes/default.json +0 -10
- package/src/config/presets/runtime/modes/embedding-bench.json +0 -28
- package/src/config/presets/runtime/modes/embedding.json +0 -54
- package/src/config/presets/runtime/modes/low-memory.json +0 -40
- package/src/config/presets/runtime/modes/production.json +0 -48
- package/src/config/presets/runtime/modes/simulation.json +0 -30
- package/src/config/presets/runtime/modes/trace-layers.json +0 -126
- package/src/config/presets/runtime/platform/metal-apple-q4k.json +0 -11
- package/src/config/quantization-contract-check.d.ts +0 -12
- package/src/config/quantization-contract-check.js +0 -91
- package/src/config/required-inference-fields-contract-check.d.ts +0 -24
- package/src/config/required-inference-fields-contract-check.js +0 -237
- package/src/config/runtime-merge.d.ts +0 -5
- package/src/config/runtime-merge.js +0 -21
- package/src/config/runtime.d.ts +0 -28
- package/src/config/runtime.js +0 -61
- package/src/config/schema/adapter.schema.d.ts +0 -53
- package/src/config/schema/adapter.schema.js +0 -60
- package/src/config/schema/backward-registry.schema.d.ts +0 -14
- package/src/config/schema/backward-registry.schema.js +0 -46
- package/src/config/schema/benchmark.schema.d.ts +0 -54
- package/src/config/schema/benchmark.schema.js +0 -74
- package/src/config/schema/bridge.schema.d.ts +0 -25
- package/src/config/schema/bridge.schema.js +0 -22
- package/src/config/schema/browser-suite-metrics.schema.d.ts +0 -17
- package/src/config/schema/browser-suite-metrics.schema.js +0 -46
- package/src/config/schema/buffer-pool.schema.d.ts +0 -92
- package/src/config/schema/buffer-pool.schema.js +0 -50
- package/src/config/schema/conversion-report.schema.d.ts +0 -40
- package/src/config/schema/conversion-report.schema.js +0 -108
- package/src/config/schema/conversion.schema.d.ts +0 -184
- package/src/config/schema/conversion.schema.js +0 -13
- package/src/config/schema/converter.schema.d.ts +0 -123
- package/src/config/schema/converter.schema.js +0 -136
- package/src/config/schema/debug.schema.d.ts +0 -250
- package/src/config/schema/debug.schema.js +0 -106
- package/src/config/schema/diffusion.schema.d.ts +0 -88
- package/src/config/schema/diffusion.schema.js +0 -62
- package/src/config/schema/distill-training.schema.d.ts +0 -48
- package/src/config/schema/distill-training.schema.js +0 -139
- package/src/config/schema/distribution.schema.d.ts +0 -155
- package/src/config/schema/distribution.schema.js +0 -81
- package/src/config/schema/doppler.schema.d.ts +0 -75
- package/src/config/schema/doppler.schema.js +0 -341
- package/src/config/schema/ecosystem.schema.d.ts +0 -255
- package/src/config/schema/ecosystem.schema.js +0 -534
- package/src/config/schema/emulation.schema.d.ts +0 -351
- package/src/config/schema/emulation.schema.js +0 -299
- package/src/config/schema/energy.schema.d.ts +0 -102
- package/src/config/schema/energy.schema.js +0 -72
- package/src/config/schema/execution-v0.schema.d.ts +0 -187
- package/src/config/schema/execution-v0.schema.js +0 -55
- package/src/config/schema/gpu-cache.schema.d.ts +0 -26
- package/src/config/schema/gpu-cache.schema.js +0 -8
- package/src/config/schema/harness.schema.d.ts +0 -32
- package/src/config/schema/harness.schema.js +0 -20
- package/src/config/schema/hotswap.schema.d.ts +0 -55
- package/src/config/schema/hotswap.schema.js +0 -18
- package/src/config/schema/index.d.ts +0 -885
- package/src/config/schema/index.js +0 -489
- package/src/config/schema/inference-defaults.schema.d.ts +0 -276
- package/src/config/schema/inference-defaults.schema.js +0 -188
- package/src/config/schema/inference.schema.d.ts +0 -298
- package/src/config/schema/inference.schema.js +0 -39
- package/src/config/schema/intent-bundle.schema.d.ts +0 -28
- package/src/config/schema/intent-bundle.schema.js +0 -12
- package/src/config/schema/kernel-path.schema.d.ts +0 -183
- package/src/config/schema/kernel-path.schema.js +0 -9
- package/src/config/schema/kernel-registry.schema.d.ts +0 -199
- package/src/config/schema/kernel-registry.schema.js +0 -46
- package/src/config/schema/kernel-thresholds.schema.d.ts +0 -302
- package/src/config/schema/kernel-thresholds.schema.js +0 -195
- package/src/config/schema/kernel-warmup.schema.d.ts +0 -19
- package/src/config/schema/kernel-warmup.schema.js +0 -5
- package/src/config/schema/kvcache.schema.d.ts +0 -131
- package/src/config/schema/kvcache.schema.js +0 -31
- package/src/config/schema/loading.schema.d.ts +0 -153
- package/src/config/schema/loading.schema.js +0 -84
- package/src/config/schema/lora.schema.d.ts +0 -12
- package/src/config/schema/lora.schema.js +0 -12
- package/src/config/schema/manifest.schema.d.ts +0 -507
- package/src/config/schema/manifest.schema.js +0 -146
- package/src/config/schema/memory-limits.schema.d.ts +0 -107
- package/src/config/schema/memory-limits.schema.js +0 -57
- package/src/config/schema/moe.schema.d.ts +0 -78
- package/src/config/schema/moe.schema.js +0 -31
- package/src/config/schema/platform.schema.d.ts +0 -121
- package/src/config/schema/platform.schema.js +0 -1
- package/src/config/schema/preset.schema.d.ts +0 -124
- package/src/config/schema/preset.schema.js +0 -1
- package/src/config/schema/quantization-defaults.schema.d.ts +0 -34
- package/src/config/schema/quantization-defaults.schema.js +0 -5
- package/src/config/schema/quantization.schema.d.ts +0 -10
- package/src/config/schema/quantization.schema.js +0 -33
- package/src/config/schema/shared-runtime.schema.d.ts +0 -75
- package/src/config/schema/shared-runtime.schema.js +0 -45
- package/src/config/schema/speculative.schema.d.ts +0 -21
- package/src/config/schema/speculative.schema.js +0 -11
- package/src/config/schema/storage.schema.d.ts +0 -123
- package/src/config/schema/storage.schema.js +0 -66
- package/src/config/schema/tooling.schema.d.ts +0 -29
- package/src/config/schema/tooling.schema.js +0 -12
- package/src/config/schema/training-metrics.schema.d.ts +0 -89
- package/src/config/schema/training-metrics.schema.js +0 -374
- package/src/config/schema/training.schema.d.ts +0 -88
- package/src/config/schema/training.schema.js +0 -106
- package/src/config/schema/tuner.schema.d.ts +0 -39
- package/src/config/schema/tuner.schema.js +0 -13
- package/src/config/schema/ul-training.schema.d.ts +0 -61
- package/src/config/schema/ul-training.schema.js +0 -140
- package/src/config/schema/units.schema.d.ts +0 -27
- package/src/config/schema/units.schema.js +0 -26
- package/src/config/training-defaults.d.ts +0 -24
- package/src/config/training-defaults.js +0 -99
- package/src/converter/conversion-plan.d.ts +0 -64
- package/src/converter/conversion-plan.js +0 -565
- package/src/converter/core.d.ts +0 -264
- package/src/converter/core.js +0 -1374
- package/src/converter/execution-v0-manifest.d.ts +0 -15
- package/src/converter/execution-v0-manifest.js +0 -149
- package/src/converter/index.d.ts +0 -99
- package/src/converter/index.js +0 -60
- package/src/converter/manifest-inference.d.ts +0 -20
- package/src/converter/manifest-inference.js +0 -513
- package/src/converter/parsers/diffusion.d.ts +0 -50
- package/src/converter/parsers/diffusion.js +0 -327
- package/src/converter/parsers/gguf.d.ts +0 -22
- package/src/converter/parsers/gguf.js +0 -46
- package/src/converter/parsers/index.d.ts +0 -21
- package/src/converter/parsers/index.js +0 -12
- package/src/converter/parsers/transformer.d.ts +0 -16
- package/src/converter/parsers/transformer.js +0 -29
- package/src/converter/quantization-info.d.ts +0 -37
- package/src/converter/quantization-info.js +0 -422
- package/src/converter/quantizer.d.ts +0 -96
- package/src/converter/quantizer.js +0 -429
- package/src/converter/rope-config.d.ts +0 -15
- package/src/converter/rope-config.js +0 -262
- package/src/converter/shard-packer.d.ts +0 -138
- package/src/converter/shard-packer.js +0 -425
- package/src/converter/tokenizer-utils.d.ts +0 -12
- package/src/converter/tokenizer-utils.js +0 -104
- package/src/debug/config.d.ts +0 -78
- package/src/debug/config.js +0 -347
- package/src/debug/history.d.ts +0 -65
- package/src/debug/history.js +0 -71
- package/src/debug/index.d.ts +0 -268
- package/src/debug/index.js +0 -192
- package/src/debug/log.d.ts +0 -46
- package/src/debug/log.js +0 -132
- package/src/debug/perf.d.ts +0 -33
- package/src/debug/perf.js +0 -51
- package/src/debug/reference/README.md +0 -114
- package/src/debug/reference/hf_attn_debug.py +0 -114
- package/src/debug/reference/hf_embed_check.py +0 -89
- package/src/debug/reference/hf_layer_out.py +0 -100
- package/src/debug/reference/hf_qwen35_linear_attn_debug.py +0 -268
- package/src/debug/reference/hf_rope_check.py +0 -116
- package/src/debug/reference/hf_weights.py +0 -75
- package/src/debug/signals.d.ts +0 -63
- package/src/debug/signals.js +0 -39
- package/src/debug/stats.d.ts +0 -47
- package/src/debug/stats.js +0 -160
- package/src/debug/tensor.d.ts +0 -125
- package/src/debug/tensor.js +0 -268
- package/src/debug/trace.d.ts +0 -17
- package/src/debug/trace.js +0 -167
- package/src/diffusion/image-regression.d.ts +0 -31
- package/src/diffusion/image-regression.js +0 -107
- package/src/diffusion/index.d.ts +0 -8
- package/src/diffusion/index.js +0 -8
- package/src/distribution/p2p-control-plane.d.ts +0 -52
- package/src/distribution/p2p-control-plane.js +0 -272
- package/src/distribution/p2p-observability.d.ts +0 -116
- package/src/distribution/p2p-observability.js +0 -303
- package/src/distribution/p2p-transport-contract.d.ts +0 -57
- package/src/distribution/p2p-transport-contract.js +0 -310
- package/src/distribution/p2p-webrtc-browser.d.ts +0 -37
- package/src/distribution/p2p-webrtc-browser.js +0 -454
- package/src/distribution/shard-delivery.d.ts +0 -251
- package/src/distribution/shard-delivery.js +0 -2152
- package/src/energy/index.d.ts +0 -2
- package/src/energy/index.js +0 -2
- package/src/errors/doppler-error.d.ts +0 -21
- package/src/errors/doppler-error.js +0 -25
- package/src/errors/index.d.ts +0 -1
- package/src/errors/index.js +0 -1
- package/src/formats/gguf/index.d.ts +0 -8
- package/src/formats/gguf/index.js +0 -4
- package/src/formats/gguf/types.d.ts +0 -137
- package/src/formats/gguf/types.js +0 -460
- package/src/formats/index.d.ts +0 -51
- package/src/formats/index.js +0 -13
- package/src/formats/rdrr/classification.d.ts +0 -39
- package/src/formats/rdrr/classification.js +0 -275
- package/src/formats/rdrr/groups.d.ts +0 -35
- package/src/formats/rdrr/groups.js +0 -73
- package/src/formats/rdrr/index.d.ts +0 -25
- package/src/formats/rdrr/index.js +0 -19
- package/src/formats/rdrr/manifest.d.ts +0 -32
- package/src/formats/rdrr/manifest.js +0 -108
- package/src/formats/rdrr/parsing.d.ts +0 -27
- package/src/formats/rdrr/parsing.js +0 -151
- package/src/formats/rdrr/tensor-config-validator.d.ts +0 -42
- package/src/formats/rdrr/tensor-config-validator.js +0 -156
- package/src/formats/rdrr/types.d.ts +0 -201
- package/src/formats/rdrr/types.js +0 -16
- package/src/formats/rdrr/validation.d.ts +0 -9
- package/src/formats/rdrr/validation.js +0 -213
- package/src/formats/safetensors/index.d.ts +0 -8
- package/src/formats/safetensors/index.js +0 -4
- package/src/formats/safetensors/types.d.ts +0 -67
- package/src/formats/safetensors/types.js +0 -102
- package/src/formats/tokenizer/index.d.ts +0 -5
- package/src/formats/tokenizer/index.js +0 -3
- package/src/formats/tokenizer/types.d.ts +0 -9
- package/src/formats/tokenizer/types.js +0 -22
- package/src/generation/index.d.ts +0 -18
- package/src/generation/index.js +0 -12
- package/src/gpu/command-recorder.d.ts +0 -175
- package/src/gpu/command-recorder.js +0 -498
- package/src/gpu/device.d.ts +0 -142
- package/src/gpu/device.js +0 -462
- package/src/gpu/kernel-runtime.d.ts +0 -20
- package/src/gpu/kernel-runtime.js +0 -37
- package/src/gpu/kernel-selection-cache.d.ts +0 -13
- package/src/gpu/kernel-selection-cache.js +0 -13
- package/src/gpu/kernel-selection-log.d.ts +0 -12
- package/src/gpu/kernel-selection-log.js +0 -28
- package/src/gpu/kernel-selector.d.ts +0 -11
- package/src/gpu/kernel-selector.js +0 -10
- package/src/gpu/kernel-tuner/benchmarks.d.ts +0 -144
- package/src/gpu/kernel-tuner/benchmarks.js +0 -902
- package/src/gpu/kernel-tuner/cache.d.ts +0 -55
- package/src/gpu/kernel-tuner/cache.js +0 -133
- package/src/gpu/kernel-tuner/index.d.ts +0 -59
- package/src/gpu/kernel-tuner/index.js +0 -38
- package/src/gpu/kernel-tuner/tuner.d.ts +0 -82
- package/src/gpu/kernel-tuner/tuner.js +0 -247
- package/src/gpu/kernel-tuner/types.d.ts +0 -101
- package/src/gpu/kernel-tuner/types.js +0 -4
- package/src/gpu/kernel-tuner.d.ts +0 -33
- package/src/gpu/kernel-tuner.js +0 -12
- package/src/gpu/kernels/README.md +0 -127
- package/src/gpu/kernels/attention.d.ts +0 -236
- package/src/gpu/kernels/attention.js +0 -1438
- package/src/gpu/kernels/attention.wgsl +0 -249
- package/src/gpu/kernels/attention_bdpa_decode_f16.wgsl +0 -246
- package/src/gpu/kernels/attention_decode.wgsl +0 -233
- package/src/gpu/kernels/attention_decode_chunked_f16.wgsl +0 -183
- package/src/gpu/kernels/attention_decode_chunked_f16kv.wgsl +0 -208
- package/src/gpu/kernels/attention_decode_f16.wgsl +0 -202
- package/src/gpu/kernels/attention_decode_f16kv.wgsl +0 -224
- package/src/gpu/kernels/attention_decode_online_f16.wgsl +0 -223
- package/src/gpu/kernels/attention_decode_online_f16kv.wgsl +0 -225
- package/src/gpu/kernels/attention_decode_optimized.wgsl +0 -445
- package/src/gpu/kernels/attention_decode_paged_f16.wgsl +0 -172
- package/src/gpu/kernels/attention_decode_paged_f16kv.wgsl +0 -174
- package/src/gpu/kernels/attention_decode_subgroup.wgsl +0 -233
- package/src/gpu/kernels/attention_decode_tiered_f16.wgsl +0 -218
- package/src/gpu/kernels/attention_decode_tiered_f16kv.wgsl +0 -220
- package/src/gpu/kernels/attention_decode_tiered_int4_f16kv.wgsl +0 -242
- package/src/gpu/kernels/attention_decode_tiered_int8_f16kv.wgsl +0 -242
- package/src/gpu/kernels/attention_f16.wgsl +0 -214
- package/src/gpu/kernels/attention_f16kv.wgsl +0 -242
- package/src/gpu/kernels/attention_small.wgsl +0 -260
- package/src/gpu/kernels/attention_small_f16.wgsl +0 -240
- package/src/gpu/kernels/attention_small_f16kv.wgsl +0 -266
- package/src/gpu/kernels/attention_streaming.wgsl +0 -149
- package/src/gpu/kernels/attention_streaming_f16.wgsl +0 -147
- package/src/gpu/kernels/attention_streaming_f16kv.wgsl +0 -151
- package/src/gpu/kernels/backward/adam.d.ts +0 -28
- package/src/gpu/kernels/backward/adam.js +0 -203
- package/src/gpu/kernels/backward/adam.wgsl +0 -50
- package/src/gpu/kernels/backward/attention_backward.d.ts +0 -22
- package/src/gpu/kernels/backward/attention_backward.js +0 -364
- package/src/gpu/kernels/backward/attention_backward.wgsl +0 -49
- package/src/gpu/kernels/backward/bias_add_backward.d.ts +0 -17
- package/src/gpu/kernels/backward/bias_add_backward.js +0 -24
- package/src/gpu/kernels/backward/bias_add_backward.wgsl +0 -33
- package/src/gpu/kernels/backward/conv2d_backward.d.ts +0 -31
- package/src/gpu/kernels/backward/conv2d_backward.js +0 -148
- package/src/gpu/kernels/backward/conv2d_backward_input.wgsl +0 -83
- package/src/gpu/kernels/backward/conv2d_backward_weight.wgsl +0 -70
- package/src/gpu/kernels/backward/cross_entropy_backward.d.ts +0 -23
- package/src/gpu/kernels/backward/cross_entropy_backward.js +0 -29
- package/src/gpu/kernels/backward/cross_entropy_backward.wgsl +0 -39
- package/src/gpu/kernels/backward/embed_backward.d.ts +0 -29
- package/src/gpu/kernels/backward/embed_backward.js +0 -118
- package/src/gpu/kernels/backward/embed_backward.wgsl +0 -73
- package/src/gpu/kernels/backward/gelu_backward.d.ts +0 -16
- package/src/gpu/kernels/backward/gelu_backward.js +0 -39
- package/src/gpu/kernels/backward/gelu_backward.wgsl +0 -38
- package/src/gpu/kernels/backward/groupnorm_backward.d.ts +0 -24
- package/src/gpu/kernels/backward/groupnorm_backward.js +0 -29
- package/src/gpu/kernels/backward/groupnorm_backward.wgsl +0 -143
- package/src/gpu/kernels/backward/index.d.ts +0 -17
- package/src/gpu/kernels/backward/index.js +0 -23
- package/src/gpu/kernels/backward/layernorm_backward.d.ts +0 -22
- package/src/gpu/kernels/backward/layernorm_backward.js +0 -135
- package/src/gpu/kernels/backward/layernorm_backward.wgsl +0 -194
- package/src/gpu/kernels/backward/matmul_backward.d.ts +0 -32
- package/src/gpu/kernels/backward/matmul_backward.js +0 -124
- package/src/gpu/kernels/backward/matmul_backward.wgsl +0 -90
- package/src/gpu/kernels/backward/matmul_transpose_a.wgsl +0 -84
- package/src/gpu/kernels/backward/pixel_shuffle_backward.d.ts +0 -22
- package/src/gpu/kernels/backward/pixel_shuffle_backward.js +0 -30
- package/src/gpu/kernels/backward/pixel_shuffle_backward.wgsl +0 -54
- package/src/gpu/kernels/backward/rmsnorm_backward.d.ts +0 -24
- package/src/gpu/kernels/backward/rmsnorm_backward.js +0 -101
- package/src/gpu/kernels/backward/rmsnorm_backward.wgsl +0 -78
- package/src/gpu/kernels/backward/rope_backward.d.ts +0 -25
- package/src/gpu/kernels/backward/rope_backward.js +0 -109
- package/src/gpu/kernels/backward/rope_backward.wgsl +0 -59
- package/src/gpu/kernels/backward/scale_backward.d.ts +0 -16
- package/src/gpu/kernels/backward/scale_backward.js +0 -84
- package/src/gpu/kernels/backward/scale_backward.wgsl +0 -27
- package/src/gpu/kernels/backward/silu_backward.d.ts +0 -16
- package/src/gpu/kernels/backward/silu_backward.js +0 -39
- package/src/gpu/kernels/backward/silu_backward.wgsl +0 -31
- package/src/gpu/kernels/backward/softmax_backward.d.ts +0 -16
- package/src/gpu/kernels/backward/softmax_backward.js +0 -43
- package/src/gpu/kernels/backward/softmax_backward.wgsl +0 -44
- package/src/gpu/kernels/backward/upsample2d_backward.d.ts +0 -21
- package/src/gpu/kernels/backward/upsample2d_backward.js +0 -30
- package/src/gpu/kernels/backward/upsample2d_backward.wgsl +0 -59
- package/src/gpu/kernels/backward/utils.d.ts +0 -45
- package/src/gpu/kernels/backward/utils.js +0 -371
- package/src/gpu/kernels/bf16_to_f16.wgsl +0 -54
- package/src/gpu/kernels/bf16_to_f32.wgsl +0 -70
- package/src/gpu/kernels/bias_add.wgsl +0 -42
- package/src/gpu/kernels/bias_add_f16.wgsl +0 -47
- package/src/gpu/kernels/cast.d.ts +0 -67
- package/src/gpu/kernels/cast.js +0 -464
- package/src/gpu/kernels/cast_f16_to_f32.wgsl +0 -31
- package/src/gpu/kernels/cast_f32_to_f16.wgsl +0 -36
- package/src/gpu/kernels/check-finiteness.d.ts +0 -15
- package/src/gpu/kernels/check-finiteness.js +0 -149
- package/src/gpu/kernels/check-stop.d.ts +0 -31
- package/src/gpu/kernels/check-stop.js +0 -170
- package/src/gpu/kernels/clamp.d.ts +0 -22
- package/src/gpu/kernels/clamp.js +0 -42
- package/src/gpu/kernels/clamp.wgsl +0 -24
- package/src/gpu/kernels/constants.d.ts +0 -168
- package/src/gpu/kernels/constants.js +0 -129
- package/src/gpu/kernels/conv2d.d.ts +0 -34
- package/src/gpu/kernels/conv2d.js +0 -91
- package/src/gpu/kernels/conv2d.wgsl +0 -70
- package/src/gpu/kernels/conv2d_f16.wgsl +0 -72
- package/src/gpu/kernels/cross_entropy_loss.d.ts +0 -21
- package/src/gpu/kernels/cross_entropy_loss.js +0 -60
- package/src/gpu/kernels/cross_entropy_loss.wgsl +0 -39
- package/src/gpu/kernels/depthwise_conv2d.d.ts +0 -29
- package/src/gpu/kernels/depthwise_conv2d.js +0 -109
- package/src/gpu/kernels/depthwise_conv2d.wgsl +0 -55
- package/src/gpu/kernels/depthwise_conv2d_f16.wgsl +0 -59
- package/src/gpu/kernels/dequant.d.ts +0 -108
- package/src/gpu/kernels/dequant.js +0 -576
- package/src/gpu/kernels/dequant_f16_out.wgsl +0 -151
- package/src/gpu/kernels/dequant_f16_out_vec4.wgsl +0 -149
- package/src/gpu/kernels/dequant_f16_rowwise.wgsl +0 -139
- package/src/gpu/kernels/dequant_f32_rowwise.wgsl +0 -133
- package/src/gpu/kernels/dequant_mxfp4.wgsl +0 -120
- package/src/gpu/kernels/dequant_mxfp4_expert.wgsl +0 -129
- package/src/gpu/kernels/dequant_mxfp4_expert_f16.wgsl +0 -105
- package/src/gpu/kernels/dequant_mxfp4_vec4.wgsl +0 -116
- package/src/gpu/kernels/dequant_q6k.wgsl +0 -140
- package/src/gpu/kernels/dequant_q8_0.wgsl +0 -98
- package/src/gpu/kernels/dequant_shared.wgsl +0 -202
- package/src/gpu/kernels/dequant_shared_vec4.wgsl +0 -153
- package/src/gpu/kernels/dequant_subgroup.wgsl +0 -202
- package/src/gpu/kernels/dispatch.d.ts +0 -157
- package/src/gpu/kernels/dispatch.js +0 -235
- package/src/gpu/kernels/energy.d.ts +0 -113
- package/src/gpu/kernels/energy.js +0 -448
- package/src/gpu/kernels/energy_eval.wgsl +0 -26
- package/src/gpu/kernels/energy_eval_f16.wgsl +0 -30
- package/src/gpu/kernels/energy_quintel_grad.wgsl +0 -92
- package/src/gpu/kernels/energy_quintel_grad_f16.wgsl +0 -96
- package/src/gpu/kernels/energy_quintel_reduce.wgsl +0 -112
- package/src/gpu/kernels/energy_quintel_reduce_f16.wgsl +0 -116
- package/src/gpu/kernels/energy_quintel_update.wgsl +0 -92
- package/src/gpu/kernels/energy_quintel_update_f16.wgsl +0 -96
- package/src/gpu/kernels/energy_update.wgsl +0 -25
- package/src/gpu/kernels/energy_update_f16.wgsl +0 -30
- package/src/gpu/kernels/feature-check.d.ts +0 -42
- package/src/gpu/kernels/feature-check.js +0 -70
- package/src/gpu/kernels/fused_ffn.d.ts +0 -65
- package/src/gpu/kernels/fused_ffn.js +0 -337
- package/src/gpu/kernels/fused_ffn.wgsl +0 -420
- package/src/gpu/kernels/fused_ffn_f16.wgsl +0 -213
- package/src/gpu/kernels/fused_ffn_q4k.wgsl +0 -375
- package/src/gpu/kernels/fused_matmul_q4.wgsl +0 -404
- package/src/gpu/kernels/fused_matmul_q4_batched.wgsl +0 -194
- package/src/gpu/kernels/fused_matmul_q4_batched_f16.wgsl +0 -170
- package/src/gpu/kernels/fused_matmul_q4_batched_f16a.wgsl +0 -154
- package/src/gpu/kernels/fused_matmul_q4_f16a.wgsl +0 -219
- package/src/gpu/kernels/fused_matmul_q4_multicol_f16.wgsl +0 -216
- package/src/gpu/kernels/fused_matmul_q4_multicol_f16a.wgsl +0 -204
- package/src/gpu/kernels/fused_matmul_residual.d.ts +0 -46
- package/src/gpu/kernels/fused_matmul_residual.js +0 -175
- package/src/gpu/kernels/fused_matmul_rmsnorm.d.ts +0 -64
- package/src/gpu/kernels/fused_matmul_rmsnorm.js +0 -290
- package/src/gpu/kernels/fused_matmul_rmsnorm.wgsl +0 -324
- package/src/gpu/kernels/fused_matmul_rmsnorm_f16.wgsl +0 -303
- package/src/gpu/kernels/fused_swiglu.wgsl +0 -63
- package/src/gpu/kernels/fused_swiglu_f16.wgsl +0 -57
- package/src/gpu/kernels/gather.d.ts +0 -64
- package/src/gpu/kernels/gather.js +0 -137
- package/src/gpu/kernels/gather.wgsl +0 -61
- package/src/gpu/kernels/gather_f16.wgsl +0 -65
- package/src/gpu/kernels/gather_f16_f16_out.wgsl +0 -55
- package/src/gpu/kernels/gather_f16_out.wgsl +0 -55
- package/src/gpu/kernels/gather_f16_vec4.wgsl +0 -76
- package/src/gpu/kernels/gather_f16_vec4_f16_out.wgsl +0 -68
- package/src/gpu/kernels/gather_vec4.wgsl +0 -74
- package/src/gpu/kernels/gather_vec4_f16_out.wgsl +0 -68
- package/src/gpu/kernels/gelu.d.ts +0 -33
- package/src/gpu/kernels/gelu.js +0 -55
- package/src/gpu/kernels/gelu.wgsl +0 -64
- package/src/gpu/kernels/gelu_f16.wgsl +0 -66
- package/src/gpu/kernels/gptoss_mxfp4_expert_fused.wgsl +0 -127
- package/src/gpu/kernels/gptoss_router_topk.wgsl +0 -119
- package/src/gpu/kernels/grouped_pointwise_conv2d.d.ts +0 -27
- package/src/gpu/kernels/grouped_pointwise_conv2d.js +0 -103
- package/src/gpu/kernels/grouped_pointwise_conv2d.wgsl +0 -44
- package/src/gpu/kernels/grouped_pointwise_conv2d_f16.wgsl +0 -48
- package/src/gpu/kernels/groupnorm.d.ts +0 -31
- package/src/gpu/kernels/groupnorm.js +0 -102
- package/src/gpu/kernels/groupnorm_apply.wgsl +0 -41
- package/src/gpu/kernels/groupnorm_apply_f16.wgsl +0 -46
- package/src/gpu/kernels/groupnorm_stats.wgsl +0 -76
- package/src/gpu/kernels/groupnorm_stats_f16.wgsl +0 -79
- package/src/gpu/kernels/index.d.ts +0 -374
- package/src/gpu/kernels/index.js +0 -315
- package/src/gpu/kernels/kernel-base.d.ts +0 -33
- package/src/gpu/kernels/kernel-base.js +0 -46
- package/src/gpu/kernels/kernel-configs.d.ts +0 -65
- package/src/gpu/kernels/kernel-configs.js +0 -50
- package/src/gpu/kernels/kernel-tuning.d.ts +0 -42
- package/src/gpu/kernels/kernel-tuning.js +0 -149
- package/src/gpu/kernels/kv-quantize.d.ts +0 -37
- package/src/gpu/kernels/kv-quantize.js +0 -141
- package/src/gpu/kernels/kv_quantize_int4.wgsl +0 -119
- package/src/gpu/kernels/kv_quantize_int8.wgsl +0 -119
- package/src/gpu/kernels/layernorm.d.ts +0 -37
- package/src/gpu/kernels/layernorm.js +0 -96
- package/src/gpu/kernels/layernorm.wgsl +0 -121
- package/src/gpu/kernels/layernorm_f16.wgsl +0 -103
- package/src/gpu/kernels/linear-attention-core.d.ts +0 -39
- package/src/gpu/kernels/linear-attention-core.js +0 -535
- package/src/gpu/kernels/logit-merge.d.ts +0 -110
- package/src/gpu/kernels/logit-merge.js +0 -394
- package/src/gpu/kernels/matmul-dispatch.d.ts +0 -38
- package/src/gpu/kernels/matmul-dispatch.js +0 -155
- package/src/gpu/kernels/matmul-selection.d.ts +0 -87
- package/src/gpu/kernels/matmul-selection.js +0 -517
- package/src/gpu/kernels/matmul.d.ts +0 -111
- package/src/gpu/kernels/matmul.js +0 -315
- package/src/gpu/kernels/matmul_f16.wgsl +0 -170
- package/src/gpu/kernels/matmul_f16_tiled.wgsl +0 -165
- package/src/gpu/kernels/matmul_f16w_f32a.wgsl +0 -89
- package/src/gpu/kernels/matmul_f16w_f32a_tiled.wgsl +0 -154
- package/src/gpu/kernels/matmul_f32.wgsl +0 -100
- package/src/gpu/kernels/matmul_gemv.wgsl +0 -80
- package/src/gpu/kernels/matmul_gemv_f16a.wgsl +0 -81
- package/src/gpu/kernels/matmul_gemv_residual.wgsl +0 -119
- package/src/gpu/kernels/matmul_gemv_residual_f16.wgsl +0 -78
- package/src/gpu/kernels/matmul_gemv_subgroup.wgsl +0 -345
- package/src/gpu/kernels/matmul_gemv_subgroup_f16a.wgsl +0 -514
- package/src/gpu/kernels/modulate.d.ts +0 -29
- package/src/gpu/kernels/modulate.js +0 -57
- package/src/gpu/kernels/modulate.wgsl +0 -40
- package/src/gpu/kernels/modulate_f16.wgsl +0 -43
- package/src/gpu/kernels/moe.d.ts +0 -164
- package/src/gpu/kernels/moe.js +0 -542
- package/src/gpu/kernels/moe_gather.wgsl +0 -170
- package/src/gpu/kernels/moe_gather_f16.wgsl +0 -82
- package/src/gpu/kernels/moe_gather_vec4.wgsl +0 -74
- package/src/gpu/kernels/moe_offsets.wgsl +0 -48
- package/src/gpu/kernels/pipeline-cache.d.ts +0 -88
- package/src/gpu/kernels/pipeline-cache.js +0 -305
- package/src/gpu/kernels/pixel_shuffle.d.ts +0 -27
- package/src/gpu/kernels/pixel_shuffle.js +0 -57
- package/src/gpu/kernels/pixel_shuffle.wgsl +0 -43
- package/src/gpu/kernels/pixel_shuffle_f16.wgsl +0 -46
- package/src/gpu/kernels/relu.d.ts +0 -18
- package/src/gpu/kernels/relu.js +0 -66
- package/src/gpu/kernels/relu.wgsl +0 -22
- package/src/gpu/kernels/relu_f16.wgsl +0 -24
- package/src/gpu/kernels/repeat_channels.d.ts +0 -21
- package/src/gpu/kernels/repeat_channels.js +0 -68
- package/src/gpu/kernels/repeat_channels.wgsl +0 -28
- package/src/gpu/kernels/repeat_channels_f16.wgsl +0 -30
- package/src/gpu/kernels/residual.d.ts +0 -74
- package/src/gpu/kernels/residual.js +0 -173
- package/src/gpu/kernels/residual.wgsl +0 -56
- package/src/gpu/kernels/residual_f16.wgsl +0 -36
- package/src/gpu/kernels/residual_f16_vec4.wgsl +0 -48
- package/src/gpu/kernels/residual_vec4.wgsl +0 -47
- package/src/gpu/kernels/rmsnorm.d.ts +0 -53
- package/src/gpu/kernels/rmsnorm.js +0 -215
- package/src/gpu/kernels/rmsnorm.wgsl +0 -425
- package/src/gpu/kernels/rmsnorm_f16.wgsl +0 -172
- package/src/gpu/kernels/rope.d.ts +0 -50
- package/src/gpu/kernels/rope.js +0 -66
- package/src/gpu/kernels/rope.wgsl +0 -344
- package/src/gpu/kernels/rope_f16.wgsl +0 -271
- package/src/gpu/kernels/rule-matcher.d.ts +0 -30
- package/src/gpu/kernels/rule-matcher.js +0 -42
- package/src/gpu/kernels/rule-registry.d.ts +0 -7
- package/src/gpu/kernels/rule-registry.js +0 -41
- package/src/gpu/kernels/sample.d.ts +0 -75
- package/src/gpu/kernels/sample.js +0 -567
- package/src/gpu/kernels/sample.wgsl +0 -377
- package/src/gpu/kernels/sample_f16.wgsl +0 -331
- package/src/gpu/kernels/sana_linear_attention.d.ts +0 -27
- package/src/gpu/kernels/sana_linear_attention.js +0 -129
- package/src/gpu/kernels/sana_linear_attention_apply.wgsl +0 -43
- package/src/gpu/kernels/sana_linear_attention_apply_f16.wgsl +0 -46
- package/src/gpu/kernels/sana_linear_attention_summary.wgsl +0 -51
- package/src/gpu/kernels/sana_linear_attention_summary_f16.wgsl +0 -53
- package/src/gpu/kernels/scale.d.ts +0 -35
- package/src/gpu/kernels/scale.js +0 -44
- package/src/gpu/kernels/scale.wgsl +0 -38
- package/src/gpu/kernels/scatter_add.wgsl +0 -88
- package/src/gpu/kernels/scatter_add_dynamic.wgsl +0 -59
- package/src/gpu/kernels/scatter_add_dynamic_f16.wgsl +0 -52
- package/src/gpu/kernels/scatter_add_dynamic_f16_weights.wgsl +0 -50
- package/src/gpu/kernels/scatter_add_vec4.wgsl +0 -70
- package/src/gpu/kernels/shader-cache.d.ts +0 -56
- package/src/gpu/kernels/shader-cache.js +0 -208
- package/src/gpu/kernels/silu.d.ts +0 -76
- package/src/gpu/kernels/silu.js +0 -406
- package/src/gpu/kernels/silu.wgsl +0 -109
- package/src/gpu/kernels/silu_f16.wgsl +0 -108
- package/src/gpu/kernels/softmax.d.ts +0 -57
- package/src/gpu/kernels/softmax.js +0 -125
- package/src/gpu/kernels/softmax.wgsl +0 -388
- package/src/gpu/kernels/softmax_subgroup.wgsl +0 -175
- package/src/gpu/kernels/split_qg.d.ts +0 -50
- package/src/gpu/kernels/split_qg.js +0 -46
- package/src/gpu/kernels/split_qg.wgsl +0 -58
- package/src/gpu/kernels/split_qg_f16.wgsl +0 -62
- package/src/gpu/kernels/split_qkv.d.ts +0 -51
- package/src/gpu/kernels/split_qkv.js +0 -51
- package/src/gpu/kernels/split_qkv.wgsl +0 -71
- package/src/gpu/kernels/split_qkv_f16.wgsl +0 -75
- package/src/gpu/kernels/topk.wgsl +0 -243
- package/src/gpu/kernels/topk_f16.wgsl +0 -108
- package/src/gpu/kernels/topk_f16_weights.wgsl +0 -101
- package/src/gpu/kernels/transpose.d.ts +0 -21
- package/src/gpu/kernels/transpose.js +0 -51
- package/src/gpu/kernels/transpose.wgsl +0 -33
- package/src/gpu/kernels/types.d.ts +0 -21
- package/src/gpu/kernels/types.js +0 -4
- package/src/gpu/kernels/uniform-utils.d.ts +0 -48
- package/src/gpu/kernels/uniform-utils.js +0 -94
- package/src/gpu/kernels/upsample2d.d.ts +0 -25
- package/src/gpu/kernels/upsample2d.js +0 -67
- package/src/gpu/kernels/upsample2d.wgsl +0 -34
- package/src/gpu/kernels/upsample2d_f16.wgsl +0 -38
- package/src/gpu/kernels/utils.d.ts +0 -106
- package/src/gpu/kernels/utils.js +0 -246
- package/src/gpu/multi-model-recorder.d.ts +0 -21
- package/src/gpu/multi-model-recorder.js +0 -31
- package/src/gpu/partitioned-buffer-pool.d.ts +0 -28
- package/src/gpu/partitioned-buffer-pool.js +0 -57
- package/src/gpu/perf-guards.d.ts +0 -25
- package/src/gpu/perf-guards.js +0 -133
- package/src/gpu/profiler.d.ts +0 -114
- package/src/gpu/profiler.js +0 -396
- package/src/gpu/readback-utils.d.ts +0 -16
- package/src/gpu/readback-utils.js +0 -41
- package/src/gpu/submit-tracker.d.ts +0 -111
- package/src/gpu/submit-tracker.js +0 -242
- package/src/gpu/tensor.d.ts +0 -69
- package/src/gpu/tensor.js +0 -75
- package/src/gpu/uniform-cache.d.ts +0 -109
- package/src/gpu/uniform-cache.js +0 -263
- package/src/gpu/weight-buffer.d.ts +0 -115
- package/src/gpu/weight-buffer.js +0 -118
- package/src/hotswap/intent-bundle.d.ts +0 -37
- package/src/hotswap/intent-bundle.js +0 -129
- package/src/hotswap/manifest.d.ts +0 -42
- package/src/hotswap/manifest.js +0 -124
- package/src/hotswap/runtime.d.ts +0 -31
- package/src/hotswap/runtime.js +0 -150
- package/src/index-browser.d.ts +0 -92
- package/src/index-browser.js +0 -68
- package/src/index-internal.d.ts +0 -2
- package/src/index-internal.js +0 -2
- package/src/index.d.ts +0 -103
- package/src/index.js +0 -76
- package/src/inference/README.md +0 -593
- package/src/inference/browser-harness-contract-helpers.d.ts +0 -5
- package/src/inference/browser-harness-contract-helpers.js +0 -28
- package/src/inference/browser-harness-diffusion-energy-suites.d.ts +0 -2
- package/src/inference/browser-harness-diffusion-energy-suites.js +0 -269
- package/src/inference/browser-harness-model-helpers.d.ts +0 -16
- package/src/inference/browser-harness-model-helpers.js +0 -217
- package/src/inference/browser-harness-report-helpers.d.ts +0 -7
- package/src/inference/browser-harness-report-helpers.js +0 -42
- package/src/inference/browser-harness-runtime-helpers.d.ts +0 -61
- package/src/inference/browser-harness-runtime-helpers.js +0 -415
- package/src/inference/browser-harness-suite-helpers.d.ts +0 -28
- package/src/inference/browser-harness-suite-helpers.js +0 -268
- package/src/inference/browser-harness-text-helpers.d.ts +0 -27
- package/src/inference/browser-harness-text-helpers.js +0 -788
- package/src/inference/browser-harness.d.ts +0 -242
- package/src/inference/browser-harness.js +0 -990
- package/src/inference/decode-buffers.d.ts +0 -108
- package/src/inference/decode-buffers.js +0 -181
- package/src/inference/decode-ring.d.ts +0 -52
- package/src/inference/decode-ring.js +0 -273
- package/src/inference/expert-router.d.ts +0 -27
- package/src/inference/expert-router.js +0 -55
- package/src/inference/functiongemma.d.ts +0 -15
- package/src/inference/functiongemma.js +0 -1
- package/src/inference/kv-cache/base.d.ts +0 -150
- package/src/inference/kv-cache/base.js +0 -1083
- package/src/inference/kv-cache/basis-decomposed-paged.d.ts +0 -50
- package/src/inference/kv-cache/basis-decomposed-paged.js +0 -276
- package/src/inference/kv-cache/index.d.ts +0 -35
- package/src/inference/kv-cache/index.js +0 -20
- package/src/inference/kv-cache/sliding-window.d.ts +0 -72
- package/src/inference/kv-cache/sliding-window.js +0 -243
- package/src/inference/kv-cache/tiered.d.ts +0 -89
- package/src/inference/kv-cache/tiered.js +0 -576
- package/src/inference/kv-cache/types.d.ts +0 -188
- package/src/inference/kv-cache/types.js +0 -80
- package/src/inference/kv-cache.d.ts +0 -36
- package/src/inference/kv-cache.js +0 -18
- package/src/inference/moe-router.d.ts +0 -212
- package/src/inference/moe-router.js +0 -585
- package/src/inference/multi-model-network.d.ts +0 -139
- package/src/inference/multi-model-network.js +0 -771
- package/src/inference/multi-pipeline-pool.d.ts +0 -62
- package/src/inference/multi-pipeline-pool.js +0 -161
- package/src/inference/network-evolution.d.ts +0 -55
- package/src/inference/network-evolution.js +0 -79
- package/src/inference/pipelines/context.d.ts +0 -21
- package/src/inference/pipelines/context.js +0 -184
- package/src/inference/pipelines/diffusion/helpers.d.ts +0 -29
- package/src/inference/pipelines/diffusion/helpers.js +0 -120
- package/src/inference/pipelines/diffusion/index.d.ts +0 -3
- package/src/inference/pipelines/diffusion/index.js +0 -3
- package/src/inference/pipelines/diffusion/init.d.ts +0 -24
- package/src/inference/pipelines/diffusion/init.js +0 -138
- package/src/inference/pipelines/diffusion/pipeline.d.ts +0 -38
- package/src/inference/pipelines/diffusion/pipeline.js +0 -771
- package/src/inference/pipelines/diffusion/sana-transformer.d.ts +0 -53
- package/src/inference/pipelines/diffusion/sana-transformer.js +0 -738
- package/src/inference/pipelines/diffusion/scheduler.d.ts +0 -35
- package/src/inference/pipelines/diffusion/scheduler.js +0 -153
- package/src/inference/pipelines/diffusion/sd3-transformer.d.ts +0 -20
- package/src/inference/pipelines/diffusion/sd3-transformer.js +0 -1194
- package/src/inference/pipelines/diffusion/sd3-weights.d.ts +0 -21
- package/src/inference/pipelines/diffusion/sd3-weights.js +0 -287
- package/src/inference/pipelines/diffusion/text-encoder-gpu.d.ts +0 -87
- package/src/inference/pipelines/diffusion/text-encoder-gpu.js +0 -1223
- package/src/inference/pipelines/diffusion/text-encoder.d.ts +0 -29
- package/src/inference/pipelines/diffusion/text-encoder.js +0 -195
- package/src/inference/pipelines/diffusion/types.d.ts +0 -116
- package/src/inference/pipelines/diffusion/types.js +0 -1
- package/src/inference/pipelines/diffusion/vae.d.ts +0 -20
- package/src/inference/pipelines/diffusion/vae.js +0 -1375
- package/src/inference/pipelines/diffusion/weights.d.ts +0 -40
- package/src/inference/pipelines/diffusion/weights.js +0 -150
- package/src/inference/pipelines/dream/energy-head-pipeline.d.ts +0 -29
- package/src/inference/pipelines/dream/energy-head-pipeline.js +0 -6
- package/src/inference/pipelines/dream/pipeline.d.ts +0 -17
- package/src/inference/pipelines/dream/pipeline.js +0 -8
- package/src/inference/pipelines/energy/index.d.ts +0 -1
- package/src/inference/pipelines/energy/index.js +0 -1
- package/src/inference/pipelines/energy/pipeline.d.ts +0 -27
- package/src/inference/pipelines/energy/pipeline.js +0 -686
- package/src/inference/pipelines/energy/quintel.d.ts +0 -92
- package/src/inference/pipelines/energy/quintel.js +0 -218
- package/src/inference/pipelines/energy/types.d.ts +0 -63
- package/src/inference/pipelines/energy/types.js +0 -1
- package/src/inference/pipelines/energy-head/index.d.ts +0 -6
- package/src/inference/pipelines/energy-head/index.js +0 -6
- package/src/inference/pipelines/energy-head/row-head-pipeline.d.ts +0 -103
- package/src/inference/pipelines/energy-head/row-head-pipeline.js +0 -491
- package/src/inference/pipelines/factory.d.ts +0 -10
- package/src/inference/pipelines/factory.js +0 -6
- package/src/inference/pipelines/index.d.ts +0 -22
- package/src/inference/pipelines/index.js +0 -19
- package/src/inference/pipelines/registry.d.ts +0 -15
- package/src/inference/pipelines/registry.js +0 -23
- package/src/inference/pipelines/rng.d.ts +0 -2
- package/src/inference/pipelines/rng.js +0 -17
- package/src/inference/pipelines/structured/index.d.ts +0 -8
- package/src/inference/pipelines/structured/index.js +0 -8
- package/src/inference/pipelines/structured/json-head-pipeline.d.ts +0 -58
- package/src/inference/pipelines/structured/json-head-pipeline.js +0 -196
- package/src/inference/pipelines/text/attention/index.d.ts +0 -24
- package/src/inference/pipelines/text/attention/index.js +0 -17
- package/src/inference/pipelines/text/attention/output-projection.d.ts +0 -12
- package/src/inference/pipelines/text/attention/output-projection.js +0 -8
- package/src/inference/pipelines/text/attention/projections.d.ts +0 -110
- package/src/inference/pipelines/text/attention/projections.js +0 -515
- package/src/inference/pipelines/text/attention/record.d.ts +0 -36
- package/src/inference/pipelines/text/attention/record.js +0 -685
- package/src/inference/pipelines/text/attention/run.d.ts +0 -38
- package/src/inference/pipelines/text/attention/run.js +0 -933
- package/src/inference/pipelines/text/attention/types.d.ts +0 -98
- package/src/inference/pipelines/text/attention/types.js +0 -67
- package/src/inference/pipelines/text/attention.d.ts +0 -23
- package/src/inference/pipelines/text/attention.js +0 -12
- package/src/inference/pipelines/text/bdpa-steamroller.d.ts +0 -22
- package/src/inference/pipelines/text/bdpa-steamroller.js +0 -158
- package/src/inference/pipelines/text/buffer-types.d.ts +0 -7
- package/src/inference/pipelines/text/buffer-types.js +0 -4
- package/src/inference/pipelines/text/chat-format.d.ts +0 -46
- package/src/inference/pipelines/text/chat-format.js +0 -390
- package/src/inference/pipelines/text/config.d.ts +0 -244
- package/src/inference/pipelines/text/config.js +0 -703
- package/src/inference/pipelines/text/debug-utils/config.d.ts +0 -144
- package/src/inference/pipelines/text/debug-utils/config.js +0 -156
- package/src/inference/pipelines/text/debug-utils/index.d.ts +0 -53
- package/src/inference/pipelines/text/debug-utils/index.js +0 -44
- package/src/inference/pipelines/text/debug-utils/logging.d.ts +0 -106
- package/src/inference/pipelines/text/debug-utils/logging.js +0 -152
- package/src/inference/pipelines/text/debug-utils/tensor.d.ts +0 -119
- package/src/inference/pipelines/text/debug-utils/tensor.js +0 -268
- package/src/inference/pipelines/text/debug-utils/utils.d.ts +0 -77
- package/src/inference/pipelines/text/debug-utils/utils.js +0 -139
- package/src/inference/pipelines/text/debug-utils.d.ts +0 -42
- package/src/inference/pipelines/text/debug-utils.js +0 -34
- package/src/inference/pipelines/text/embed.d.ts +0 -67
- package/src/inference/pipelines/text/embed.js +0 -455
- package/src/inference/pipelines/text/execution-plan.d.ts +0 -116
- package/src/inference/pipelines/text/execution-plan.js +0 -329
- package/src/inference/pipelines/text/execution-v0-contract-helpers.d.ts +0 -59
- package/src/inference/pipelines/text/execution-v0-contract-helpers.js +0 -937
- package/src/inference/pipelines/text/execution-v0-runtime-builders.d.ts +0 -15
- package/src/inference/pipelines/text/execution-v0-runtime-builders.js +0 -279
- package/src/inference/pipelines/text/execution-v0.d.ts +0 -66
- package/src/inference/pipelines/text/execution-v0.js +0 -255
- package/src/inference/pipelines/text/ffn/dense.d.ts +0 -40
- package/src/inference/pipelines/text/ffn/dense.js +0 -759
- package/src/inference/pipelines/text/ffn/index.d.ts +0 -23
- package/src/inference/pipelines/text/ffn/index.js +0 -16
- package/src/inference/pipelines/text/ffn/moe.d.ts +0 -21
- package/src/inference/pipelines/text/ffn/moe.js +0 -49
- package/src/inference/pipelines/text/ffn/sandwich.d.ts +0 -25
- package/src/inference/pipelines/text/ffn/sandwich.js +0 -196
- package/src/inference/pipelines/text/ffn/standard.d.ts +0 -23
- package/src/inference/pipelines/text/ffn/standard.js +0 -87
- package/src/inference/pipelines/text/ffn/types.d.ts +0 -30
- package/src/inference/pipelines/text/ffn/types.js +0 -25
- package/src/inference/pipelines/text/ffn.d.ts +0 -31
- package/src/inference/pipelines/text/ffn.js +0 -18
- package/src/inference/pipelines/text/finiteness-guard-status.d.ts +0 -11
- package/src/inference/pipelines/text/finiteness-guard-status.js +0 -21
- package/src/inference/pipelines/text/finiteness-policy.d.ts +0 -35
- package/src/inference/pipelines/text/finiteness-policy.js +0 -45
- package/src/inference/pipelines/text/generator-helpers.d.ts +0 -34
- package/src/inference/pipelines/text/generator-helpers.js +0 -175
- package/src/inference/pipelines/text/generator-runtime.d.ts +0 -93
- package/src/inference/pipelines/text/generator-runtime.js +0 -378
- package/src/inference/pipelines/text/generator-steps.d.ts +0 -127
- package/src/inference/pipelines/text/generator-steps.js +0 -1197
- package/src/inference/pipelines/text/generator.d.ts +0 -41
- package/src/inference/pipelines/text/generator.js +0 -1361
- package/src/inference/pipelines/text/index.d.ts +0 -5
- package/src/inference/pipelines/text/index.js +0 -6
- package/src/inference/pipelines/text/init.d.ts +0 -312
- package/src/inference/pipelines/text/init.js +0 -1112
- package/src/inference/pipelines/text/kernel-path-auto-select.d.ts +0 -12
- package/src/inference/pipelines/text/kernel-path-auto-select.js +0 -92
- package/src/inference/pipelines/text/kernel-trace.d.ts +0 -152
- package/src/inference/pipelines/text/kernel-trace.js +0 -330
- package/src/inference/pipelines/text/layer-plan.d.ts +0 -65
- package/src/inference/pipelines/text/layer-plan.js +0 -249
- package/src/inference/pipelines/text/layer.d.ts +0 -56
- package/src/inference/pipelines/text/layer.js +0 -922
- package/src/inference/pipelines/text/linear-attention.d.ts +0 -109
- package/src/inference/pipelines/text/linear-attention.js +0 -907
- package/src/inference/pipelines/text/logits/cpu.d.ts +0 -81
- package/src/inference/pipelines/text/logits/cpu.js +0 -91
- package/src/inference/pipelines/text/logits/gpu.d.ts +0 -113
- package/src/inference/pipelines/text/logits/gpu.js +0 -411
- package/src/inference/pipelines/text/logits/index.d.ts +0 -62
- package/src/inference/pipelines/text/logits/index.js +0 -306
- package/src/inference/pipelines/text/logits/types.d.ts +0 -46
- package/src/inference/pipelines/text/logits/types.js +0 -4
- package/src/inference/pipelines/text/logits/utils.d.ts +0 -56
- package/src/inference/pipelines/text/logits/utils.js +0 -68
- package/src/inference/pipelines/text/logits.d.ts +0 -27
- package/src/inference/pipelines/text/logits.js +0 -16
- package/src/inference/pipelines/text/lora-apply.d.ts +0 -28
- package/src/inference/pipelines/text/lora-apply.js +0 -76
- package/src/inference/pipelines/text/lora-types.d.ts +0 -39
- package/src/inference/pipelines/text/lora-types.js +0 -18
- package/src/inference/pipelines/text/lora.d.ts +0 -18
- package/src/inference/pipelines/text/lora.js +0 -12
- package/src/inference/pipelines/text/model-load.d.ts +0 -58
- package/src/inference/pipelines/text/model-load.js +0 -739
- package/src/inference/pipelines/text/moe-cache.d.ts +0 -32
- package/src/inference/pipelines/text/moe-cache.js +0 -108
- package/src/inference/pipelines/text/moe-cpu-gptoss.d.ts +0 -9
- package/src/inference/pipelines/text/moe-cpu-gptoss.js +0 -115
- package/src/inference/pipelines/text/moe-cpu.d.ts +0 -13
- package/src/inference/pipelines/text/moe-cpu.js +0 -120
- package/src/inference/pipelines/text/moe-gpu.d.ts +0 -13
- package/src/inference/pipelines/text/moe-gpu.js +0 -635
- package/src/inference/pipelines/text/moe-helpers.d.ts +0 -12
- package/src/inference/pipelines/text/moe-helpers.js +0 -21
- package/src/inference/pipelines/text/moe-impl.d.ts +0 -117
- package/src/inference/pipelines/text/moe-impl.js +0 -9
- package/src/inference/pipelines/text/moe-shape-validator.d.ts +0 -31
- package/src/inference/pipelines/text/moe-shape-validator.js +0 -78
- package/src/inference/pipelines/text/ops.d.ts +0 -167
- package/src/inference/pipelines/text/ops.js +0 -367
- package/src/inference/pipelines/text/probes.d.ts +0 -31
- package/src/inference/pipelines/text/probes.js +0 -170
- package/src/inference/pipelines/text/sampling.d.ts +0 -54
- package/src/inference/pipelines/text/sampling.js +0 -249
- package/src/inference/pipelines/text/state.d.ts +0 -112
- package/src/inference/pipelines/text/state.js +0 -152
- package/src/inference/pipelines/text/types.d.ts +0 -627
- package/src/inference/pipelines/text/types.js +0 -4
- package/src/inference/pipelines/text/weights.d.ts +0 -110
- package/src/inference/pipelines/text/weights.js +0 -173
- package/src/inference/pipelines/text.d.ts +0 -157
- package/src/inference/pipelines/text.js +0 -608
- package/src/inference/speculative.d.ts +0 -239
- package/src/inference/speculative.js +0 -402
- package/src/inference/test-harness.d.ts +0 -178
- package/src/inference/test-harness.js +0 -359
- package/src/inference/tokenizer.d.ts +0 -72
- package/src/inference/tokenizer.js +0 -239
- package/src/inference/tokenizers/base.d.ts +0 -39
- package/src/inference/tokenizers/base.js +0 -69
- package/src/inference/tokenizers/bpe.d.ts +0 -27
- package/src/inference/tokenizers/bpe.js +0 -180
- package/src/inference/tokenizers/bundled.d.ts +0 -63
- package/src/inference/tokenizers/bundled.js +0 -1009
- package/src/inference/tokenizers/sentencepiece.d.ts +0 -28
- package/src/inference/tokenizers/sentencepiece.js +0 -401
- package/src/inference/tokenizers/types.d.ts +0 -166
- package/src/inference/tokenizers/types.js +0 -7
- package/src/loader/doppler-loader.d.ts +0 -134
- package/src/loader/doppler-loader.js +0 -1052
- package/src/loader/dtype-utils.d.ts +0 -40
- package/src/loader/dtype-utils.js +0 -61
- package/src/loader/embedding-loader.d.ts +0 -56
- package/src/loader/embedding-loader.js +0 -211
- package/src/loader/experts/expert-cache.d.ts +0 -156
- package/src/loader/experts/expert-cache.js +0 -382
- package/src/loader/experts/expert-loader.d.ts +0 -108
- package/src/loader/experts/expert-loader.js +0 -388
- package/src/loader/final-weights-loader.d.ts +0 -68
- package/src/loader/final-weights-loader.js +0 -268
- package/src/loader/index.d.ts +0 -150
- package/src/loader/index.js +0 -124
- package/src/loader/layer-loader.d.ts +0 -63
- package/src/loader/layer-loader.js +0 -418
- package/src/loader/loader-state.d.ts +0 -51
- package/src/loader/loader-state.js +0 -142
- package/src/loader/loader-types.d.ts +0 -236
- package/src/loader/loader-types.js +0 -4
- package/src/loader/manifest-config.d.ts +0 -97
- package/src/loader/manifest-config.js +0 -132
- package/src/loader/memory-monitor.d.ts +0 -112
- package/src/loader/memory-monitor.js +0 -284
- package/src/loader/multi-model-loader.d.ts +0 -51
- package/src/loader/multi-model-loader.js +0 -133
- package/src/loader/quantization-constants.d.ts +0 -23
- package/src/loader/quantization-constants.js +0 -14
- package/src/loader/shard-cache.d.ts +0 -60
- package/src/loader/shard-cache.js +0 -638
- package/src/loader/shard-resolver.d.ts +0 -12
- package/src/loader/shard-resolver.js +0 -105
- package/src/loader/tensors/tensor-loader.d.ts +0 -154
- package/src/loader/tensors/tensor-loader.js +0 -497
- package/src/loader/tensors/tensor-reader.d.ts +0 -22
- package/src/loader/tensors/tensor-reader.js +0 -113
- package/src/loader/tensors/tensor-role.d.ts +0 -7
- package/src/loader/tensors/tensor-role.js +0 -12
- package/src/loader/weight-downcast.d.ts +0 -62
- package/src/loader/weight-downcast.js +0 -213
- package/src/loader/weights.d.ts +0 -22
- package/src/loader/weights.js +0 -4
- package/src/memory/address-table.d.ts +0 -104
- package/src/memory/address-table.js +0 -114
- package/src/memory/buffer-pool.d.ts +0 -204
- package/src/memory/buffer-pool.js +0 -821
- package/src/memory/capability.d.ts +0 -49
- package/src/memory/capability.js +0 -95
- package/src/memory/heap-manager.d.ts +0 -104
- package/src/memory/heap-manager.js +0 -264
- package/src/memory/unified-detect.d.ts +0 -59
- package/src/memory/unified-detect.js +0 -192
- package/src/rules/converter/execution.rules.json +0 -20
- package/src/rules/converter/tensor-roles.rules.json +0 -13
- package/src/rules/converter/tokenizer.rules.json +0 -7
- package/src/rules/execution-rules-contract-check.d.ts +0 -17
- package/src/rules/execution-rules-contract-check.js +0 -245
- package/src/rules/inference/attention.rules.json +0 -54
- package/src/rules/inference/config.rules.json +0 -58
- package/src/rules/inference/dtype.rules.json +0 -99
- package/src/rules/inference/execution.rules.json +0 -45
- package/src/rules/inference/ffn.rules.json +0 -35
- package/src/rules/inference/kernel-path.rules.json +0 -92
- package/src/rules/inference/layer-pattern.rules.json +0 -16
- package/src/rules/inference/layer.rules.json +0 -7
- package/src/rules/inference/moe.rules.json +0 -48
- package/src/rules/kernels/attention.rules.json +0 -61
- package/src/rules/kernels/conv2d.rules.json +0 -6
- package/src/rules/kernels/depthwise-conv2d.rules.json +0 -6
- package/src/rules/kernels/dequant.rules.json +0 -58
- package/src/rules/kernels/energy.rules.json +0 -22
- package/src/rules/kernels/fused-ffn.rules.json +0 -13
- package/src/rules/kernels/fused-matmul-residual.rules.json +0 -6
- package/src/rules/kernels/fused-matmul-rmsnorm.rules.json +0 -8
- package/src/rules/kernels/gather.rules.json +0 -12
- package/src/rules/kernels/gelu.rules.json +0 -11
- package/src/rules/kernels/grouped-pointwise-conv2d.rules.json +0 -6
- package/src/rules/kernels/groupnorm.rules.json +0 -10
- package/src/rules/kernels/kernel-validator.d.ts +0 -24
- package/src/rules/kernels/kernel-validator.js +0 -160
- package/src/rules/kernels/kv_quantize.rules.json +0 -7
- package/src/rules/kernels/layernorm.rules.json +0 -6
- package/src/rules/kernels/matmul.rules.json +0 -60
- package/src/rules/kernels/modulate.rules.json +0 -6
- package/src/rules/kernels/moe.rules.gptoss.json +0 -105
- package/src/rules/kernels/moe.rules.json +0 -11
- package/src/rules/kernels/pixel_shuffle.rules.json +0 -6
- package/src/rules/kernels/relu.rules.json +0 -6
- package/src/rules/kernels/repeat-channels.rules.json +0 -6
- package/src/rules/kernels/residual.rules.json +0 -12
- package/src/rules/kernels/rmsnorm.rules.json +0 -11
- package/src/rules/kernels/rope.rules.json +0 -6
- package/src/rules/kernels/sample.rules.json +0 -6
- package/src/rules/kernels/sana-linear-attention.rules.json +0 -6
- package/src/rules/kernels/scale.rules.json +0 -6
- package/src/rules/kernels/silu.rules.json +0 -21
- package/src/rules/kernels/softmax.rules.json +0 -23
- package/src/rules/kernels/split-qg.rules.json +0 -6
- package/src/rules/kernels/split-qkv.rules.json +0 -6
- package/src/rules/kernels/upsample2d.rules.json +0 -6
- package/src/rules/layer-pattern-contract-check.d.ts +0 -17
- package/src/rules/layer-pattern-contract-check.js +0 -231
- package/src/rules/loader/tensor-loader.rules.json +0 -15
- package/src/rules/loader/weights.rules.json +0 -41
- package/src/rules/rule-registry.d.ts +0 -76
- package/src/rules/rule-registry.js +0 -241
- package/src/rules/tooling/command-runtime.rules.json +0 -56
- package/src/storage/backends/idb-store.d.ts +0 -52
- package/src/storage/backends/idb-store.js +0 -590
- package/src/storage/backends/memory-store.d.ts +0 -36
- package/src/storage/backends/memory-store.js +0 -242
- package/src/storage/backends/opfs-store.d.ts +0 -41
- package/src/storage/backends/opfs-store.js +0 -473
- package/src/storage/blake3.d.ts +0 -17
- package/src/storage/blake3.js +0 -269
- package/src/storage/download-types.d.ts +0 -157
- package/src/storage/download-types.js +0 -48
- package/src/storage/downloader.d.ts +0 -103
- package/src/storage/downloader.js +0 -1121
- package/src/storage/emulated-vram.d.ts +0 -264
- package/src/storage/emulated-vram.js +0 -576
- package/src/storage/export.d.ts +0 -20
- package/src/storage/export.js +0 -159
- package/src/storage/index.d.ts +0 -256
- package/src/storage/index.js +0 -188
- package/src/storage/inventory.d.ts +0 -26
- package/src/storage/inventory.js +0 -218
- package/src/storage/preflight.d.ts +0 -144
- package/src/storage/preflight.js +0 -316
- package/src/storage/quickstart-downloader.d.ts +0 -154
- package/src/storage/quickstart-downloader.js +0 -271
- package/src/storage/quota.d.ts +0 -150
- package/src/storage/quota.js +0 -304
- package/src/storage/registry.d.ts +0 -28
- package/src/storage/registry.js +0 -131
- package/src/storage/reports.d.ts +0 -20
- package/src/storage/reports.js +0 -94
- package/src/storage/shard-manager.d.ts +0 -151
- package/src/storage/shard-manager.js +0 -850
- package/src/storage/source-artifact-store.d.ts +0 -52
- package/src/storage/source-artifact-store.js +0 -234
- package/src/sw.d.ts +0 -1
- package/src/sw.js +0 -187
- package/src/tooling/browser-command-runner.d.ts +0 -28
- package/src/tooling/browser-command-runner.js +0 -82
- package/src/tooling/command-api-constants.d.ts +0 -9
- package/src/tooling/command-api-constants.js +0 -9
- package/src/tooling/command-api-family-normalizers.d.ts +0 -9
- package/src/tooling/command-api-family-normalizers.js +0 -343
- package/src/tooling/command-api-helpers.d.ts +0 -25
- package/src/tooling/command-api-helpers.js +0 -262
- package/src/tooling/command-api.d.ts +0 -173
- package/src/tooling/command-api.js +0 -76
- package/src/tooling/command-envelope.d.ts +0 -81
- package/src/tooling/command-envelope.js +0 -198
- package/src/tooling/command-runner-shared.d.ts +0 -73
- package/src/tooling/command-runner-shared.js +0 -180
- package/src/tooling/command-runner.html +0 -45
- package/src/tooling/conversion-config-materializer.d.ts +0 -24
- package/src/tooling/conversion-config-materializer.js +0 -97
- package/src/tooling/lean-execution-contract-runner.d.ts +0 -43
- package/src/tooling/lean-execution-contract-runner.js +0 -158
- package/src/tooling/lean-execution-contract.d.ts +0 -16
- package/src/tooling/lean-execution-contract.js +0 -228
- package/src/tooling/node-browser-command-runner.d.ts +0 -34
- package/src/tooling/node-browser-command-runner.js +0 -813
- package/src/tooling/node-command-runner.d.ts +0 -36
- package/src/tooling/node-command-runner.js +0 -168
- package/src/tooling/node-convert-worker-pool.d.ts +0 -16
- package/src/tooling/node-convert-worker-pool.js +0 -186
- package/src/tooling/node-convert-worker.d.ts +0 -1
- package/src/tooling/node-convert-worker.js +0 -60
- package/src/tooling/node-convert.d.ts +0 -54
- package/src/tooling/node-converter.d.ts +0 -1
- package/src/tooling/node-converter.js +0 -1315
- package/src/tooling/node-file-fetch.d.ts +0 -1
- package/src/tooling/node-file-fetch.js +0 -38
- package/src/tooling/node-source-runtime.d.ts +0 -19
- package/src/tooling/node-source-runtime.js +0 -586
- package/src/tooling/node-webgpu.d.ts +0 -6
- package/src/tooling/node-webgpu.js +0 -267
- package/src/tooling/opfs-cache.d.ts +0 -11
- package/src/tooling/opfs-cache.js +0 -191
- package/src/tooling/runtime-input-composition.d.ts +0 -38
- package/src/tooling/runtime-input-composition.js +0 -86
- package/src/tooling/source-runtime-bundle.d.ts +0 -137
- package/src/tooling/source-runtime-bundle.js +0 -711
- package/src/tooling/source-runtime-materializer.d.ts +0 -6
- package/src/tooling/source-runtime-materializer.js +0 -93
- package/src/tooling-exports.browser.d.ts +0 -7
- package/src/tooling-exports.browser.js +0 -2
- package/src/tooling-exports.d.ts +0 -22
- package/src/tooling-exports.js +0 -7
- package/src/tooling-exports.shared.d.ts +0 -105
- package/src/tooling-exports.shared.js +0 -92
- package/src/training/README.md +0 -153
- package/src/training/artifacts.d.ts +0 -160
- package/src/training/artifacts.js +0 -896
- package/src/training/attention-backward.d.ts +0 -30
- package/src/training/attention-backward.js +0 -232
- package/src/training/attention-forward.d.ts +0 -22
- package/src/training/attention-forward.js +0 -82
- package/src/training/autograd.d.ts +0 -51
- package/src/training/autograd.js +0 -408
- package/src/training/checkpoint-watch.d.ts +0 -8
- package/src/training/checkpoint-watch.js +0 -139
- package/src/training/checkpoint.d.ts +0 -36
- package/src/training/checkpoint.js +0 -277
- package/src/training/clip.d.ts +0 -9
- package/src/training/clip.js +0 -55
- package/src/training/dataloader.d.ts +0 -8
- package/src/training/dataloader.js +0 -44
- package/src/training/datasets/index.d.ts +0 -12
- package/src/training/datasets/index.js +0 -6
- package/src/training/datasets/jsonl.d.ts +0 -11
- package/src/training/datasets/jsonl.js +0 -50
- package/src/training/datasets/reploid.d.ts +0 -3
- package/src/training/datasets/reploid.js +0 -36
- package/src/training/datasets/text-pairs.d.ts +0 -21
- package/src/training/datasets/text-pairs.js +0 -42
- package/src/training/datasets/token-batch.d.ts +0 -21
- package/src/training/datasets/token-batch.js +0 -52
- package/src/training/datasets/translation-pairs.d.ts +0 -34
- package/src/training/datasets/translation-pairs.js +0 -49
- package/src/training/distillation/artifacts.d.ts +0 -71
- package/src/training/distillation/artifacts.js +0 -132
- package/src/training/distillation/checkpoint-watch.d.ts +0 -10
- package/src/training/distillation/checkpoint-watch.js +0 -58
- package/src/training/distillation/dataset.d.ts +0 -59
- package/src/training/distillation/dataset.js +0 -337
- package/src/training/distillation/eval.d.ts +0 -34
- package/src/training/distillation/eval.js +0 -310
- package/src/training/distillation/index.d.ts +0 -29
- package/src/training/distillation/index.js +0 -29
- package/src/training/distillation/runtime.d.ts +0 -20
- package/src/training/distillation/runtime.js +0 -121
- package/src/training/distillation/scoreboard.d.ts +0 -6
- package/src/training/distillation/scoreboard.js +0 -8
- package/src/training/distillation/stage-a.d.ts +0 -45
- package/src/training/distillation/stage-a.js +0 -338
- package/src/training/distillation/stage-b.d.ts +0 -24
- package/src/training/distillation/stage-b.js +0 -20
- package/src/training/distillation/student-fixture.d.ts +0 -22
- package/src/training/distillation/student-fixture.js +0 -846
- package/src/training/distillation/suite-data.d.ts +0 -45
- package/src/training/distillation/suite-data.js +0 -189
- package/src/training/export.d.ts +0 -32
- package/src/training/export.js +0 -112
- package/src/training/index.d.ts +0 -62
- package/src/training/index.js +0 -51
- package/src/training/lora-pipeline.d.ts +0 -40
- package/src/training/lora-pipeline.js +0 -793
- package/src/training/lora.d.ts +0 -19
- package/src/training/lora.js +0 -71
- package/src/training/loss-scaling.d.ts +0 -21
- package/src/training/loss-scaling.js +0 -80
- package/src/training/loss.d.ts +0 -10
- package/src/training/loss.js +0 -40
- package/src/training/objectives/base.d.ts +0 -58
- package/src/training/objectives/base.js +0 -38
- package/src/training/objectives/cross_entropy.d.ts +0 -18
- package/src/training/objectives/cross_entropy.js +0 -34
- package/src/training/objectives/distill_kd.d.ts +0 -16
- package/src/training/objectives/distill_kd.js +0 -365
- package/src/training/objectives/distill_triplet.d.ts +0 -16
- package/src/training/objectives/distill_triplet.js +0 -408
- package/src/training/objectives/index.d.ts +0 -12
- package/src/training/objectives/index.js +0 -6
- package/src/training/objectives/ul_stage1_joint.d.ts +0 -16
- package/src/training/objectives/ul_stage1_joint.js +0 -188
- package/src/training/objectives/ul_stage2_base.d.ts +0 -16
- package/src/training/objectives/ul_stage2_base.js +0 -218
- package/src/training/operator-artifacts.d.ts +0 -62
- package/src/training/operator-artifacts.js +0 -140
- package/src/training/operator-command.d.ts +0 -5
- package/src/training/operator-command.js +0 -455
- package/src/training/operator-eval.d.ts +0 -48
- package/src/training/operator-eval.js +0 -230
- package/src/training/operator-scoreboard.d.ts +0 -5
- package/src/training/operator-scoreboard.js +0 -44
- package/src/training/optimizer.d.ts +0 -22
- package/src/training/optimizer.js +0 -127
- package/src/training/runner.d.ts +0 -248
- package/src/training/runner.js +0 -1220
- package/src/training/suite.d.ts +0 -299
- package/src/training/suite.js +0 -2196
- package/src/training/tensor-factory.d.ts +0 -9
- package/src/training/tensor-factory.js +0 -13
- package/src/training/trainer.d.ts +0 -89
- package/src/training/trainer.js +0 -299
- package/src/training/ul_dataset.d.ts +0 -47
- package/src/training/ul_dataset.js +0 -151
- package/src/training/ul_schedule.d.ts +0 -6
- package/src/training/ul_schedule.js +0 -29
- package/src/training/workloads.d.ts +0 -164
- package/src/training/workloads.js +0 -530
- package/src/types/chrome.d.ts +0 -36
- package/src/types/chrome.js +0 -1
- package/src/types/gpu.d.ts +0 -185
- package/src/types/gpu.js +0 -5
- package/src/types/index.d.ts +0 -3
- package/src/types/index.js +0 -3
- package/src/types/inference.d.ts +0 -197
- package/src/types/inference.js +0 -5
- package/src/types/model.d.ts +0 -130
- package/src/types/model.js +0 -5
- package/src/utils/index.d.ts +0 -7
- package/src/utils/index.js +0 -7
- package/src/utils/load-json.d.ts +0 -5
- package/src/utils/load-json.js +0 -23
- package/src/utils/plain-object.d.ts +0 -1
- package/src/utils/plain-object.js +0 -3
- package/src/utils/sha256.d.ts +0 -4
- package/src/utils/sha256.js +0 -135
- package/src/version.d.ts +0 -2
- package/src/version.js +0 -2
- package/tools/convert-safetensors-node.js +0 -233
- package/tools/doppler-cli.js +0 -1452
|
@@ -1,937 +0,0 @@
|
|
|
1
|
-
import {
|
|
2
|
-
buildExecutionV0KernelProfileKey,
|
|
3
|
-
indexExecutionV0KernelProfiles,
|
|
4
|
-
normalizeExecutionV0Dtype,
|
|
5
|
-
resolveExecutionV0KernelProfile,
|
|
6
|
-
resolveExecutionV0KVIO,
|
|
7
|
-
resolveExecutionV0Precision,
|
|
8
|
-
} from '../../../config/execution-v0-contract-check.js';
|
|
9
|
-
import {
|
|
10
|
-
EXECUTION_V0_SCHEMA_ID,
|
|
11
|
-
isExecutionV0Digest,
|
|
12
|
-
isExecutionV0Semver,
|
|
13
|
-
} from '../../../config/schema/execution-v0.schema.js';
|
|
14
|
-
import { KERNEL_CONFIGS } from '../../../gpu/kernels/kernel-configs.js';
|
|
15
|
-
import { buildKernelRefFromKernelEntry, isKernelRefBoundToKernel } from '../../../config/kernels/kernel-ref.js';
|
|
16
|
-
|
|
17
|
-
const PATCH_SET_MUTABLE_FIELDS = new Set(['precision', 'kvIO', 'constants', 'entry']);
|
|
18
|
-
const EXECUTION_V0_RUNTIME_KEYS = new Set(['session', 'executionPatch']);
|
|
19
|
-
|
|
20
|
-
const KERNEL_OUTPUT_CAPABILITIES = (() => {
|
|
21
|
-
const byKernelEntry = new Map();
|
|
22
|
-
for (const variants of Object.values(KERNEL_CONFIGS)) {
|
|
23
|
-
for (const config of Object.values(variants)) {
|
|
24
|
-
const kernel = config?.shaderFile;
|
|
25
|
-
const entry = config?.entryPoint ?? 'main';
|
|
26
|
-
if (typeof kernel !== 'string' || kernel.length === 0) continue;
|
|
27
|
-
const key = `${kernel}#${entry}`;
|
|
28
|
-
if (!byKernelEntry.has(key)) {
|
|
29
|
-
byKernelEntry.set(key, new Set());
|
|
30
|
-
}
|
|
31
|
-
const outputDtype = config?.outputDtype;
|
|
32
|
-
if (typeof outputDtype === 'string' && outputDtype.length > 0) {
|
|
33
|
-
byKernelEntry.get(key).add(String(outputDtype).toLowerCase());
|
|
34
|
-
}
|
|
35
|
-
}
|
|
36
|
-
}
|
|
37
|
-
return byKernelEntry;
|
|
38
|
-
})();
|
|
39
|
-
|
|
40
|
-
const normalizeDtype = normalizeExecutionV0Dtype;
|
|
41
|
-
const resolvePrecision = resolveExecutionV0Precision;
|
|
42
|
-
const resolveKVIO = resolveExecutionV0KVIO;
|
|
43
|
-
const indexKernelProfiles = indexExecutionV0KernelProfiles;
|
|
44
|
-
const buildKernelProfileKey = buildExecutionV0KernelProfileKey;
|
|
45
|
-
|
|
46
|
-
function getKernelOutputCapabilities(step) {
|
|
47
|
-
const kernel = String(step?.kernel ?? '').trim();
|
|
48
|
-
const entry = String(step?.entry ?? 'main').trim() || 'main';
|
|
49
|
-
if (!kernel) {
|
|
50
|
-
return null;
|
|
51
|
-
}
|
|
52
|
-
return KERNEL_OUTPUT_CAPABILITIES.get(`${kernel}#${entry}`) ?? null;
|
|
53
|
-
}
|
|
54
|
-
|
|
55
|
-
export function cloneJson(value) {
|
|
56
|
-
if (typeof structuredClone === 'function') {
|
|
57
|
-
return structuredClone(value);
|
|
58
|
-
}
|
|
59
|
-
return JSON.parse(JSON.stringify(value));
|
|
60
|
-
}
|
|
61
|
-
|
|
62
|
-
function normalizeStopCheckMode(value, label) {
|
|
63
|
-
const normalized = String(value ?? '').trim().toLowerCase();
|
|
64
|
-
if (normalized !== 'batch' && normalized !== 'per-token') {
|
|
65
|
-
throw new Error(`[ExecutionV0] ${label} must be "batch" or "per-token".`);
|
|
66
|
-
}
|
|
67
|
-
return normalized;
|
|
68
|
-
}
|
|
69
|
-
|
|
70
|
-
function normalizeKVLayout(value, label) {
|
|
71
|
-
if (value == null) {
|
|
72
|
-
return null;
|
|
73
|
-
}
|
|
74
|
-
const normalized = String(value).trim().toLowerCase();
|
|
75
|
-
if (!normalized) {
|
|
76
|
-
return null;
|
|
77
|
-
}
|
|
78
|
-
return normalized;
|
|
79
|
-
}
|
|
80
|
-
|
|
81
|
-
function requirePlainObject(value, label) {
|
|
82
|
-
if (!value || typeof value !== 'object' || Array.isArray(value)) {
|
|
83
|
-
throw new Error(`[ExecutionV0] ${label} must be an object.`);
|
|
84
|
-
}
|
|
85
|
-
return value;
|
|
86
|
-
}
|
|
87
|
-
|
|
88
|
-
function requireOwnProperty(root, key, label) {
|
|
89
|
-
if (!Object.prototype.hasOwnProperty.call(root, key)) {
|
|
90
|
-
throw new Error(`[ExecutionV0] ${label} is required.`);
|
|
91
|
-
}
|
|
92
|
-
return root[key];
|
|
93
|
-
}
|
|
94
|
-
|
|
95
|
-
function requireNullableObject(root, key, label) {
|
|
96
|
-
const value = requireOwnProperty(root, key, label);
|
|
97
|
-
if (value === null) {
|
|
98
|
-
return null;
|
|
99
|
-
}
|
|
100
|
-
return requirePlainObject(value, label);
|
|
101
|
-
}
|
|
102
|
-
|
|
103
|
-
function requireArrayProperty(root, key, label) {
|
|
104
|
-
const value = requireOwnProperty(root, key, label);
|
|
105
|
-
if (!Array.isArray(value)) {
|
|
106
|
-
throw new Error(`[ExecutionV0] ${label} must be an array.`);
|
|
107
|
-
}
|
|
108
|
-
return value;
|
|
109
|
-
}
|
|
110
|
-
|
|
111
|
-
function requirePositiveInteger(value, label) {
|
|
112
|
-
if (!Number.isInteger(value) || value < 1) {
|
|
113
|
-
throw new Error(`[ExecutionV0] ${label} must be a positive integer.`);
|
|
114
|
-
}
|
|
115
|
-
return value;
|
|
116
|
-
}
|
|
117
|
-
|
|
118
|
-
function requireOptionalBoolean(value, label) {
|
|
119
|
-
if (value === undefined) {
|
|
120
|
-
return undefined;
|
|
121
|
-
}
|
|
122
|
-
if (typeof value !== 'boolean') {
|
|
123
|
-
throw new Error(`[ExecutionV0] ${label} must be a boolean when provided.`);
|
|
124
|
-
}
|
|
125
|
-
return value;
|
|
126
|
-
}
|
|
127
|
-
|
|
128
|
-
function requireDtypeProperty(root, key, label) {
|
|
129
|
-
const value = requireOwnProperty(root, key, label);
|
|
130
|
-
if (value == null) {
|
|
131
|
-
throw new Error(`[ExecutionV0] ${label} is required.`);
|
|
132
|
-
}
|
|
133
|
-
return normalizeDtype(value, label);
|
|
134
|
-
}
|
|
135
|
-
|
|
136
|
-
function validateDecodeLoopContract(sessionDefaults) {
|
|
137
|
-
const decodeLoop = requireNullableObject(sessionDefaults, 'decodeLoop', 'sessionDefaults.decodeLoop');
|
|
138
|
-
if (decodeLoop === null) {
|
|
139
|
-
return;
|
|
140
|
-
}
|
|
141
|
-
requirePositiveInteger(
|
|
142
|
-
decodeLoop.batchSize,
|
|
143
|
-
'sessionDefaults.decodeLoop.batchSize'
|
|
144
|
-
);
|
|
145
|
-
requirePositiveInteger(
|
|
146
|
-
decodeLoop.readbackInterval,
|
|
147
|
-
'sessionDefaults.decodeLoop.readbackInterval'
|
|
148
|
-
);
|
|
149
|
-
normalizeStopCheckMode(
|
|
150
|
-
decodeLoop.stopCheckMode,
|
|
151
|
-
'sessionDefaults.decodeLoop.stopCheckMode'
|
|
152
|
-
);
|
|
153
|
-
if (decodeLoop.ringTokens !== undefined) {
|
|
154
|
-
requirePositiveInteger(
|
|
155
|
-
decodeLoop.ringTokens,
|
|
156
|
-
'sessionDefaults.decodeLoop.ringTokens'
|
|
157
|
-
);
|
|
158
|
-
}
|
|
159
|
-
if (decodeLoop.ringStop !== undefined) {
|
|
160
|
-
requirePositiveInteger(
|
|
161
|
-
decodeLoop.ringStop,
|
|
162
|
-
'sessionDefaults.decodeLoop.ringStop'
|
|
163
|
-
);
|
|
164
|
-
}
|
|
165
|
-
if (decodeLoop.ringStaging !== undefined) {
|
|
166
|
-
requirePositiveInteger(
|
|
167
|
-
decodeLoop.ringStaging,
|
|
168
|
-
'sessionDefaults.decodeLoop.ringStaging'
|
|
169
|
-
);
|
|
170
|
-
}
|
|
171
|
-
requireOptionalBoolean(
|
|
172
|
-
decodeLoop.disableCommandBatching,
|
|
173
|
-
'sessionDefaults.decodeLoop.disableCommandBatching'
|
|
174
|
-
);
|
|
175
|
-
}
|
|
176
|
-
|
|
177
|
-
export function validateManifestSessionDefaultsContract(manifestInference) {
|
|
178
|
-
const sessionDefaults = requirePlainObject(
|
|
179
|
-
manifestInference?.sessionDefaults,
|
|
180
|
-
'manifest.inference.sessionDefaults'
|
|
181
|
-
);
|
|
182
|
-
const compute = requirePlainObject(
|
|
183
|
-
requireOwnProperty(sessionDefaults, 'compute', 'sessionDefaults.compute'),
|
|
184
|
-
'sessionDefaults.compute'
|
|
185
|
-
);
|
|
186
|
-
const computeDefaults = requirePlainObject(
|
|
187
|
-
requireOwnProperty(compute, 'defaults', 'sessionDefaults.compute.defaults'),
|
|
188
|
-
'sessionDefaults.compute.defaults'
|
|
189
|
-
);
|
|
190
|
-
requireDtypeProperty(
|
|
191
|
-
computeDefaults,
|
|
192
|
-
'activationDtype',
|
|
193
|
-
'sessionDefaults.compute.defaults.activationDtype'
|
|
194
|
-
);
|
|
195
|
-
requireDtypeProperty(
|
|
196
|
-
computeDefaults,
|
|
197
|
-
'mathDtype',
|
|
198
|
-
'sessionDefaults.compute.defaults.mathDtype'
|
|
199
|
-
);
|
|
200
|
-
requireDtypeProperty(
|
|
201
|
-
computeDefaults,
|
|
202
|
-
'accumDtype',
|
|
203
|
-
'sessionDefaults.compute.defaults.accumDtype'
|
|
204
|
-
);
|
|
205
|
-
requireDtypeProperty(
|
|
206
|
-
computeDefaults,
|
|
207
|
-
'outputDtype',
|
|
208
|
-
'sessionDefaults.compute.defaults.outputDtype'
|
|
209
|
-
);
|
|
210
|
-
requireArrayProperty(
|
|
211
|
-
compute,
|
|
212
|
-
'kernelProfiles',
|
|
213
|
-
'sessionDefaults.compute.kernelProfiles'
|
|
214
|
-
);
|
|
215
|
-
const kvcache = requireNullableObject(sessionDefaults, 'kvcache', 'sessionDefaults.kvcache');
|
|
216
|
-
if (kvcache !== null) {
|
|
217
|
-
requireDtypeProperty(
|
|
218
|
-
kvcache,
|
|
219
|
-
'kvDtype',
|
|
220
|
-
'sessionDefaults.kvcache.kvDtype'
|
|
221
|
-
);
|
|
222
|
-
}
|
|
223
|
-
validateDecodeLoopContract(sessionDefaults);
|
|
224
|
-
}
|
|
225
|
-
|
|
226
|
-
function assertKernelRef(kernelRef, label) {
|
|
227
|
-
if (!kernelRef) return;
|
|
228
|
-
if (typeof kernelRef.id !== 'string' || kernelRef.id.trim().length === 0) {
|
|
229
|
-
throw new Error(`[ExecutionV0] ${label}.id is required`);
|
|
230
|
-
}
|
|
231
|
-
if (!isExecutionV0Semver(kernelRef.version)) {
|
|
232
|
-
throw new Error(`[ExecutionV0] ${label}.version must be semver; got "${kernelRef.version}"`);
|
|
233
|
-
}
|
|
234
|
-
if (!isExecutionV0Digest(kernelRef.digest)) {
|
|
235
|
-
throw new Error(`[ExecutionV0] ${label}.digest must match sha256:<64-hex>`);
|
|
236
|
-
}
|
|
237
|
-
}
|
|
238
|
-
|
|
239
|
-
export function isPhaseMatch(phase, targetPhase) {
|
|
240
|
-
return phase === 'both' || phase === targetPhase;
|
|
241
|
-
}
|
|
242
|
-
|
|
243
|
-
export function stepHasLayer(step, layerIdx) {
|
|
244
|
-
if (step.layers === 'all') return true;
|
|
245
|
-
if (!Array.isArray(step.layers)) return false;
|
|
246
|
-
return step.layers.includes(layerIdx);
|
|
247
|
-
}
|
|
248
|
-
|
|
249
|
-
export function normalizePhase(value, label) {
|
|
250
|
-
const normalized = String(value ?? '').trim().toLowerCase();
|
|
251
|
-
if (normalized !== 'prefill' && normalized !== 'decode' && normalized !== 'both') {
|
|
252
|
-
throw new Error(`[ExecutionV0] ${label} must be prefill|decode|both; got "${value}"`);
|
|
253
|
-
}
|
|
254
|
-
return normalized;
|
|
255
|
-
}
|
|
256
|
-
|
|
257
|
-
export function normalizeSection(value, label) {
|
|
258
|
-
const normalized = String(value ?? '').trim();
|
|
259
|
-
if (!['preLayer', 'layer', 'postLayer', 'sampling'].includes(normalized)) {
|
|
260
|
-
throw new Error(`[ExecutionV0] ${label} must be preLayer|layer|postLayer|sampling; got "${value}"`);
|
|
261
|
-
}
|
|
262
|
-
return normalized;
|
|
263
|
-
}
|
|
264
|
-
|
|
265
|
-
export function normalizeSlot(value, label) {
|
|
266
|
-
if (typeof value !== 'string' || value.trim().length === 0) {
|
|
267
|
-
throw new Error(`[ExecutionV0] ${label} must be a non-empty string`);
|
|
268
|
-
}
|
|
269
|
-
return value.trim();
|
|
270
|
-
}
|
|
271
|
-
|
|
272
|
-
function assertKernelPrecisionCapability(step, resolvedPrecision, policies) {
|
|
273
|
-
if (step.op === 'cast') {
|
|
274
|
-
return;
|
|
275
|
-
}
|
|
276
|
-
if (policies.unsupportedPrecision !== 'error') {
|
|
277
|
-
return;
|
|
278
|
-
}
|
|
279
|
-
const kernel = String(step.kernel ?? '').trim();
|
|
280
|
-
const entry = String(step.entry ?? 'main').trim() || 'main';
|
|
281
|
-
const supportedOutputDtypes = getKernelOutputCapabilities(step);
|
|
282
|
-
if (!supportedOutputDtypes) {
|
|
283
|
-
throw new Error(
|
|
284
|
-
`[ExecutionV0] step "${step.id}" kernel "${kernel}#${entry}" ` +
|
|
285
|
-
'is not present in kernel registry; cannot validate precision capability.'
|
|
286
|
-
);
|
|
287
|
-
}
|
|
288
|
-
if (supportedOutputDtypes.size === 0) {
|
|
289
|
-
return;
|
|
290
|
-
}
|
|
291
|
-
const outputDtype = normalizeDtype(resolvedPrecision.outputDtype, `${step.id}.precision.outputDtype`);
|
|
292
|
-
if (!supportedOutputDtypes.has(outputDtype)) {
|
|
293
|
-
throw new Error(
|
|
294
|
-
`[ExecutionV0] step "${step.id}" outputDtype=${outputDtype} is unsupported by ` +
|
|
295
|
-
`kernel "${kernel}#${entry}" (supported: ${[...supportedOutputDtypes].join(', ') || 'none'}).`
|
|
296
|
-
);
|
|
297
|
-
}
|
|
298
|
-
}
|
|
299
|
-
|
|
300
|
-
export function createSourceTrace() {
|
|
301
|
-
return {
|
|
302
|
-
session: {},
|
|
303
|
-
steps: {},
|
|
304
|
-
};
|
|
305
|
-
}
|
|
306
|
-
|
|
307
|
-
export function setSourceTrace(trace, path, source) {
|
|
308
|
-
if (!trace || typeof path !== 'string' || path.length === 0) return;
|
|
309
|
-
trace[path] = { source };
|
|
310
|
-
}
|
|
311
|
-
|
|
312
|
-
function setStepSourceTrace(trace, stepId, path, source) {
|
|
313
|
-
if (!trace || !stepId || !path) return;
|
|
314
|
-
if (!trace.steps[stepId]) {
|
|
315
|
-
trace.steps[stepId] = {};
|
|
316
|
-
}
|
|
317
|
-
trace.steps[stepId][path] = { source };
|
|
318
|
-
}
|
|
319
|
-
|
|
320
|
-
function isExecutionV0PlainObject(value) {
|
|
321
|
-
return value != null && typeof value === 'object' && !Array.isArray(value);
|
|
322
|
-
}
|
|
323
|
-
|
|
324
|
-
export function collectLeafPaths(value, prefix = [], out = []) {
|
|
325
|
-
if (Array.isArray(value)) {
|
|
326
|
-
if (prefix.length > 0) {
|
|
327
|
-
out.push(prefix);
|
|
328
|
-
}
|
|
329
|
-
return out;
|
|
330
|
-
}
|
|
331
|
-
if (!isExecutionV0PlainObject(value)) {
|
|
332
|
-
if (prefix.length > 0) {
|
|
333
|
-
out.push(prefix);
|
|
334
|
-
}
|
|
335
|
-
return out;
|
|
336
|
-
}
|
|
337
|
-
for (const [key, child] of Object.entries(value)) {
|
|
338
|
-
collectLeafPaths(child, [...prefix, key], out);
|
|
339
|
-
}
|
|
340
|
-
return out;
|
|
341
|
-
}
|
|
342
|
-
|
|
343
|
-
export function hasDefinedPath(root, pathSegments) {
|
|
344
|
-
let current = root;
|
|
345
|
-
for (const segment of pathSegments) {
|
|
346
|
-
if (!isExecutionV0PlainObject(current) || !Object.prototype.hasOwnProperty.call(current, segment)) {
|
|
347
|
-
return false;
|
|
348
|
-
}
|
|
349
|
-
current = current[segment];
|
|
350
|
-
}
|
|
351
|
-
return current !== undefined;
|
|
352
|
-
}
|
|
353
|
-
|
|
354
|
-
function resolveProfile(profileIndex, step) {
|
|
355
|
-
return resolveExecutionV0KernelProfile(profileIndex, step);
|
|
356
|
-
}
|
|
357
|
-
|
|
358
|
-
export function validateStepShape(step, index) {
|
|
359
|
-
if (!step || typeof step !== 'object') {
|
|
360
|
-
throw new Error(`[ExecutionV0] execution.steps[${index}] must be an object`);
|
|
361
|
-
}
|
|
362
|
-
if (typeof step.id !== 'string' || step.id.trim().length === 0) {
|
|
363
|
-
throw new Error(`[ExecutionV0] execution.steps[${index}].id is required`);
|
|
364
|
-
}
|
|
365
|
-
if (typeof step.op !== 'string' || step.op.trim().length === 0) {
|
|
366
|
-
throw new Error(`[ExecutionV0] execution.steps[${index}].op is required`);
|
|
367
|
-
}
|
|
368
|
-
normalizePhase(step.phase, `execution.steps[${index}].phase`);
|
|
369
|
-
normalizeSection(step.section, `execution.steps[${index}].section`);
|
|
370
|
-
normalizeSlot(step.src, `execution.steps[${index}].src`);
|
|
371
|
-
normalizeSlot(step.dst, `execution.steps[${index}].dst`);
|
|
372
|
-
if (step.layers !== 'all' && !Array.isArray(step.layers)) {
|
|
373
|
-
throw new Error(`[ExecutionV0] execution.steps[${index}].layers must be "all" or number[]`);
|
|
374
|
-
}
|
|
375
|
-
if (step.layers !== 'all') {
|
|
376
|
-
for (const layer of step.layers) {
|
|
377
|
-
if (!Number.isInteger(layer) || layer < 0) {
|
|
378
|
-
throw new Error(`[ExecutionV0] execution.steps[${index}].layers must contain non-negative integers`);
|
|
379
|
-
}
|
|
380
|
-
}
|
|
381
|
-
}
|
|
382
|
-
if (step.op === 'cast') {
|
|
383
|
-
normalizeDtype(step.toDtype, `execution.steps[${index}].toDtype`);
|
|
384
|
-
if (step.fromDtype != null) {
|
|
385
|
-
normalizeDtype(step.fromDtype, `execution.steps[${index}].fromDtype`);
|
|
386
|
-
}
|
|
387
|
-
} else {
|
|
388
|
-
if (typeof step.kernel !== 'string' || step.kernel.trim().length === 0) {
|
|
389
|
-
throw new Error(
|
|
390
|
-
`[ExecutionV0] execution.steps[${index}] "${step.id}" requires kernel (non-cast op)`
|
|
391
|
-
);
|
|
392
|
-
}
|
|
393
|
-
if (!step.kernelRef || typeof step.kernelRef !== 'object' || Array.isArray(step.kernelRef)) {
|
|
394
|
-
throw new Error(
|
|
395
|
-
`[ExecutionV0] execution.steps[${index}] "${step.id}" requires kernelRef {id, version, digest} (non-cast op)`
|
|
396
|
-
);
|
|
397
|
-
}
|
|
398
|
-
assertKernelRef(step.kernelRef, `execution.steps[${index}].kernelRef`);
|
|
399
|
-
const entry = String(step.entry ?? 'main').trim() || 'main';
|
|
400
|
-
let expectedKernelRef;
|
|
401
|
-
try {
|
|
402
|
-
expectedKernelRef = buildKernelRefFromKernelEntry(step.kernel, entry);
|
|
403
|
-
} catch (error) {
|
|
404
|
-
const message = error instanceof Error ? error.message : String(error);
|
|
405
|
-
throw new Error(
|
|
406
|
-
`[ExecutionV0] execution.steps[${index}] "${step.id}" kernel "${step.kernel}#${entry}" ` +
|
|
407
|
-
`cannot be content-pinned: ${message}`
|
|
408
|
-
);
|
|
409
|
-
}
|
|
410
|
-
if (!isKernelRefBoundToKernel(step.kernelRef, step.kernel, entry)) {
|
|
411
|
-
throw new Error(
|
|
412
|
-
`[ExecutionV0] execution.steps[${index}] "${step.id}" kernelRef does not match kernel binding ` +
|
|
413
|
-
`("${step.kernel}#${entry}"). Expected ${expectedKernelRef.id}@${expectedKernelRef.version} ${expectedKernelRef.digest}.`
|
|
414
|
-
);
|
|
415
|
-
}
|
|
416
|
-
}
|
|
417
|
-
}
|
|
418
|
-
|
|
419
|
-
export function assertExecutionRuntimeOverlay(runtimeInference) {
|
|
420
|
-
if (!runtimeInference || typeof runtimeInference !== 'object') {
|
|
421
|
-
return;
|
|
422
|
-
}
|
|
423
|
-
const unknownKeys = Object.keys(runtimeInference).filter((key) => !EXECUTION_V0_RUNTIME_KEYS.has(key));
|
|
424
|
-
if (unknownKeys.length > 0) {
|
|
425
|
-
throw new Error(
|
|
426
|
-
`[ExecutionV0] runtime.inference overlay supports only ${[...EXECUTION_V0_RUNTIME_KEYS].join(', ')}; ` +
|
|
427
|
-
`got unsupported keys: ${unknownKeys.join(', ')}.`
|
|
428
|
-
);
|
|
429
|
-
}
|
|
430
|
-
}
|
|
431
|
-
|
|
432
|
-
export function validateUniqueStepIds(steps) {
|
|
433
|
-
const ids = new Set();
|
|
434
|
-
for (const step of steps) {
|
|
435
|
-
if (ids.has(step.id)) {
|
|
436
|
-
throw new Error(`[ExecutionV0] duplicate step id "${step.id}"`);
|
|
437
|
-
}
|
|
438
|
-
ids.add(step.id);
|
|
439
|
-
}
|
|
440
|
-
}
|
|
441
|
-
|
|
442
|
-
export function hasExecutionV0(manifestInference) {
|
|
443
|
-
return !!manifestInference?.execution && Array.isArray(manifestInference.execution.steps);
|
|
444
|
-
}
|
|
445
|
-
|
|
446
|
-
export function assertExecutionV0Schema(manifestInference) {
|
|
447
|
-
if (!hasExecutionV0(manifestInference)) return;
|
|
448
|
-
const discriminator = manifestInference?.schema ?? null;
|
|
449
|
-
if (discriminator !== EXECUTION_V0_SCHEMA_ID) {
|
|
450
|
-
throw new Error(
|
|
451
|
-
`[ExecutionV0] manifest.inference.schema must be "${EXECUTION_V0_SCHEMA_ID}" ` +
|
|
452
|
-
`when execution is present; got "${discriminator}".`
|
|
453
|
-
);
|
|
454
|
-
}
|
|
455
|
-
}
|
|
456
|
-
|
|
457
|
-
export function applyExecutionPatchAtomic(baseSteps, patch) {
|
|
458
|
-
if (!patch) {
|
|
459
|
-
return baseSteps;
|
|
460
|
-
}
|
|
461
|
-
const steps = cloneJson(baseSteps);
|
|
462
|
-
const byId = new Map(steps.map((step, index) => [step.id, index]));
|
|
463
|
-
|
|
464
|
-
for (const entry of patch.set ?? []) {
|
|
465
|
-
if (!entry || typeof entry !== 'object' || typeof entry.id !== 'string') {
|
|
466
|
-
throw new Error('[ExecutionV0] executionPatch.set entries require id');
|
|
467
|
-
}
|
|
468
|
-
if (!byId.has(entry.id)) {
|
|
469
|
-
throw new Error(`[ExecutionV0] executionPatch.set target "${entry.id}" does not exist`);
|
|
470
|
-
}
|
|
471
|
-
for (const key of Object.keys(entry)) {
|
|
472
|
-
if (key === 'id') continue;
|
|
473
|
-
if (!PATCH_SET_MUTABLE_FIELDS.has(key)) {
|
|
474
|
-
throw new Error(`[ExecutionV0] executionPatch.set "${entry.id}" cannot mutate "${key}"`);
|
|
475
|
-
}
|
|
476
|
-
}
|
|
477
|
-
}
|
|
478
|
-
|
|
479
|
-
for (const entry of patch.remove ?? []) {
|
|
480
|
-
if (!entry || typeof entry !== 'object' || typeof entry.id !== 'string') {
|
|
481
|
-
throw new Error('[ExecutionV0] executionPatch.remove entries require id');
|
|
482
|
-
}
|
|
483
|
-
if (!byId.has(entry.id)) {
|
|
484
|
-
throw new Error(`[ExecutionV0] executionPatch.remove target "${entry.id}" does not exist`);
|
|
485
|
-
}
|
|
486
|
-
}
|
|
487
|
-
|
|
488
|
-
for (const entry of patch.set ?? []) {
|
|
489
|
-
const index = byId.get(entry.id);
|
|
490
|
-
const target = steps[index];
|
|
491
|
-
if (entry.precision !== undefined) target.precision = cloneJson(entry.precision);
|
|
492
|
-
if (entry.kvIO !== undefined) target.kvIO = cloneJson(entry.kvIO);
|
|
493
|
-
if (entry.constants !== undefined) target.constants = cloneJson(entry.constants);
|
|
494
|
-
if (entry.entry !== undefined) target.entry = entry.entry;
|
|
495
|
-
}
|
|
496
|
-
|
|
497
|
-
const removeIds = new Set((patch.remove ?? []).map((entry) => entry.id));
|
|
498
|
-
const removedSteps = steps.filter((step) => !removeIds.has(step.id));
|
|
499
|
-
|
|
500
|
-
let current = removedSteps;
|
|
501
|
-
const insertedAfterAnchors = new Map();
|
|
502
|
-
for (const entry of patch.add ?? []) {
|
|
503
|
-
if (!entry?.step || typeof entry.step !== 'object') {
|
|
504
|
-
throw new Error('[ExecutionV0] executionPatch.add requires a step payload');
|
|
505
|
-
}
|
|
506
|
-
const hasBefore = typeof entry.insertBefore === 'string' && entry.insertBefore.length > 0;
|
|
507
|
-
const hasAfter = typeof entry.insertAfter === 'string' && entry.insertAfter.length > 0;
|
|
508
|
-
if (hasBefore === hasAfter) {
|
|
509
|
-
throw new Error('[ExecutionV0] executionPatch.add requires exactly one of insertBefore or insertAfter');
|
|
510
|
-
}
|
|
511
|
-
if (current.some((step) => step.id === entry.step.id)) {
|
|
512
|
-
throw new Error(`[ExecutionV0] executionPatch.add step id "${entry.step.id}" already exists`);
|
|
513
|
-
}
|
|
514
|
-
const anchorId = hasBefore ? entry.insertBefore : entry.insertAfter;
|
|
515
|
-
const anchorIndex = current.findIndex((step) => step.id === anchorId);
|
|
516
|
-
if (anchorIndex < 0) {
|
|
517
|
-
throw new Error(`[ExecutionV0] executionPatch.add anchor "${anchorId}" not found`);
|
|
518
|
-
}
|
|
519
|
-
let insertIndex = hasBefore ? anchorIndex : anchorIndex + 1;
|
|
520
|
-
if (!hasBefore) {
|
|
521
|
-
const insertedIds = insertedAfterAnchors.get(anchorId) ?? [];
|
|
522
|
-
while (insertIndex < current.length && insertedIds.includes(current[insertIndex].id)) {
|
|
523
|
-
insertIndex += 1;
|
|
524
|
-
}
|
|
525
|
-
}
|
|
526
|
-
current = [
|
|
527
|
-
...current.slice(0, insertIndex),
|
|
528
|
-
cloneJson(entry.step),
|
|
529
|
-
...current.slice(insertIndex),
|
|
530
|
-
];
|
|
531
|
-
if (!hasBefore) {
|
|
532
|
-
const insertedIds = insertedAfterAnchors.get(anchorId) ?? [];
|
|
533
|
-
insertedIds.push(entry.step.id);
|
|
534
|
-
insertedAfterAnchors.set(anchorId, insertedIds);
|
|
535
|
-
}
|
|
536
|
-
}
|
|
537
|
-
|
|
538
|
-
validateUniqueStepIds(current);
|
|
539
|
-
return current;
|
|
540
|
-
}
|
|
541
|
-
|
|
542
|
-
export function indexRuntimePatchMeta(patch) {
|
|
543
|
-
const meta = {
|
|
544
|
-
addedSteps: new Set(),
|
|
545
|
-
precisionFieldsByStep: new Map(),
|
|
546
|
-
kvIOFieldsByStep: new Set(),
|
|
547
|
-
};
|
|
548
|
-
if (!patch || typeof patch !== 'object') {
|
|
549
|
-
return meta;
|
|
550
|
-
}
|
|
551
|
-
|
|
552
|
-
for (const add of patch.add ?? []) {
|
|
553
|
-
const stepId = add?.step?.id;
|
|
554
|
-
if (typeof stepId === 'string' && stepId.length > 0) {
|
|
555
|
-
meta.addedSteps.add(stepId);
|
|
556
|
-
}
|
|
557
|
-
}
|
|
558
|
-
|
|
559
|
-
for (const set of patch.set ?? []) {
|
|
560
|
-
const stepId = set?.id;
|
|
561
|
-
if (typeof stepId !== 'string' || stepId.length === 0) continue;
|
|
562
|
-
if (set.precision && typeof set.precision === 'object') {
|
|
563
|
-
meta.precisionFieldsByStep.set(stepId, new Set(Object.keys(set.precision)));
|
|
564
|
-
}
|
|
565
|
-
if (set.kvIO && typeof set.kvIO === 'object') {
|
|
566
|
-
meta.kvIOFieldsByStep.add(stepId);
|
|
567
|
-
}
|
|
568
|
-
}
|
|
569
|
-
return meta;
|
|
570
|
-
}
|
|
571
|
-
|
|
572
|
-
export function requireSessionActivationDtype(
|
|
573
|
-
sessionDefaults,
|
|
574
|
-
label = 'sessionDefaults.compute.defaults.activationDtype'
|
|
575
|
-
) {
|
|
576
|
-
const activationDtype = sessionDefaults?.compute?.defaults?.activationDtype;
|
|
577
|
-
if (activationDtype == null) {
|
|
578
|
-
throw new Error(`[ExecutionV0] ${label} is required.`);
|
|
579
|
-
}
|
|
580
|
-
return normalizeDtype(activationDtype, label);
|
|
581
|
-
}
|
|
582
|
-
|
|
583
|
-
export function createInitialSlotDtypes(sessionDefaults) {
|
|
584
|
-
const activationDefault = requireSessionActivationDtype(
|
|
585
|
-
sessionDefaults,
|
|
586
|
-
'sessionDefaults.compute.defaults.activationDtype'
|
|
587
|
-
);
|
|
588
|
-
return new Map([['state', activationDefault]]);
|
|
589
|
-
}
|
|
590
|
-
|
|
591
|
-
function ensureCompatibleKV(step, kvIO, sessionDefaults) {
|
|
592
|
-
if (step.op !== 'attention' || !kvIO) {
|
|
593
|
-
return;
|
|
594
|
-
}
|
|
595
|
-
const runtimeKvDtypeRaw = sessionDefaults?.kvcache?.kvDtype;
|
|
596
|
-
if (runtimeKvDtypeRaw == null) {
|
|
597
|
-
return;
|
|
598
|
-
}
|
|
599
|
-
const runtimeKvDtype = normalizeDtype(runtimeKvDtypeRaw, 'sessionDefaults.kvcache.kvDtype');
|
|
600
|
-
if (kvIO.readDtype !== runtimeKvDtype || kvIO.writeDtype !== runtimeKvDtype) {
|
|
601
|
-
throw new Error(
|
|
602
|
-
`[ExecutionV0] step "${step.id}" kvIO read/write (${kvIO.readDtype}/${kvIO.writeDtype}) ` +
|
|
603
|
-
`must match sessionDefaults.kvcache.kvDtype (${runtimeKvDtype}).`
|
|
604
|
-
);
|
|
605
|
-
}
|
|
606
|
-
}
|
|
607
|
-
|
|
608
|
-
export function resolvePhaseSteps(phase, steps, sessionDefaults, profileIndex, policies, options = {}) {
|
|
609
|
-
const slotDtypes = options.initialSlotDtypes
|
|
610
|
-
? new Map(options.initialSlotDtypes)
|
|
611
|
-
: createInitialSlotDtypes(sessionDefaults);
|
|
612
|
-
const resolved = [];
|
|
613
|
-
const sourceTrace = options.sourceTrace ?? null;
|
|
614
|
-
const sessionDefaultSources = options.sessionDefaultSources ?? {};
|
|
615
|
-
const runtimePatchMeta = options.runtimePatchMeta ?? {
|
|
616
|
-
addedSteps: new Set(),
|
|
617
|
-
precisionFieldsByStep: new Map(),
|
|
618
|
-
kvIOFieldsByStep: new Set(),
|
|
619
|
-
};
|
|
620
|
-
|
|
621
|
-
for (const step of steps) {
|
|
622
|
-
const stepPhase = normalizePhase(step.phase, `${step.id}.phase`);
|
|
623
|
-
if (!isPhaseMatch(stepPhase, phase)) continue;
|
|
624
|
-
const profile = resolveProfile(profileIndex, step);
|
|
625
|
-
if (
|
|
626
|
-
step.kernelRef
|
|
627
|
-
&& !profile
|
|
628
|
-
&& policies.unresolvedKernel === 'error'
|
|
629
|
-
) {
|
|
630
|
-
throw new Error(
|
|
631
|
-
`[ExecutionV0] step "${step.id}" references kernel profile ` +
|
|
632
|
-
`${step.kernelRef.id}@${step.kernelRef.version} (${step.kernelRef.digest}) ` +
|
|
633
|
-
'but no matching sessionDefaults.compute.kernelProfiles entry was found.'
|
|
634
|
-
);
|
|
635
|
-
}
|
|
636
|
-
const { precision, sources: precisionSources } = resolvePrecision(step, profile, sessionDefaults);
|
|
637
|
-
const src = normalizeSlot(step.src, `${step.id}.src`);
|
|
638
|
-
const dst = normalizeSlot(step.dst, `${step.id}.dst`);
|
|
639
|
-
if (!slotDtypes.has(src)) {
|
|
640
|
-
throw new Error(
|
|
641
|
-
`[ExecutionV0] step "${step.id}" reads slot "${src}" before it is produced. ` +
|
|
642
|
-
'Add an explicit producer step or cast/load bridge.'
|
|
643
|
-
);
|
|
644
|
-
}
|
|
645
|
-
const derivedInput = slotDtypes.get(src);
|
|
646
|
-
const inputDtype = normalizeDtype(precision.inputDtype ?? derivedInput, `${step.id}.precision.inputDtype`);
|
|
647
|
-
|
|
648
|
-
if (
|
|
649
|
-
policies.dtypeTransition === 'require_cast_step'
|
|
650
|
-
&& step.op !== 'cast'
|
|
651
|
-
&& inputDtype !== derivedInput
|
|
652
|
-
) {
|
|
653
|
-
throw new Error(
|
|
654
|
-
`[ExecutionV0] step "${step.id}" requires inputDtype=${inputDtype} ` +
|
|
655
|
-
`but slot "${src}" currently holds ${derivedInput}. Insert explicit cast step.`
|
|
656
|
-
);
|
|
657
|
-
}
|
|
658
|
-
|
|
659
|
-
let outputDtype = normalizeDtype(precision.outputDtype, `${step.id}.precision.outputDtype`);
|
|
660
|
-
let outputDtypeSource = precisionSources.outputDtype;
|
|
661
|
-
if (step.op !== 'cast' && outputDtypeSource === 'sessionDefault') {
|
|
662
|
-
const declaredOutputDtypes = getKernelOutputCapabilities(step);
|
|
663
|
-
if (declaredOutputDtypes && declaredOutputDtypes.size === 1) {
|
|
664
|
-
outputDtype = [...declaredOutputDtypes][0];
|
|
665
|
-
outputDtypeSource = 'derived';
|
|
666
|
-
}
|
|
667
|
-
}
|
|
668
|
-
if (step.op === 'cast') {
|
|
669
|
-
outputDtype = normalizeDtype(step.toDtype, `${step.id}.toDtype`);
|
|
670
|
-
outputDtypeSource = 'manifest';
|
|
671
|
-
const fromDtype = step.fromDtype
|
|
672
|
-
? normalizeDtype(step.fromDtype, `${step.id}.fromDtype`)
|
|
673
|
-
: derivedInput;
|
|
674
|
-
if (fromDtype !== derivedInput) {
|
|
675
|
-
throw new Error(
|
|
676
|
-
`[ExecutionV0] cast step "${step.id}" fromDtype=${fromDtype} does not match slot "${src}" dtype=${derivedInput}`
|
|
677
|
-
);
|
|
678
|
-
}
|
|
679
|
-
}
|
|
680
|
-
|
|
681
|
-
const resolvedPrecision = {
|
|
682
|
-
inputDtype,
|
|
683
|
-
mathDtype: normalizeDtype(precision.mathDtype, `${step.id}.precision.mathDtype`),
|
|
684
|
-
accumDtype: normalizeDtype(precision.accumDtype, `${step.id}.precision.accumDtype`),
|
|
685
|
-
outputDtype,
|
|
686
|
-
};
|
|
687
|
-
assertKernelPrecisionCapability(step, resolvedPrecision, policies);
|
|
688
|
-
slotDtypes.set(dst, outputDtype);
|
|
689
|
-
|
|
690
|
-
const kvIOResolved = step.op === 'attention'
|
|
691
|
-
? resolveKVIO(step, profile, sessionDefaults)
|
|
692
|
-
: null;
|
|
693
|
-
const kvIO = kvIOResolved?.value ?? null;
|
|
694
|
-
ensureCompatibleKV(step, kvIO, sessionDefaults);
|
|
695
|
-
|
|
696
|
-
if (sourceTrace) {
|
|
697
|
-
const precisionFieldsPatched = runtimePatchMeta.precisionFieldsByStep.get(step.id) ?? new Set();
|
|
698
|
-
const isAddedStep = runtimePatchMeta.addedSteps.has(step.id);
|
|
699
|
-
const inputPatched = isAddedStep
|
|
700
|
-
? step.precision?.inputDtype != null
|
|
701
|
-
: precisionFieldsPatched.has('inputDtype');
|
|
702
|
-
const mathPatched = isAddedStep
|
|
703
|
-
? step.precision?.mathDtype != null
|
|
704
|
-
: precisionFieldsPatched.has('mathDtype');
|
|
705
|
-
const accumPatched = isAddedStep
|
|
706
|
-
? step.precision?.accumDtype != null
|
|
707
|
-
: precisionFieldsPatched.has('accumDtype');
|
|
708
|
-
const outputPatched = isAddedStep
|
|
709
|
-
? step.precision?.outputDtype != null
|
|
710
|
-
: precisionFieldsPatched.has('outputDtype');
|
|
711
|
-
const mathSource = precisionSources.mathDtype === 'sessionDefault'
|
|
712
|
-
? sessionDefaultSources.mathDtype ?? 'derived'
|
|
713
|
-
: precisionSources.mathDtype;
|
|
714
|
-
const accumSource = precisionSources.accumDtype === 'sessionDefault'
|
|
715
|
-
? sessionDefaultSources.accumDtype ?? 'derived'
|
|
716
|
-
: precisionSources.accumDtype;
|
|
717
|
-
const outputSource = precisionSources.outputDtype === 'sessionDefault'
|
|
718
|
-
? outputDtypeSource === 'sessionDefault'
|
|
719
|
-
? (sessionDefaultSources.outputDtype ?? 'derived')
|
|
720
|
-
: outputDtypeSource
|
|
721
|
-
: outputDtypeSource;
|
|
722
|
-
setStepSourceTrace(sourceTrace, step.id, 'precision.inputDtype',
|
|
723
|
-
inputPatched
|
|
724
|
-
? 'runtime.patch'
|
|
725
|
-
: precision.inputDtype != null
|
|
726
|
-
? precisionSources.inputDtype
|
|
727
|
-
: 'derived');
|
|
728
|
-
setStepSourceTrace(sourceTrace, step.id, 'precision.mathDtype', mathPatched ? 'runtime.patch' : mathSource);
|
|
729
|
-
setStepSourceTrace(sourceTrace, step.id, 'precision.accumDtype', accumPatched ? 'runtime.patch' : accumSource);
|
|
730
|
-
setStepSourceTrace(sourceTrace, step.id, 'precision.outputDtype', outputPatched ? 'runtime.patch' : outputSource);
|
|
731
|
-
if (step.op === 'attention') {
|
|
732
|
-
const kvPatched = runtimePatchMeta.kvIOFieldsByStep.has(step.id)
|
|
733
|
-
|| (isAddedStep && !!step.kvIO);
|
|
734
|
-
const kvSource = kvIOResolved?.source === 'sessionDefault'
|
|
735
|
-
? sessionDefaultSources.kvDtype ?? 'derived'
|
|
736
|
-
: kvIOResolved?.source ?? 'derived';
|
|
737
|
-
const resolvedKvSource = kvPatched ? 'runtime.patch' : kvSource;
|
|
738
|
-
setStepSourceTrace(sourceTrace, step.id, 'kvIO.readDtype', resolvedKvSource);
|
|
739
|
-
setStepSourceTrace(sourceTrace, step.id, 'kvIO.writeDtype', resolvedKvSource);
|
|
740
|
-
}
|
|
741
|
-
}
|
|
742
|
-
|
|
743
|
-
resolved.push({
|
|
744
|
-
...step,
|
|
745
|
-
src,
|
|
746
|
-
dst,
|
|
747
|
-
phase: stepPhase,
|
|
748
|
-
section: normalizeSection(step.section, `${step.id}.section`),
|
|
749
|
-
precision: resolvedPrecision,
|
|
750
|
-
kvIO,
|
|
751
|
-
});
|
|
752
|
-
}
|
|
753
|
-
|
|
754
|
-
return {
|
|
755
|
-
steps: resolved,
|
|
756
|
-
finalSlotDtypes: slotDtypes,
|
|
757
|
-
};
|
|
758
|
-
}
|
|
759
|
-
|
|
760
|
-
function stripSchemaDefaultComputeDefaults(compute, manifestComputeDefaults, defaultComputeDefaults) {
|
|
761
|
-
if (!compute?.defaults || !manifestComputeDefaults || !defaultComputeDefaults) {
|
|
762
|
-
return compute;
|
|
763
|
-
}
|
|
764
|
-
const dtypeKeys = ['activationDtype', 'mathDtype', 'accumDtype', 'outputDtype'];
|
|
765
|
-
const hasManifestDtype = dtypeKeys.some(
|
|
766
|
-
(key) => manifestComputeDefaults[key] !== undefined && manifestComputeDefaults[key] !== null
|
|
767
|
-
);
|
|
768
|
-
if (!hasManifestDtype) {
|
|
769
|
-
return compute;
|
|
770
|
-
}
|
|
771
|
-
const nextDefaults = { ...compute.defaults };
|
|
772
|
-
let changed = false;
|
|
773
|
-
for (const key of dtypeKeys) {
|
|
774
|
-
if (
|
|
775
|
-
manifestComputeDefaults[key] !== undefined
|
|
776
|
-
&& manifestComputeDefaults[key] !== null
|
|
777
|
-
&& nextDefaults[key] === defaultComputeDefaults[key]
|
|
778
|
-
) {
|
|
779
|
-
delete nextDefaults[key];
|
|
780
|
-
changed = true;
|
|
781
|
-
}
|
|
782
|
-
}
|
|
783
|
-
if (!changed) {
|
|
784
|
-
return compute;
|
|
785
|
-
}
|
|
786
|
-
if (Object.keys(nextDefaults).length === 0) {
|
|
787
|
-
const nextCompute = { ...compute };
|
|
788
|
-
delete nextCompute.defaults;
|
|
789
|
-
return Object.keys(nextCompute).length === 0 ? null : nextCompute;
|
|
790
|
-
}
|
|
791
|
-
return { ...compute, defaults: nextDefaults };
|
|
792
|
-
}
|
|
793
|
-
|
|
794
|
-
export function normalizeRuntimeSessionForExecutionV0(
|
|
795
|
-
runtimeSession,
|
|
796
|
-
manifestInference,
|
|
797
|
-
defaultComputeDefaults
|
|
798
|
-
) {
|
|
799
|
-
const manifestSessionDefaults = manifestInference?.sessionDefaults ?? null;
|
|
800
|
-
const manifestProfiles = manifestSessionDefaults?.compute?.kernelProfiles;
|
|
801
|
-
const hasManifestProfiles = Array.isArray(manifestProfiles) && manifestProfiles.length > 0;
|
|
802
|
-
const manifestComputeDefaults = manifestSessionDefaults?.compute?.defaults ?? null;
|
|
803
|
-
const hasManifestKVCache = manifestSessionDefaults?.kvcache != null;
|
|
804
|
-
const hasManifestDecodeLoop = manifestSessionDefaults?.decodeLoop != null;
|
|
805
|
-
|
|
806
|
-
if (!runtimeSession || typeof runtimeSession !== 'object') {
|
|
807
|
-
return runtimeSession;
|
|
808
|
-
}
|
|
809
|
-
|
|
810
|
-
let compute = runtimeSession.compute ?? null;
|
|
811
|
-
let kvcache = Object.prototype.hasOwnProperty.call(runtimeSession, 'kvcache')
|
|
812
|
-
? runtimeSession.kvcache
|
|
813
|
-
: undefined;
|
|
814
|
-
let decodeLoop = Object.prototype.hasOwnProperty.call(runtimeSession, 'decodeLoop')
|
|
815
|
-
? runtimeSession.decodeLoop
|
|
816
|
-
: undefined;
|
|
817
|
-
let changed = false;
|
|
818
|
-
|
|
819
|
-
if (manifestComputeDefaults) {
|
|
820
|
-
const stripped = stripSchemaDefaultComputeDefaults(
|
|
821
|
-
compute,
|
|
822
|
-
manifestComputeDefaults,
|
|
823
|
-
defaultComputeDefaults
|
|
824
|
-
);
|
|
825
|
-
if (stripped !== compute) {
|
|
826
|
-
compute = stripped;
|
|
827
|
-
changed = true;
|
|
828
|
-
}
|
|
829
|
-
}
|
|
830
|
-
|
|
831
|
-
if (compute && Object.prototype.hasOwnProperty.call(compute, 'kernelProfiles')) {
|
|
832
|
-
const kernelProfiles = compute.kernelProfiles;
|
|
833
|
-
if (Array.isArray(kernelProfiles) && kernelProfiles.length === 0 && hasManifestProfiles) {
|
|
834
|
-
const nextCompute = { ...compute };
|
|
835
|
-
delete nextCompute.kernelProfiles;
|
|
836
|
-
compute = Object.keys(nextCompute).length === 0 ? null : nextCompute;
|
|
837
|
-
changed = true;
|
|
838
|
-
}
|
|
839
|
-
}
|
|
840
|
-
|
|
841
|
-
if (kvcache === null && hasManifestKVCache) {
|
|
842
|
-
kvcache = undefined;
|
|
843
|
-
changed = true;
|
|
844
|
-
}
|
|
845
|
-
|
|
846
|
-
if (decodeLoop === null && hasManifestDecodeLoop) {
|
|
847
|
-
decodeLoop = undefined;
|
|
848
|
-
changed = true;
|
|
849
|
-
}
|
|
850
|
-
|
|
851
|
-
if (!changed) {
|
|
852
|
-
return runtimeSession;
|
|
853
|
-
}
|
|
854
|
-
|
|
855
|
-
const nextRuntimeSession = { ...runtimeSession };
|
|
856
|
-
if (!compute) {
|
|
857
|
-
delete nextRuntimeSession.compute;
|
|
858
|
-
} else {
|
|
859
|
-
nextRuntimeSession.compute = compute;
|
|
860
|
-
}
|
|
861
|
-
if (kvcache === undefined) {
|
|
862
|
-
delete nextRuntimeSession.kvcache;
|
|
863
|
-
} else {
|
|
864
|
-
nextRuntimeSession.kvcache = kvcache;
|
|
865
|
-
}
|
|
866
|
-
if (decodeLoop === undefined) {
|
|
867
|
-
delete nextRuntimeSession.decodeLoop;
|
|
868
|
-
} else {
|
|
869
|
-
nextRuntimeSession.decodeLoop = decodeLoop;
|
|
870
|
-
}
|
|
871
|
-
|
|
872
|
-
return Object.keys(nextRuntimeSession).length === 0 ? {} : nextRuntimeSession;
|
|
873
|
-
}
|
|
874
|
-
|
|
875
|
-
export function validatePhaseBoundaryCompatibility(options) {
|
|
876
|
-
const {
|
|
877
|
-
steps,
|
|
878
|
-
prefillFinalSlotDtypes,
|
|
879
|
-
decodeInitialSlotDtypes,
|
|
880
|
-
sessionDefaults,
|
|
881
|
-
profileIndex,
|
|
882
|
-
policies,
|
|
883
|
-
} = options;
|
|
884
|
-
const decodeSteps = steps.filter((step) => isPhaseMatch(normalizePhase(step.phase, `${step.id}.phase`), 'decode'));
|
|
885
|
-
if (decodeSteps.length === 0) {
|
|
886
|
-
return;
|
|
887
|
-
}
|
|
888
|
-
const writtenSlots = new Set();
|
|
889
|
-
for (const step of decodeSteps) {
|
|
890
|
-
const src = normalizeSlot(step.src, `${step.id}.src`);
|
|
891
|
-
const dst = normalizeSlot(step.dst, `${step.id}.dst`);
|
|
892
|
-
const readsCarriedSlot = !writtenSlots.has(src) && prefillFinalSlotDtypes.has(src);
|
|
893
|
-
if (readsCarriedSlot && step.op !== 'cast') {
|
|
894
|
-
const profile = resolveProfile(profileIndex, step);
|
|
895
|
-
const { precision } = resolvePrecision(step, profile, sessionDefaults);
|
|
896
|
-
const carriedDtype = prefillFinalSlotDtypes.get(src);
|
|
897
|
-
const decodeInput = normalizeDtype(
|
|
898
|
-
precision.inputDtype
|
|
899
|
-
?? carriedDtype
|
|
900
|
-
?? decodeInitialSlotDtypes.get(src),
|
|
901
|
-
`${step.id}.precision.inputDtype`
|
|
902
|
-
);
|
|
903
|
-
if (decodeInput !== carriedDtype) {
|
|
904
|
-
throw new Error(
|
|
905
|
-
`[ExecutionV0] decode step "${step.id}" reads carried slot "${src}" as ${decodeInput} ` +
|
|
906
|
-
`but prefill left ${carriedDtype}. Add explicit cast at phase boundary.`
|
|
907
|
-
);
|
|
908
|
-
}
|
|
909
|
-
}
|
|
910
|
-
writtenSlots.add(dst);
|
|
911
|
-
}
|
|
912
|
-
}
|
|
913
|
-
|
|
914
|
-
export function assertKVLayoutExecutionCompatibility(steps, sessionDefaults) {
|
|
915
|
-
const kvLayout = normalizeKVLayout(sessionDefaults?.kvcache?.layout, 'sessionDefaults.kvcache.layout');
|
|
916
|
-
if (kvLayout !== 'bdpa') {
|
|
917
|
-
return;
|
|
918
|
-
}
|
|
919
|
-
const incompatibleStep = steps.find((step) => (
|
|
920
|
-
step?.op === 'attention'
|
|
921
|
-
&& isPhaseMatch(normalizePhase(step.phase, `${step.id}.phase`), 'prefill')
|
|
922
|
-
));
|
|
923
|
-
if (!incompatibleStep) {
|
|
924
|
-
return;
|
|
925
|
-
}
|
|
926
|
-
throw new Error(
|
|
927
|
-
`[ExecutionV0] sessionDefaults.kvcache.layout="bdpa" is decode-only, ` +
|
|
928
|
-
`but step "${incompatibleStep.id}" declares prefill attention. ` +
|
|
929
|
-
'Use a non-BDPA KV layout for prefill-capable models or remove prefill attention from the execution contract.'
|
|
930
|
-
);
|
|
931
|
-
}
|
|
932
|
-
|
|
933
|
-
export {
|
|
934
|
-
buildKernelProfileKey,
|
|
935
|
-
indexKernelProfiles,
|
|
936
|
-
normalizeDtype,
|
|
937
|
-
};
|