@simulatte/doppler 0.1.9 → 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 -116
- package/package.json +5 -162
- package/BRANDING.md +0 -14
- package/CHANGELOG.md +0 -158
- 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 -310
- package/src/client/doppler-api.d.ts +0 -83
- package/src/client/doppler-api.js +0 -323
- 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 -86
- package/src/client/doppler-registry.json +0 -40
- 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 -228
- 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/moe/mixtral.paths.json +0 -46
- 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 -584
- 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 -155
- 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-f32a-small-attn.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 -145
- 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 -78
- package/src/config/presets/models/gemma4.json +0 -61
- package/src/config/presets/models/gpt-oss.json +0 -68
- package/src/config/presets/models/granite-docling.json +0 -70
- 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 -88
- 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/qwen3_vl.json +0 -40
- 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 -54
- 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/experiments/verify/lfm2-verify.json +0 -46
- package/src/config/presets/runtime/experiments/verify/translategemma-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 -127
- package/src/config/presets/runtime/platform/metal-apple-q4k.json +0 -11
- package/src/config/presets/runtime/tiers/gemma4-16gb.json +0 -69
- package/src/config/presets/runtime/tiers/gemma4-24gb.json +0 -66
- package/src/config/presets/runtime/tiers/gemma4-32gb.json +0 -66
- 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 -64
- 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 -290
- package/src/config/schema/debug.schema.js +0 -134
- 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 -491
- 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 -184
- 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 -1383
- 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 -101
- package/src/converter/quantizer.js +0 -444
- 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 -2186
- 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 -307
- 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 -39
- 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 -1439
- 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 -153
- package/src/gpu/kernels/dequant_f16_out_vec4.wgsl +0 -152
- 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 -204
- package/src/gpu/kernels/dequant_shared_vec4.wgsl +0 -155
- package/src/gpu/kernels/dequant_subgroup.wgsl +0 -206
- 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/gated-short-conv.d.ts +0 -63
- package/src/gpu/kernels/gated-short-conv.js +0 -284
- 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 -555
- 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 -518
- package/src/gpu/kernels/matmul.d.ts +0 -114
- package/src/gpu/kernels/matmul.js +0 -384
- 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 -343
- 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 -565
- package/src/gpu/kernels/sample.wgsl +0 -407
- package/src/gpu/kernels/sample_f16.wgsl +0 -361
- 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 -213
- 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 -1076
- 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 -772
- 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 -1224
- 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 -113
- package/src/inference/pipelines/text/attention/projections.js +0 -526
- package/src/inference/pipelines/text/attention/record.d.ts +0 -36
- package/src/inference/pipelines/text/attention/record.js +0 -686
- package/src/inference/pipelines/text/attention/run.d.ts +0 -38
- package/src/inference/pipelines/text/attention/run.js +0 -942
- 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 -245
- package/src/inference/pipelines/text/config.js +0 -731
- 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 -474
- 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 -286
- package/src/inference/pipelines/text/execution-v0.d.ts +0 -66
- package/src/inference/pipelines/text/execution-v0.js +0 -266
- 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 -176
- package/src/inference/pipelines/text/generator-runtime.d.ts +0 -93
- package/src/inference/pipelines/text/generator-runtime.js +0 -392
- package/src/inference/pipelines/text/generator-steps.d.ts +0 -136
- package/src/inference/pipelines/text/generator-steps.js +0 -1214
- package/src/inference/pipelines/text/generator.d.ts +0 -46
- package/src/inference/pipelines/text/generator.js +0 -1515
- 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 -314
- package/src/inference/pipelines/text/init.js +0 -1126
- 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 -951
- 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 -653
- 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 -40
- package/src/inference/pipelines/text/moe-shape-validator.js +0 -98
- package/src/inference/pipelines/text/ops.d.ts +0 -167
- package/src/inference/pipelines/text/ops.js +0 -437
- package/src/inference/pipelines/text/probes.d.ts +0 -31
- package/src/inference/pipelines/text/probes.js +0 -171
- 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 -154
- 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 -162
- package/src/inference/pipelines/text.js +0 -666
- package/src/inference/pipelines/vision/encoder.js +0 -386
- package/src/inference/pipelines/vision/image-preprocess.js +0 -151
- package/src/inference/pipelines/vision/index.js +0 -173
- package/src/inference/pipelines/vision/ops.js +0 -78
- package/src/inference/pipelines/vision/patch-embed.js +0 -151
- 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 -361
- 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 -137
- package/src/loader/doppler-loader.js +0 -1069
- 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 -386
- package/src/loader/experts/expert-loader.d.ts +0 -108
- package/src/loader/experts/expert-loader.js +0 -392
- 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 -457
- 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 -134
- 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 -157
- package/src/loader/tensors/tensor-loader.js +0 -618
- 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/moe.rules.mixtral.json +0 -75
- 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 -25
- 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 -77
- package/src/rules/rule-registry.js +0 -243
- 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 -157
- package/src/storage/quickstart-downloader.js +0 -268
- 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-converter.d.ts +0 -1
- package/src/tooling/node-converter.js +0 -1333
- 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 -610
- package/src/tooling/node-webgpu.d.ts +0 -6
- package/src/tooling/node-webgpu.js +0 -284
- 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/hf-resolve-url.d.ts +0 -16
- package/src/utils/hf-resolve-url.js +0 -17
- 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
|
@@ -1,123 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Converter Config Schema Definitions
|
|
3
|
-
*
|
|
4
|
-
* Converter configuration for GGUF/SafeTensors -> RDRR output.
|
|
5
|
-
*
|
|
6
|
-
* @module config/schema/converter
|
|
7
|
-
*/
|
|
8
|
-
|
|
9
|
-
import type {
|
|
10
|
-
HashAlgorithm,
|
|
11
|
-
QuantizationValue,
|
|
12
|
-
RuntimeOptimizationsSchema,
|
|
13
|
-
ConversionInfoSchema,
|
|
14
|
-
} from './manifest.schema.js';
|
|
15
|
-
import type {
|
|
16
|
-
ExecutionV0ConfigSchema,
|
|
17
|
-
ExecutionV0SessionDefaultsSchema,
|
|
18
|
-
} from './execution-v0.schema.js';
|
|
19
|
-
|
|
20
|
-
export type ComputePrecision = 'f16' | 'f32' | 'auto' | null;
|
|
21
|
-
/** Q4K layout: 'row' = fused kernel (fast), 'col' = dequant fallback */
|
|
22
|
-
export type Q4KLayout = 'row' | 'col' | null;
|
|
23
|
-
export type ConverterWorkerCountPolicy = 'cap' | 'error';
|
|
24
|
-
|
|
25
|
-
export interface ConverterQuantizationConfigSchema {
|
|
26
|
-
weights: QuantizationValue | null;
|
|
27
|
-
embeddings: QuantizationValue | null;
|
|
28
|
-
lmHead: QuantizationValue | null;
|
|
29
|
-
vision: QuantizationValue | null;
|
|
30
|
-
audio: QuantizationValue | null;
|
|
31
|
-
projector: QuantizationValue | null;
|
|
32
|
-
q4kLayout: Q4KLayout;
|
|
33
|
-
computePrecision: ComputePrecision;
|
|
34
|
-
}
|
|
35
|
-
|
|
36
|
-
export interface ConverterShardingConfigSchema {
|
|
37
|
-
shardSizeBytes: number;
|
|
38
|
-
}
|
|
39
|
-
|
|
40
|
-
export interface ConverterStreamingConfigSchema {
|
|
41
|
-
chunkSizeBytes: number;
|
|
42
|
-
}
|
|
43
|
-
|
|
44
|
-
export interface ConverterHttpConfigSchema {
|
|
45
|
-
allowDownloadFallback: boolean;
|
|
46
|
-
maxDownloadBytes: number | null;
|
|
47
|
-
}
|
|
48
|
-
|
|
49
|
-
export interface ConverterWeightLayoutConfigSchema {
|
|
50
|
-
transposeWeights: boolean;
|
|
51
|
-
fuseGateUp: boolean;
|
|
52
|
-
}
|
|
53
|
-
|
|
54
|
-
export interface ConverterManifestConfigSchema {
|
|
55
|
-
hashAlgorithm: HashAlgorithm;
|
|
56
|
-
optimizations: RuntimeOptimizationsSchema | null;
|
|
57
|
-
conversion: ConversionInfoSchema | null;
|
|
58
|
-
}
|
|
59
|
-
|
|
60
|
-
export interface ConverterInferenceConfigSchema {
|
|
61
|
-
defaultKernelPath: string | null;
|
|
62
|
-
sessionDefaults: ExecutionV0SessionDefaultsSchema | null;
|
|
63
|
-
execution: ExecutionV0ConfigSchema['execution'] | null;
|
|
64
|
-
}
|
|
65
|
-
|
|
66
|
-
export interface ConverterOutputConfigSchema {
|
|
67
|
-
modelBaseId: string | null;
|
|
68
|
-
dir: string | null;
|
|
69
|
-
baseDir: string | null;
|
|
70
|
-
textOnly: boolean;
|
|
71
|
-
fast: boolean;
|
|
72
|
-
}
|
|
73
|
-
|
|
74
|
-
export interface GGUFParserDefaultsSchema {
|
|
75
|
-
contextLength: number;
|
|
76
|
-
attentionLayerNormEpsilon: number;
|
|
77
|
-
attentionLayerNormRMSEpsilon: number;
|
|
78
|
-
ropeFreqBase: number;
|
|
79
|
-
}
|
|
80
|
-
|
|
81
|
-
export interface ConverterPresetConfigSchema {
|
|
82
|
-
model: string | null;
|
|
83
|
-
}
|
|
84
|
-
|
|
85
|
-
/** Node convert command execution defaults (worker scheduling policy) */
|
|
86
|
-
export interface ConverterExecutionConfigSchema {
|
|
87
|
-
workers: number;
|
|
88
|
-
workerCountPolicy: ConverterWorkerCountPolicy;
|
|
89
|
-
rowChunkRows: number | null;
|
|
90
|
-
rowChunkMinTensorBytes: number;
|
|
91
|
-
maxInFlightJobs: number | null;
|
|
92
|
-
useGpuCast: boolean;
|
|
93
|
-
gpuCastMinTensorBytes: number;
|
|
94
|
-
}
|
|
95
|
-
|
|
96
|
-
export interface ConverterConfigSchema {
|
|
97
|
-
quantization: ConverterQuantizationConfigSchema;
|
|
98
|
-
sharding: ConverterShardingConfigSchema;
|
|
99
|
-
streaming: ConverterStreamingConfigSchema;
|
|
100
|
-
http: ConverterHttpConfigSchema;
|
|
101
|
-
weightLayout: ConverterWeightLayoutConfigSchema;
|
|
102
|
-
manifest: ConverterManifestConfigSchema;
|
|
103
|
-
inference: ConverterInferenceConfigSchema;
|
|
104
|
-
output: ConverterOutputConfigSchema;
|
|
105
|
-
presets: ConverterPresetConfigSchema;
|
|
106
|
-
}
|
|
107
|
-
|
|
108
|
-
export declare const DEFAULT_CONVERTER_QUANTIZATION_CONFIG: ConverterQuantizationConfigSchema;
|
|
109
|
-
export declare const DEFAULT_CONVERTER_SHARDING_CONFIG: ConverterShardingConfigSchema;
|
|
110
|
-
export declare const DEFAULT_CONVERTER_STREAMING_CONFIG: ConverterStreamingConfigSchema;
|
|
111
|
-
export declare const DEFAULT_CONVERTER_HTTP_CONFIG: ConverterHttpConfigSchema;
|
|
112
|
-
export declare const DEFAULT_CONVERTER_WEIGHT_LAYOUT_CONFIG: ConverterWeightLayoutConfigSchema;
|
|
113
|
-
export declare const DEFAULT_CONVERTER_MANIFEST_CONFIG: ConverterManifestConfigSchema;
|
|
114
|
-
export declare const DEFAULT_CONVERTER_INFERENCE_CONFIG: ConverterInferenceConfigSchema;
|
|
115
|
-
export declare const DEFAULT_CONVERTER_OUTPUT_CONFIG: ConverterOutputConfigSchema;
|
|
116
|
-
export declare const DEFAULT_CONVERTER_PRESET_CONFIG: ConverterPresetConfigSchema;
|
|
117
|
-
export declare const DEFAULT_GGUF_PARSER_DEFAULTS: GGUFParserDefaultsSchema;
|
|
118
|
-
export declare const DEFAULT_CONVERTER_EXECUTION_CONFIG: ConverterExecutionConfigSchema;
|
|
119
|
-
export declare const DEFAULT_CONVERTER_CONFIG: ConverterConfigSchema;
|
|
120
|
-
|
|
121
|
-
export declare function createConverterConfig(
|
|
122
|
-
overrides?: Partial<ConverterConfigSchema>
|
|
123
|
-
): ConverterConfigSchema;
|
|
@@ -1,136 +0,0 @@
|
|
|
1
|
-
import { DEFAULT_RMS_NORM_EPS, SHARD_SIZE } from './manifest.schema.js';
|
|
2
|
-
import { DEFAULT_QUANTIZATION_DEFAULTS } from './quantization-defaults.schema.js';
|
|
3
|
-
import { MB } from './units.schema.js';
|
|
4
|
-
|
|
5
|
-
// =============================================================================
|
|
6
|
-
// Converter Config
|
|
7
|
-
// =============================================================================
|
|
8
|
-
|
|
9
|
-
export const DEFAULT_CONVERTER_QUANTIZATION_CONFIG = {
|
|
10
|
-
weights: null,
|
|
11
|
-
embeddings: null,
|
|
12
|
-
lmHead: null,
|
|
13
|
-
vision: DEFAULT_QUANTIZATION_DEFAULTS.visionDtype,
|
|
14
|
-
audio: DEFAULT_QUANTIZATION_DEFAULTS.audioDtype,
|
|
15
|
-
projector: DEFAULT_QUANTIZATION_DEFAULTS.projectorDtype,
|
|
16
|
-
// Q4K layout: 'row' (fused kernel compatible, fast) or 'col' (dequant fallback)
|
|
17
|
-
q4kLayout: 'row',
|
|
18
|
-
computePrecision: 'f16',
|
|
19
|
-
};
|
|
20
|
-
|
|
21
|
-
export const DEFAULT_CONVERTER_SHARDING_CONFIG = {
|
|
22
|
-
shardSizeBytes: SHARD_SIZE,
|
|
23
|
-
};
|
|
24
|
-
|
|
25
|
-
export const DEFAULT_CONVERTER_STREAMING_CONFIG = {
|
|
26
|
-
chunkSizeBytes: 64 * MB,
|
|
27
|
-
};
|
|
28
|
-
|
|
29
|
-
export const DEFAULT_CONVERTER_HTTP_CONFIG = {
|
|
30
|
-
allowDownloadFallback: true,
|
|
31
|
-
maxDownloadBytes: null,
|
|
32
|
-
};
|
|
33
|
-
|
|
34
|
-
export const DEFAULT_CONVERTER_WEIGHT_LAYOUT_CONFIG = {
|
|
35
|
-
transposeWeights: false,
|
|
36
|
-
fuseGateUp: false,
|
|
37
|
-
};
|
|
38
|
-
|
|
39
|
-
export const DEFAULT_CONVERTER_MANIFEST_CONFIG = {
|
|
40
|
-
hashAlgorithm: 'sha256',
|
|
41
|
-
optimizations: null,
|
|
42
|
-
conversion: null,
|
|
43
|
-
};
|
|
44
|
-
|
|
45
|
-
export const DEFAULT_CONVERTER_INFERENCE_CONFIG = {
|
|
46
|
-
defaultKernelPath: null,
|
|
47
|
-
sessionDefaults: null,
|
|
48
|
-
execution: null,
|
|
49
|
-
};
|
|
50
|
-
|
|
51
|
-
export const DEFAULT_GGUF_PARSER_DEFAULTS = {
|
|
52
|
-
contextLength: 2048,
|
|
53
|
-
attentionLayerNormEpsilon: DEFAULT_RMS_NORM_EPS,
|
|
54
|
-
attentionLayerNormRMSEpsilon: DEFAULT_RMS_NORM_EPS,
|
|
55
|
-
ropeFreqBase: 10000,
|
|
56
|
-
};
|
|
57
|
-
|
|
58
|
-
export const DEFAULT_CONVERTER_OUTPUT_CONFIG = {
|
|
59
|
-
modelBaseId: null,
|
|
60
|
-
dir: null,
|
|
61
|
-
baseDir: null,
|
|
62
|
-
textOnly: false,
|
|
63
|
-
fast: false,
|
|
64
|
-
};
|
|
65
|
-
|
|
66
|
-
export const DEFAULT_CONVERTER_PRESET_CONFIG = {
|
|
67
|
-
model: null,
|
|
68
|
-
};
|
|
69
|
-
|
|
70
|
-
export const DEFAULT_CONVERTER_EXECUTION_CONFIG = {
|
|
71
|
-
workers: 8,
|
|
72
|
-
workerCountPolicy: 'cap',
|
|
73
|
-
rowChunkRows: null,
|
|
74
|
-
rowChunkMinTensorBytes: 32 * MB,
|
|
75
|
-
maxInFlightJobs: null,
|
|
76
|
-
useGpuCast: false,
|
|
77
|
-
gpuCastMinTensorBytes: 32 * MB,
|
|
78
|
-
};
|
|
79
|
-
|
|
80
|
-
export const DEFAULT_CONVERTER_CONFIG = {
|
|
81
|
-
quantization: DEFAULT_CONVERTER_QUANTIZATION_CONFIG,
|
|
82
|
-
sharding: DEFAULT_CONVERTER_SHARDING_CONFIG,
|
|
83
|
-
streaming: DEFAULT_CONVERTER_STREAMING_CONFIG,
|
|
84
|
-
http: DEFAULT_CONVERTER_HTTP_CONFIG,
|
|
85
|
-
weightLayout: DEFAULT_CONVERTER_WEIGHT_LAYOUT_CONFIG,
|
|
86
|
-
manifest: DEFAULT_CONVERTER_MANIFEST_CONFIG,
|
|
87
|
-
inference: DEFAULT_CONVERTER_INFERENCE_CONFIG,
|
|
88
|
-
output: DEFAULT_CONVERTER_OUTPUT_CONFIG,
|
|
89
|
-
presets: DEFAULT_CONVERTER_PRESET_CONFIG,
|
|
90
|
-
};
|
|
91
|
-
|
|
92
|
-
export function createConverterConfig(overrides) {
|
|
93
|
-
if (!overrides) {
|
|
94
|
-
return {
|
|
95
|
-
quantization: { ...DEFAULT_CONVERTER_QUANTIZATION_CONFIG },
|
|
96
|
-
sharding: { ...DEFAULT_CONVERTER_SHARDING_CONFIG },
|
|
97
|
-
streaming: { ...DEFAULT_CONVERTER_STREAMING_CONFIG },
|
|
98
|
-
http: { ...DEFAULT_CONVERTER_HTTP_CONFIG },
|
|
99
|
-
weightLayout: { ...DEFAULT_CONVERTER_WEIGHT_LAYOUT_CONFIG },
|
|
100
|
-
manifest: { ...DEFAULT_CONVERTER_MANIFEST_CONFIG },
|
|
101
|
-
inference: { ...DEFAULT_CONVERTER_INFERENCE_CONFIG },
|
|
102
|
-
output: { ...DEFAULT_CONVERTER_OUTPUT_CONFIG },
|
|
103
|
-
presets: { ...DEFAULT_CONVERTER_PRESET_CONFIG },
|
|
104
|
-
};
|
|
105
|
-
}
|
|
106
|
-
|
|
107
|
-
return {
|
|
108
|
-
quantization: overrides.quantization
|
|
109
|
-
? { ...DEFAULT_CONVERTER_QUANTIZATION_CONFIG, ...overrides.quantization }
|
|
110
|
-
: { ...DEFAULT_CONVERTER_QUANTIZATION_CONFIG },
|
|
111
|
-
sharding: overrides.sharding
|
|
112
|
-
? { ...DEFAULT_CONVERTER_SHARDING_CONFIG, ...overrides.sharding }
|
|
113
|
-
: { ...DEFAULT_CONVERTER_SHARDING_CONFIG },
|
|
114
|
-
streaming: overrides.streaming
|
|
115
|
-
? { ...DEFAULT_CONVERTER_STREAMING_CONFIG, ...overrides.streaming }
|
|
116
|
-
: { ...DEFAULT_CONVERTER_STREAMING_CONFIG },
|
|
117
|
-
http: overrides.http
|
|
118
|
-
? { ...DEFAULT_CONVERTER_HTTP_CONFIG, ...overrides.http }
|
|
119
|
-
: { ...DEFAULT_CONVERTER_HTTP_CONFIG },
|
|
120
|
-
weightLayout: overrides.weightLayout
|
|
121
|
-
? { ...DEFAULT_CONVERTER_WEIGHT_LAYOUT_CONFIG, ...overrides.weightLayout }
|
|
122
|
-
: { ...DEFAULT_CONVERTER_WEIGHT_LAYOUT_CONFIG },
|
|
123
|
-
manifest: overrides.manifest
|
|
124
|
-
? { ...DEFAULT_CONVERTER_MANIFEST_CONFIG, ...overrides.manifest }
|
|
125
|
-
: { ...DEFAULT_CONVERTER_MANIFEST_CONFIG },
|
|
126
|
-
inference: overrides.inference
|
|
127
|
-
? { ...DEFAULT_CONVERTER_INFERENCE_CONFIG, ...overrides.inference }
|
|
128
|
-
: { ...DEFAULT_CONVERTER_INFERENCE_CONFIG },
|
|
129
|
-
output: overrides.output
|
|
130
|
-
? { ...DEFAULT_CONVERTER_OUTPUT_CONFIG, ...overrides.output }
|
|
131
|
-
: { ...DEFAULT_CONVERTER_OUTPUT_CONFIG },
|
|
132
|
-
presets: overrides.presets
|
|
133
|
-
? { ...DEFAULT_CONVERTER_PRESET_CONFIG, ...overrides.presets }
|
|
134
|
-
: { ...DEFAULT_CONVERTER_PRESET_CONFIG },
|
|
135
|
-
};
|
|
136
|
-
}
|
|
@@ -1,290 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* Debug Config Schema
|
|
3
|
-
*
|
|
4
|
-
* Configuration for the DOPPLER debug module, including log history limits,
|
|
5
|
-
* default log levels, trace categories, and decode step limits.
|
|
6
|
-
*
|
|
7
|
-
* @module config/schema/debug
|
|
8
|
-
*/
|
|
9
|
-
|
|
10
|
-
/**
|
|
11
|
-
* Configuration for log output destinations.
|
|
12
|
-
*
|
|
13
|
-
* Controls where logs are written: stdout, file, or both.
|
|
14
|
-
*/
|
|
15
|
-
export interface LogOutputConfigSchema {
|
|
16
|
-
/** Write logs to stdout/console (default: true) */
|
|
17
|
-
stdout: boolean;
|
|
18
|
-
/** Path to log file (null = no file output) */
|
|
19
|
-
file: string | null;
|
|
20
|
-
/** Append to existing file vs overwrite (default: true) */
|
|
21
|
-
append: boolean;
|
|
22
|
-
}
|
|
23
|
-
|
|
24
|
-
/** Default log output configuration */
|
|
25
|
-
export declare const DEFAULT_LOG_OUTPUT_CONFIG: LogOutputConfigSchema;
|
|
26
|
-
|
|
27
|
-
/**
|
|
28
|
-
* Configuration for log history retention.
|
|
29
|
-
*
|
|
30
|
-
* Controls how many log entries are kept in memory for debugging
|
|
31
|
-
* and diagnostic purposes.
|
|
32
|
-
*/
|
|
33
|
-
export interface LogHistoryConfigSchema {
|
|
34
|
-
/** Maximum number of log entries to retain in memory */
|
|
35
|
-
maxLogHistoryEntries: number;
|
|
36
|
-
}
|
|
37
|
-
|
|
38
|
-
/** Default log history configuration */
|
|
39
|
-
export declare const DEFAULT_LOG_HISTORY_CONFIG: LogHistoryConfigSchema;
|
|
40
|
-
|
|
41
|
-
/** Valid log levels */
|
|
42
|
-
export declare const LOG_LEVELS: readonly ['debug', 'verbose', 'info', 'warn', 'error', 'silent'];
|
|
43
|
-
|
|
44
|
-
/** Log level type */
|
|
45
|
-
export type LogLevel = typeof LOG_LEVELS[number];
|
|
46
|
-
|
|
47
|
-
/**
|
|
48
|
-
* Configuration for default log level.
|
|
49
|
-
*
|
|
50
|
-
* Controls the initial verbosity level when the debug module initializes.
|
|
51
|
-
*/
|
|
52
|
-
export interface LogLevelConfigSchema {
|
|
53
|
-
/** Default log level (debug, verbose, info, warn, error, silent) */
|
|
54
|
-
defaultLogLevel: LogLevel;
|
|
55
|
-
}
|
|
56
|
-
|
|
57
|
-
/** Default log level configuration */
|
|
58
|
-
export declare const DEFAULT_LOG_LEVEL_CONFIG: LogLevelConfigSchema;
|
|
59
|
-
|
|
60
|
-
/** Available trace categories */
|
|
61
|
-
export type TraceCategory =
|
|
62
|
-
| 'loader'
|
|
63
|
-
| 'kernels'
|
|
64
|
-
| 'logits'
|
|
65
|
-
| 'embed'
|
|
66
|
-
| 'attn'
|
|
67
|
-
| 'ffn'
|
|
68
|
-
| 'kv'
|
|
69
|
-
| 'sample'
|
|
70
|
-
| 'buffers'
|
|
71
|
-
| 'perf'
|
|
72
|
-
| 'all';
|
|
73
|
-
|
|
74
|
-
/**
|
|
75
|
-
* Configuration for trace output.
|
|
76
|
-
*
|
|
77
|
-
* Controls trace categories, output destination, and limits.
|
|
78
|
-
*/
|
|
79
|
-
export interface TraceConfigSchema {
|
|
80
|
-
/** Enable tracing (default: false) */
|
|
81
|
-
enabled: boolean;
|
|
82
|
-
/** Trace categories to enable (default: all) */
|
|
83
|
-
categories: TraceCategory[];
|
|
84
|
-
/** Filter to specific layer indices (null = all layers) */
|
|
85
|
-
layers: number[] | null;
|
|
86
|
-
/** Maximum decode steps to trace (0 = unlimited) */
|
|
87
|
-
maxDecodeSteps: number;
|
|
88
|
-
/** Path to trace file for JSONL output (null = no file) */
|
|
89
|
-
file: string | null;
|
|
90
|
-
}
|
|
91
|
-
|
|
92
|
-
/** Default trace configuration */
|
|
93
|
-
export declare const DEFAULT_TRACE_CONFIG: TraceConfigSchema;
|
|
94
|
-
|
|
95
|
-
/** Debug categories for pipeline debug-utils (kernel/layer inspection) */
|
|
96
|
-
export type PipelineDebugCategory =
|
|
97
|
-
| 'embed'
|
|
98
|
-
| 'layer'
|
|
99
|
-
| 'attn'
|
|
100
|
-
| 'ffn'
|
|
101
|
-
| 'kv'
|
|
102
|
-
| 'logits'
|
|
103
|
-
| 'sample'
|
|
104
|
-
| 'io'
|
|
105
|
-
| 'perf'
|
|
106
|
-
| 'kernel'
|
|
107
|
-
| 'all';
|
|
108
|
-
|
|
109
|
-
/**
|
|
110
|
-
* Pipeline debug configuration.
|
|
111
|
-
*
|
|
112
|
-
* Controls debug-utils categories and expensive readback helpers.
|
|
113
|
-
*/
|
|
114
|
-
export interface PipelineDebugConfigSchema {
|
|
115
|
-
/** Enable pipeline debug (default: false) */
|
|
116
|
-
enabled: boolean;
|
|
117
|
-
/** Debug categories to enable (default: none) */
|
|
118
|
-
categories: PipelineDebugCategory[];
|
|
119
|
-
/** Filter to specific layer indices (null = all layers) */
|
|
120
|
-
layers: number[] | null;
|
|
121
|
-
/** Maximum decode steps to log (0 = unlimited) */
|
|
122
|
-
maxDecodeSteps: number;
|
|
123
|
-
/** Warn if maxAbs exceeds this */
|
|
124
|
-
maxAbsThreshold: number;
|
|
125
|
-
/** Enable expensive GPU buffer stats */
|
|
126
|
-
bufferStats: boolean;
|
|
127
|
-
/** Maximum bytes to readback for debug samples (default: 512) */
|
|
128
|
-
readbackSampleSize: number;
|
|
129
|
-
}
|
|
130
|
-
|
|
131
|
-
/** Default pipeline debug configuration */
|
|
132
|
-
export declare const DEFAULT_PIPELINE_DEBUG_CONFIG: PipelineDebugConfigSchema;
|
|
133
|
-
|
|
134
|
-
/** Loader debug configuration (Q4K dequant and related probes). */
|
|
135
|
-
export interface LoaderDebugConfigSchema {
|
|
136
|
-
/** Enable loader debug behavior (default: false) */
|
|
137
|
-
enabled: boolean;
|
|
138
|
-
/** Force GPU dequant for Q4K tensors even when CPU fallback is eligible. */
|
|
139
|
-
forceGpuDequant: boolean;
|
|
140
|
-
/** Prefer CPU dequant for F32 output when eligible (default: false, GPU is preferred). */
|
|
141
|
-
preferCpuDequant: boolean;
|
|
142
|
-
/** Throw when CPU dequant fallback is taken. */
|
|
143
|
-
failOnCpuDequantPath: boolean;
|
|
144
|
-
/** Enable dtype-aware GPU-vs-CPU parity checks during Q4K dequant. */
|
|
145
|
-
runQ4KDequantParity: boolean;
|
|
146
|
-
/** Number of values to read back for parity checks. */
|
|
147
|
-
q4kDequantParitySamples: number;
|
|
148
|
-
}
|
|
149
|
-
|
|
150
|
-
/** Default loader debug configuration. */
|
|
151
|
-
export declare const DEFAULT_LOADER_DEBUG_CONFIG: LoaderDebugConfigSchema;
|
|
152
|
-
|
|
153
|
-
/** Matmul debug configuration (attention split/shape diagnostics). */
|
|
154
|
-
export interface MatmulDebugConfigSchema {
|
|
155
|
-
/** Enable matmul debug behavior (default: false) */
|
|
156
|
-
enabled: boolean;
|
|
157
|
-
/** Force split (non-fused) Q/K/V projection path for diagnostics. */
|
|
158
|
-
forceSplitQKV: boolean;
|
|
159
|
-
/** Validate B tensor layout/buffer bytes for attention projection roles. */
|
|
160
|
-
validateAttentionWeightBuffer: boolean;
|
|
161
|
-
/** Throw if validation fails due small B tensor. */
|
|
162
|
-
failOnSmallAttentionWeightBuffer: boolean;
|
|
163
|
-
/** Emit attention B-buffer diagnostics. */
|
|
164
|
-
logAttentionWeightBuffer: boolean;
|
|
165
|
-
/** Log first-8 projection output values for layer 0 decode (diagnostic). */
|
|
166
|
-
logProjectionValues: boolean;
|
|
167
|
-
}
|
|
168
|
-
|
|
169
|
-
/** Default matmul debug configuration. */
|
|
170
|
-
export declare const DEFAULT_MATMUL_DEBUG_CONFIG: MatmulDebugConfigSchema;
|
|
171
|
-
|
|
172
|
-
/**
|
|
173
|
-
* Profiler configuration.
|
|
174
|
-
*/
|
|
175
|
-
export interface ProfilerConfigSchema {
|
|
176
|
-
/** Enable GPU profiling */
|
|
177
|
-
enabled: boolean;
|
|
178
|
-
/** Maximum number of timestamp pairs to allocate */
|
|
179
|
-
queryCapacity: number;
|
|
180
|
-
/** Absolute upper bound for query capacity */
|
|
181
|
-
maxQueries: number;
|
|
182
|
-
/** Fallback query capacity when device limit is unavailable */
|
|
183
|
-
defaultQueryLimit: number;
|
|
184
|
-
/** Maximum samples retained per label */
|
|
185
|
-
maxSamples: number;
|
|
186
|
-
/** GPU timing sanity limit before falling back to CPU */
|
|
187
|
-
maxDurationMs: number;
|
|
188
|
-
/** Log every N decode profile steps (<=1 logs all steps) */
|
|
189
|
-
logEveryDecodeSteps: number;
|
|
190
|
-
}
|
|
191
|
-
|
|
192
|
-
/** Default profiler configuration */
|
|
193
|
-
export declare const DEFAULT_PROFILER_CONFIG: ProfilerConfigSchema;
|
|
194
|
-
|
|
195
|
-
/**
|
|
196
|
-
* Performance guard configuration.
|
|
197
|
-
*
|
|
198
|
-
* Controls GPU readback and tracking behavior.
|
|
199
|
-
*/
|
|
200
|
-
export interface PerfGuardsConfigSchema {
|
|
201
|
-
/** Allow GPU readbacks (default: true) */
|
|
202
|
-
allowGPUReadback: boolean;
|
|
203
|
-
/** Track queue.submit() calls (default: false) */
|
|
204
|
-
trackSubmitCount: boolean;
|
|
205
|
-
/** Track buffer allocations (default: false) */
|
|
206
|
-
trackAllocations: boolean;
|
|
207
|
-
/** Log expensive operations (default: false) */
|
|
208
|
-
logExpensiveOps: boolean;
|
|
209
|
-
/** Throw on blocked operations (default: false) */
|
|
210
|
-
strictMode: boolean;
|
|
211
|
-
}
|
|
212
|
-
|
|
213
|
-
/** Default performance guard configuration */
|
|
214
|
-
export declare const DEFAULT_PERF_GUARDS_CONFIG: PerfGuardsConfigSchema;
|
|
215
|
-
|
|
216
|
-
/** Pipeline probe stages */
|
|
217
|
-
export type ProbeStage =
|
|
218
|
-
| 'embed_out'
|
|
219
|
-
// Attention stages (per-layer)
|
|
220
|
-
| 'attn_input' // Input to attention (after residual from previous layer)
|
|
221
|
-
| 'attn_normed' // After input RMSNorm
|
|
222
|
-
| 'linear_qkv_proj' // Linear-attention fused QKV projection output
|
|
223
|
-
| 'linear_z_proj' // Linear-attention z projection output
|
|
224
|
-
| 'linear_a_proj' // Linear-attention a projection output
|
|
225
|
-
| 'linear_b_proj' // Linear-attention b projection output
|
|
226
|
-
| 'linear_core_out' // Linear-attention recurrent core output (before o_proj)
|
|
227
|
-
| 'q_proj' // Q projection output
|
|
228
|
-
| 'k_proj' // K projection output
|
|
229
|
-
| 'v_proj' // V projection output
|
|
230
|
-
| 'q_rope' // Q after RoPE
|
|
231
|
-
| 'k_rope' // K after RoPE
|
|
232
|
-
| 'attn_scores' // Attention scores (pre-softmax)
|
|
233
|
-
| 'attn_out' // Attention output (before o_proj)
|
|
234
|
-
| 'o_proj' // Output projection
|
|
235
|
-
| 'post_attn' // After attention residual
|
|
236
|
-
// FFN stages (per-layer)
|
|
237
|
-
| 'ffn_normed' // After post-attention RMSNorm
|
|
238
|
-
| 'ffn_in' // FFN gate/up input
|
|
239
|
-
| 'ffn_gate' // Gate projection output
|
|
240
|
-
| 'ffn_up' // Up projection output
|
|
241
|
-
| 'ffn_act' // After activation
|
|
242
|
-
| 'ffn_out' // Down projection output
|
|
243
|
-
| 'layer_out' // Final layer output (after FFN residual)
|
|
244
|
-
// Final stages
|
|
245
|
-
| 'pre_final_norm'
|
|
246
|
-
| 'final_norm'
|
|
247
|
-
| 'logits'
|
|
248
|
-
| 'logits_final';
|
|
249
|
-
|
|
250
|
-
/**
|
|
251
|
-
* Probe configuration for targeted value inspection.
|
|
252
|
-
*
|
|
253
|
-
* Probes read specific token/dimension values from GPU buffers at
|
|
254
|
-
* named pipeline stages.
|
|
255
|
-
*/
|
|
256
|
-
export interface ProbeConfigSchema {
|
|
257
|
-
/** Optional probe id (included in logs) */
|
|
258
|
-
id?: string;
|
|
259
|
-
/** Stage to probe */
|
|
260
|
-
stage: ProbeStage;
|
|
261
|
-
/** Restrict to specific layers (null = all layers) */
|
|
262
|
-
layers?: number[] | null;
|
|
263
|
-
/** Token indices to sample (null = default to token 0) */
|
|
264
|
-
tokens?: number[] | null;
|
|
265
|
-
/** Dimension indices to sample */
|
|
266
|
-
dims: number[];
|
|
267
|
-
/** Override trace category (defaults to stage category) */
|
|
268
|
-
category?: TraceCategory;
|
|
269
|
-
}
|
|
270
|
-
|
|
271
|
-
/**
|
|
272
|
-
* Complete debug configuration schema.
|
|
273
|
-
*
|
|
274
|
-
* Combines log output, log history, log level, and trace settings.
|
|
275
|
-
*/
|
|
276
|
-
export interface DebugConfigSchema {
|
|
277
|
-
logOutput: LogOutputConfigSchema;
|
|
278
|
-
logHistory: LogHistoryConfigSchema;
|
|
279
|
-
logLevel: LogLevelConfigSchema;
|
|
280
|
-
trace: TraceConfigSchema;
|
|
281
|
-
pipeline: PipelineDebugConfigSchema;
|
|
282
|
-
loader: LoaderDebugConfigSchema;
|
|
283
|
-
matmul: MatmulDebugConfigSchema;
|
|
284
|
-
probes: ProbeConfigSchema[];
|
|
285
|
-
profiler: ProfilerConfigSchema;
|
|
286
|
-
perfGuards: PerfGuardsConfigSchema;
|
|
287
|
-
}
|
|
288
|
-
|
|
289
|
-
/** Default debug configuration */
|
|
290
|
-
export declare const DEFAULT_DEBUG_CONFIG: DebugConfigSchema;
|
|
@@ -1,134 +0,0 @@
|
|
|
1
|
-
// =============================================================================
|
|
2
|
-
// Log Output Config
|
|
3
|
-
// =============================================================================
|
|
4
|
-
|
|
5
|
-
export const DEFAULT_LOG_OUTPUT_CONFIG = {
|
|
6
|
-
stdout: true,
|
|
7
|
-
file: null,
|
|
8
|
-
append: true,
|
|
9
|
-
};
|
|
10
|
-
|
|
11
|
-
// =============================================================================
|
|
12
|
-
// Log History Config
|
|
13
|
-
// =============================================================================
|
|
14
|
-
|
|
15
|
-
export const DEFAULT_LOG_HISTORY_CONFIG = {
|
|
16
|
-
maxLogHistoryEntries: 1000,
|
|
17
|
-
};
|
|
18
|
-
|
|
19
|
-
// =============================================================================
|
|
20
|
-
// Log Level Config
|
|
21
|
-
// =============================================================================
|
|
22
|
-
|
|
23
|
-
export const LOG_LEVELS = ['debug', 'verbose', 'info', 'warn', 'error', 'silent'];
|
|
24
|
-
|
|
25
|
-
export const DEFAULT_LOG_LEVEL_CONFIG = {
|
|
26
|
-
defaultLogLevel: 'info',
|
|
27
|
-
};
|
|
28
|
-
|
|
29
|
-
// =============================================================================
|
|
30
|
-
// Trace Config
|
|
31
|
-
// =============================================================================
|
|
32
|
-
|
|
33
|
-
export const DEFAULT_TRACE_CONFIG = {
|
|
34
|
-
enabled: false,
|
|
35
|
-
categories: ['all'],
|
|
36
|
-
layers: null,
|
|
37
|
-
maxDecodeSteps: 0,
|
|
38
|
-
file: null,
|
|
39
|
-
};
|
|
40
|
-
|
|
41
|
-
// =============================================================================
|
|
42
|
-
// Loader Debug Config
|
|
43
|
-
// =============================================================================
|
|
44
|
-
|
|
45
|
-
export const DEFAULT_LOADER_DEBUG_CONFIG = {
|
|
46
|
-
enabled: false,
|
|
47
|
-
forceGpuDequant: false,
|
|
48
|
-
preferCpuDequant: false,
|
|
49
|
-
failOnCpuDequantPath: false,
|
|
50
|
-
runQ4KDequantParity: false,
|
|
51
|
-
q4kDequantParitySamples: 256,
|
|
52
|
-
};
|
|
53
|
-
|
|
54
|
-
// =============================================================================
|
|
55
|
-
// Kernel Debug Config
|
|
56
|
-
// =============================================================================
|
|
57
|
-
|
|
58
|
-
export const DEFAULT_MATMUL_DEBUG_CONFIG = {
|
|
59
|
-
enabled: false,
|
|
60
|
-
forceSplitQKV: false,
|
|
61
|
-
validateAttentionWeightBuffer: false,
|
|
62
|
-
failOnSmallAttentionWeightBuffer: false,
|
|
63
|
-
logAttentionWeightBuffer: false,
|
|
64
|
-
logProjectionValues: false,
|
|
65
|
-
};
|
|
66
|
-
|
|
67
|
-
// =============================================================================
|
|
68
|
-
// Kernel Trace Config (kernel-trace.js anomaly detection)
|
|
69
|
-
// =============================================================================
|
|
70
|
-
|
|
71
|
-
export const DEFAULT_KERNEL_TRACE_CONFIG = {
|
|
72
|
-
layers: [],
|
|
73
|
-
breakOnAnomaly: false,
|
|
74
|
-
explosionThreshold: 10,
|
|
75
|
-
collapseThreshold: 1e-6,
|
|
76
|
-
maxSteps: 5000,
|
|
77
|
-
};
|
|
78
|
-
|
|
79
|
-
// =============================================================================
|
|
80
|
-
// Pipeline Debug Config (debug-utils)
|
|
81
|
-
// =============================================================================
|
|
82
|
-
|
|
83
|
-
export const DEFAULT_PIPELINE_DEBUG_CONFIG = {
|
|
84
|
-
enabled: false,
|
|
85
|
-
categories: [],
|
|
86
|
-
layers: null,
|
|
87
|
-
maxDecodeSteps: 0,
|
|
88
|
-
maxAbsThreshold: 10000,
|
|
89
|
-
bufferStats: false,
|
|
90
|
-
readbackSampleSize: 512,
|
|
91
|
-
};
|
|
92
|
-
|
|
93
|
-
// =============================================================================
|
|
94
|
-
// Profiler Config
|
|
95
|
-
// =============================================================================
|
|
96
|
-
|
|
97
|
-
export const DEFAULT_PROFILER_CONFIG = {
|
|
98
|
-
enabled: false,
|
|
99
|
-
queryCapacity: 256,
|
|
100
|
-
maxQueries: 16384,
|
|
101
|
-
defaultQueryLimit: 4096,
|
|
102
|
-
maxSamples: 100,
|
|
103
|
-
maxDurationMs: 60000,
|
|
104
|
-
logEveryDecodeSteps: 64,
|
|
105
|
-
};
|
|
106
|
-
|
|
107
|
-
// =============================================================================
|
|
108
|
-
// Perf Guard Config
|
|
109
|
-
// =============================================================================
|
|
110
|
-
|
|
111
|
-
export const DEFAULT_PERF_GUARDS_CONFIG = {
|
|
112
|
-
allowGPUReadback: true,
|
|
113
|
-
trackSubmitCount: false,
|
|
114
|
-
trackAllocations: false,
|
|
115
|
-
logExpensiveOps: false,
|
|
116
|
-
strictMode: false,
|
|
117
|
-
};
|
|
118
|
-
|
|
119
|
-
// =============================================================================
|
|
120
|
-
// Complete Debug Config
|
|
121
|
-
// =============================================================================
|
|
122
|
-
|
|
123
|
-
export const DEFAULT_DEBUG_CONFIG = {
|
|
124
|
-
logOutput: DEFAULT_LOG_OUTPUT_CONFIG,
|
|
125
|
-
logHistory: DEFAULT_LOG_HISTORY_CONFIG,
|
|
126
|
-
logLevel: DEFAULT_LOG_LEVEL_CONFIG,
|
|
127
|
-
trace: DEFAULT_TRACE_CONFIG,
|
|
128
|
-
pipeline: DEFAULT_PIPELINE_DEBUG_CONFIG,
|
|
129
|
-
loader: DEFAULT_LOADER_DEBUG_CONFIG,
|
|
130
|
-
matmul: DEFAULT_MATMUL_DEBUG_CONFIG,
|
|
131
|
-
probes: [],
|
|
132
|
-
profiler: DEFAULT_PROFILER_CONFIG,
|
|
133
|
-
perfGuards: DEFAULT_PERF_GUARDS_CONFIG,
|
|
134
|
-
};
|