@simulatte/doppler 0.1.5 → 0.1.7
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +126 -0
- package/README.md +25 -17
- package/package.json +20 -4
- package/src/adapters/adapter-registry.js +12 -1
- package/src/adapters/lora-loader.js +23 -6
- package/src/bridge/extension-client.d.ts +5 -0
- package/src/bridge/extension-client.js +40 -0
- package/src/bridge/index.d.ts +2 -1
- package/src/bridge/index.js +6 -4
- package/src/browser/browser-converter.js +26 -1
- package/src/browser/file-picker.js +6 -0
- package/src/browser/safetensors-parser-browser.js +84 -1
- package/src/browser/shard-io-browser.js +2 -2
- package/src/browser/tensor-source-download.js +8 -2
- package/src/browser/tensor-source-http.d.ts +1 -0
- package/src/browser/tensor-source-http.js +5 -1
- package/src/client/doppler-api.browser.js +20 -4
- package/src/client/doppler-api.js +19 -3
- package/src/client/doppler-provider/generation.js +12 -0
- package/src/client/doppler-provider/model-manager.d.ts +10 -0
- package/src/client/doppler-provider/model-manager.js +91 -19
- package/src/client/doppler-provider/source-runtime.d.ts +2 -1
- package/src/client/doppler-provider/source-runtime.js +132 -13
- package/src/client/doppler-registry.json +8 -7
- package/src/config/backward-registry-loader.js +17 -2
- package/src/config/execution-v0-contract-check.js +113 -15
- package/src/config/kernel-path-contract-check.js +57 -29
- package/src/config/kernel-path-loader.js +5 -36
- package/src/config/kernels/kernel-ref-digests.js +39 -39
- package/src/config/kernels/registry.js +14 -1
- package/src/config/kernels/registry.json +49 -7
- package/src/config/loader.d.ts +1 -1
- package/src/config/loader.js +43 -4
- package/src/config/merge-contract-check.js +59 -4
- package/src/config/merge-helpers.js +128 -7
- package/src/config/merge.d.ts +1 -0
- package/src/config/merge.js +28 -0
- package/src/config/param-validator.js +47 -2
- package/src/config/presets/kernel-paths/{gemma2-q4k-dequant-f32a.json → gemma2-q4k-dequant-f32a-nosubgroups.json} +3 -3
- package/src/config/presets/kernel-paths/gemma3-f16-fused-f32a-online-streamingprefill.json +223 -0
- package/src/config/presets/kernel-paths/{gemma3-q4k-dequant-f32a.json → gemma3-q4k-dequant-f32a-nosubgroups.json} +3 -3
- package/src/config/presets/kernel-paths/registry.json +29 -8
- package/src/config/presets/models/gemma2.json +2 -2
- package/src/config/presets/models/qwen3.json +9 -2
- package/src/config/presets/models/transformer.json +5 -0
- package/src/config/presets/runtime/experiments/bench/gemma3-bench-q4k.json +1 -1
- package/src/config/presets/runtime/experiments/debug/gemma3-debug-q4k.json +1 -1
- package/src/config/presets/runtime/experiments/verify/gemma3-verify.json +1 -1
- package/src/config/presets/runtime/kernels/dequant-f16-q4k.json +6 -13
- package/src/config/presets/runtime/kernels/dequant-f32-q4k.json +6 -13
- package/src/config/presets/runtime/kernels/embeddinggemma-q4k-dequant-f32a.json +37 -0
- package/src/config/presets/runtime/kernels/fused-q4k.json +6 -13
- package/src/config/presets/runtime/kernels/gemma2-q4k-dequant-f16a.json +33 -0
- package/src/config/presets/runtime/kernels/gemma2-q4k-dequant-f32a-nosubgroups.json +33 -0
- package/src/config/presets/runtime/kernels/gemma2-q4k-fused-f32a.json +33 -0
- package/src/config/presets/runtime/kernels/safe-q4k.json +6 -13
- package/src/config/presets/runtime/platform/metal-apple-q4k.json +1 -1
- package/src/config/required-inference-fields-contract-check.js +6 -0
- package/src/config/runtime.js +6 -1
- package/src/config/schema/debug.schema.d.ts +5 -0
- package/src/config/schema/doppler.schema.js +16 -21
- package/src/config/schema/inference-defaults.schema.js +6 -3
- package/src/config/schema/inference.schema.d.ts +9 -0
- package/src/config/schema/kernel-path.schema.d.ts +11 -1
- package/src/config/schema/kernel-thresholds.schema.js +12 -4
- package/src/config/schema/manifest.schema.d.ts +8 -1
- package/src/config/schema/manifest.schema.js +19 -3
- package/src/config/training-defaults.js +30 -22
- package/src/converter/conversion-plan.js +94 -9
- package/src/converter/core.d.ts +7 -0
- package/src/converter/core.js +14 -9
- package/src/converter/execution-v0-manifest.js +4 -1
- package/src/converter/index.d.ts +1 -0
- package/src/converter/index.js +1 -0
- package/src/converter/manifest-inference.js +43 -12
- package/src/converter/parsers/diffusion.js +0 -3
- package/src/converter/quantization-info.js +35 -15
- package/src/converter/rope-config.js +42 -0
- package/src/converter/shard-packer.d.ts +1 -1
- package/src/converter/shard-packer.js +4 -1
- package/src/debug/config.js +123 -11
- package/src/debug/signals.js +7 -1
- package/src/debug/tensor.d.ts +2 -0
- package/src/debug/tensor.js +13 -2
- package/src/distribution/p2p-control-plane.js +52 -12
- package/src/distribution/p2p-observability.js +43 -7
- package/src/distribution/p2p-webrtc-browser.js +20 -0
- package/src/distribution/shard-delivery.js +77 -26
- package/src/formats/gguf/types.js +33 -16
- package/src/formats/rdrr/groups.d.ts +12 -4
- package/src/formats/rdrr/groups.js +3 -6
- package/src/formats/rdrr/parsing.js +39 -2
- package/src/formats/rdrr/types.d.ts +2 -1
- package/src/gpu/command-recorder.js +86 -61
- package/src/gpu/device.d.ts +1 -0
- package/src/gpu/device.js +131 -19
- package/src/gpu/kernel-tuner/benchmarks.js +326 -316
- package/src/gpu/kernel-tuner/cache.js +71 -4
- package/src/gpu/kernel-tuner/tuner.js +22 -4
- package/src/gpu/kernels/attention.js +113 -34
- package/src/gpu/kernels/backward/adam.js +62 -58
- package/src/gpu/kernels/backward/attention_backward.js +257 -169
- package/src/gpu/kernels/backward/conv2d_backward.js +14 -1
- package/src/gpu/kernels/bias_add.wgsl +8 -6
- package/src/gpu/kernels/bias_add_f16.wgsl +8 -5
- package/src/gpu/kernels/cast.js +191 -149
- package/src/gpu/kernels/check-stop.js +33 -44
- package/src/gpu/kernels/conv2d.js +27 -17
- package/src/gpu/kernels/conv2d.wgsl +7 -8
- package/src/gpu/kernels/conv2d_f16.wgsl +7 -8
- package/src/gpu/kernels/cross_entropy_loss.js +21 -15
- package/src/gpu/kernels/depthwise_conv2d.js +37 -26
- package/src/gpu/kernels/depthwise_conv2d.wgsl +6 -9
- package/src/gpu/kernels/depthwise_conv2d_f16.wgsl +6 -9
- package/src/gpu/kernels/dequant.js +178 -126
- package/src/gpu/kernels/energy.d.ts +3 -21
- package/src/gpu/kernels/energy.js +111 -88
- package/src/gpu/kernels/feature-check.js +1 -1
- package/src/gpu/kernels/fused_ffn.js +84 -65
- package/src/gpu/kernels/fused_matmul_residual.js +56 -33
- package/src/gpu/kernels/fused_matmul_rmsnorm.js +62 -45
- package/src/gpu/kernels/gather.js +33 -15
- package/src/gpu/kernels/gelu.js +19 -11
- package/src/gpu/kernels/grouped_pointwise_conv2d.js +34 -23
- package/src/gpu/kernels/grouped_pointwise_conv2d.wgsl +6 -9
- package/src/gpu/kernels/grouped_pointwise_conv2d_f16.wgsl +6 -9
- package/src/gpu/kernels/groupnorm.js +34 -23
- package/src/gpu/kernels/kv-quantize.js +5 -2
- package/src/gpu/kernels/layernorm.js +35 -19
- package/src/gpu/kernels/logit-merge.js +5 -3
- package/src/gpu/kernels/matmul.js +83 -39
- package/src/gpu/kernels/modulate.js +23 -15
- package/src/gpu/kernels/moe.js +221 -175
- package/src/gpu/kernels/pixel_shuffle.js +22 -14
- package/src/gpu/kernels/pixel_shuffle.wgsl +4 -5
- package/src/gpu/kernels/pixel_shuffle_f16.wgsl +4 -5
- package/src/gpu/kernels/relu.js +31 -10
- package/src/gpu/kernels/relu.wgsl +2 -1
- package/src/gpu/kernels/relu_f16.wgsl +2 -1
- package/src/gpu/kernels/repeat_channels.js +25 -17
- package/src/gpu/kernels/repeat_channels.wgsl +4 -5
- package/src/gpu/kernels/repeat_channels_f16.wgsl +4 -5
- package/src/gpu/kernels/residual.js +69 -23
- package/src/gpu/kernels/residual.wgsl +6 -3
- package/src/gpu/kernels/residual_f16.wgsl +2 -1
- package/src/gpu/kernels/residual_f16_vec4.wgsl +2 -1
- package/src/gpu/kernels/residual_vec4.wgsl +2 -1
- package/src/gpu/kernels/rmsnorm.js +96 -28
- package/src/gpu/kernels/rmsnorm.wgsl +14 -6
- package/src/gpu/kernels/rmsnorm_f16.wgsl +10 -2
- package/src/gpu/kernels/rope.d.ts +2 -0
- package/src/gpu/kernels/rope.js +14 -1
- package/src/gpu/kernels/rope.wgsl +56 -40
- package/src/gpu/kernels/sample.js +27 -38
- package/src/gpu/kernels/sana_linear_attention.js +19 -12
- package/src/gpu/kernels/sana_linear_attention_apply.wgsl +4 -5
- package/src/gpu/kernels/sana_linear_attention_apply_f16.wgsl +4 -5
- package/src/gpu/kernels/sana_linear_attention_summary.wgsl +4 -0
- package/src/gpu/kernels/sana_linear_attention_summary_f16.wgsl +4 -0
- package/src/gpu/kernels/scale.js +18 -11
- package/src/gpu/kernels/shader-cache.js +4 -2
- package/src/gpu/kernels/silu.d.ts +1 -0
- package/src/gpu/kernels/silu.js +148 -82
- package/src/gpu/kernels/silu.wgsl +19 -9
- package/src/gpu/kernels/silu_f16.wgsl +19 -9
- package/src/gpu/kernels/softmax.js +44 -25
- package/src/gpu/kernels/split_qkv.js +23 -13
- package/src/gpu/kernels/transpose.js +31 -10
- package/src/gpu/kernels/transpose.wgsl +6 -5
- package/src/gpu/kernels/upsample2d.js +22 -13
- package/src/gpu/kernels/upsample2d.wgsl +6 -9
- package/src/gpu/kernels/upsample2d_f16.wgsl +6 -9
- package/src/gpu/kernels/utils.js +35 -13
- package/src/gpu/partitioned-buffer-pool.js +10 -2
- package/src/gpu/perf-guards.js +2 -9
- package/src/gpu/profiler.js +27 -22
- package/src/gpu/readback-utils.d.ts +16 -0
- package/src/gpu/readback-utils.js +41 -0
- package/src/gpu/submit-tracker.js +13 -0
- package/src/gpu/uniform-cache.d.ts +1 -0
- package/src/gpu/uniform-cache.js +30 -9
- package/src/hotswap/intent-bundle.js +6 -0
- package/src/hotswap/manifest.d.ts +10 -1
- package/src/hotswap/manifest.js +12 -2
- package/src/hotswap/runtime.js +30 -8
- package/src/index-browser.d.ts +44 -0
- package/src/index-browser.js +14 -0
- package/src/inference/browser-harness-contract-helpers.d.ts +5 -0
- package/src/inference/browser-harness-contract-helpers.js +28 -0
- package/src/inference/browser-harness-diffusion-energy-suites.d.ts +2 -0
- package/src/inference/browser-harness-diffusion-energy-suites.js +269 -0
- package/src/inference/browser-harness-model-helpers.d.ts +16 -0
- package/src/inference/browser-harness-model-helpers.js +217 -0
- package/src/inference/browser-harness-report-helpers.d.ts +7 -0
- package/src/inference/browser-harness-report-helpers.js +42 -0
- package/src/inference/browser-harness-runtime-helpers.d.ts +61 -0
- package/src/inference/browser-harness-runtime-helpers.js +415 -0
- package/src/inference/browser-harness-suite-helpers.d.ts +28 -0
- package/src/inference/browser-harness-suite-helpers.js +268 -0
- package/src/inference/browser-harness-text-helpers.d.ts +27 -0
- package/src/inference/browser-harness-text-helpers.js +788 -0
- package/src/inference/browser-harness.d.ts +6 -0
- package/src/inference/browser-harness.js +130 -1950
- package/src/inference/kv-cache/base.js +140 -94
- package/src/inference/kv-cache/tiered.js +5 -3
- package/src/inference/moe-router.js +88 -56
- package/src/inference/multi-model-network.js +5 -3
- package/src/inference/network-evolution.d.ts +11 -2
- package/src/inference/network-evolution.js +20 -21
- package/src/inference/pipelines/context.d.ts +3 -0
- package/src/inference/pipelines/context.js +142 -2
- package/src/inference/pipelines/diffusion/helpers.js +7 -2
- package/src/inference/pipelines/diffusion/pipeline.js +17 -7
- package/src/inference/pipelines/diffusion/sd3-transformer.js +10 -10
- package/src/inference/pipelines/diffusion/text-encoder-gpu.d.ts +5 -0
- package/src/inference/pipelines/diffusion/text-encoder-gpu.js +27 -15
- package/src/inference/pipelines/diffusion/vae.js +3 -7
- package/src/inference/pipelines/energy/pipeline.js +27 -21
- package/src/inference/pipelines/energy/quintel.d.ts +5 -0
- package/src/inference/pipelines/energy/quintel.js +11 -0
- package/src/inference/pipelines/energy-head/row-head-pipeline.js +17 -13
- package/src/inference/pipelines/structured/json-head-pipeline.js +26 -11
- package/src/inference/pipelines/text/attention/projections.js +151 -101
- package/src/inference/pipelines/text/attention/record.js +73 -10
- package/src/inference/pipelines/text/attention/run.js +73 -10
- package/src/inference/pipelines/text/chat-format.js +25 -1
- package/src/inference/pipelines/text/config.d.ts +4 -0
- package/src/inference/pipelines/text/config.js +71 -5
- package/src/inference/pipelines/text/embed.js +2 -8
- package/src/inference/pipelines/text/execution-plan.js +64 -50
- package/src/inference/pipelines/text/execution-v0-contract-helpers.d.ts +59 -0
- package/src/inference/pipelines/text/execution-v0-contract-helpers.js +937 -0
- package/src/inference/pipelines/text/execution-v0-runtime-builders.d.ts +15 -0
- package/src/inference/pipelines/text/execution-v0-runtime-builders.js +279 -0
- package/src/inference/pipelines/text/execution-v0.js +78 -1002
- package/src/inference/pipelines/text/ffn/standard.js +3 -0
- package/src/inference/pipelines/text/generator-steps.d.ts +46 -0
- package/src/inference/pipelines/text/generator-steps.js +298 -207
- package/src/inference/pipelines/text/generator.js +6 -23
- package/src/inference/pipelines/text/init.d.ts +4 -0
- package/src/inference/pipelines/text/init.js +134 -29
- package/src/inference/pipelines/text/kernel-path-auto-select.js +2 -0
- package/src/inference/pipelines/text/kernel-trace.d.ts +2 -0
- package/src/inference/pipelines/text/kernel-trace.js +6 -0
- package/src/inference/pipelines/text/layer.js +14 -9
- package/src/inference/pipelines/text/linear-attention.d.ts +10 -0
- package/src/inference/pipelines/text/linear-attention.js +80 -6
- package/src/inference/pipelines/text/logits/gpu.js +10 -5
- package/src/inference/pipelines/text/logits/index.js +10 -11
- package/src/inference/pipelines/text/logits/utils.d.ts +7 -0
- package/src/inference/pipelines/text/logits/utils.js +9 -0
- package/src/inference/pipelines/text/lora-apply.js +50 -32
- package/src/inference/pipelines/text/model-load.js +279 -104
- package/src/inference/pipelines/text/moe-cache.js +5 -4
- package/src/inference/pipelines/text/moe-cpu-gptoss.js +74 -69
- package/src/inference/pipelines/text/moe-cpu.js +42 -38
- package/src/inference/pipelines/text/moe-gpu.js +110 -86
- package/src/inference/pipelines/text/ops.js +90 -90
- package/src/inference/pipelines/text/probes.js +9 -9
- package/src/inference/pipelines/text/weights.js +17 -7
- package/src/inference/pipelines/text.js +17 -1
- package/src/inference/speculative.d.ts +2 -2
- package/src/inference/speculative.js +4 -18
- package/src/inference/test-harness.d.ts +1 -1
- package/src/inference/test-harness.js +15 -5
- package/src/inference/tokenizer.d.ts +0 -5
- package/src/inference/tokenizer.js +4 -23
- package/src/inference/tokenizers/bpe.js +9 -0
- package/src/inference/tokenizers/bundled.js +176 -33
- package/src/inference/tokenizers/sentencepiece.js +12 -0
- package/src/loader/doppler-loader.js +38 -22
- package/src/loader/dtype-utils.js +3 -44
- package/src/loader/embedding-loader.js +7 -3
- package/src/loader/experts/expert-cache.js +13 -6
- package/src/loader/experts/expert-loader.js +10 -6
- package/src/loader/final-weights-loader.js +8 -4
- package/src/loader/layer-loader.js +2 -1
- package/src/loader/loader-state.js +2 -2
- package/src/loader/memory-monitor.js +8 -0
- package/src/loader/multi-model-loader.d.ts +14 -0
- package/src/loader/multi-model-loader.js +70 -24
- package/src/loader/shard-cache.js +81 -12
- package/src/loader/shard-resolver.js +25 -3
- package/src/loader/tensors/tensor-loader.js +209 -144
- package/src/loader/tensors/tensor-reader.js +76 -19
- package/src/loader/weight-downcast.js +1 -1
- package/src/memory/buffer-pool.d.ts +9 -1
- package/src/memory/buffer-pool.js +109 -44
- package/src/memory/unified-detect.js +1 -1
- package/src/rules/inference/kernel-path.rules.json +24 -8
- package/src/rules/rule-registry.js +25 -1
- package/src/rules/tooling/command-runtime.rules.json +18 -0
- package/src/storage/backends/opfs-store.js +68 -24
- package/src/storage/downloader.js +364 -83
- package/src/storage/index.d.ts +3 -0
- package/src/storage/index.js +3 -0
- package/src/storage/preflight.d.ts +2 -2
- package/src/storage/preflight.js +24 -2
- package/src/storage/quickstart-downloader.js +11 -5
- package/src/storage/registry.js +10 -4
- package/src/storage/reports.js +1 -1
- package/src/storage/shard-manager.d.ts +15 -1
- package/src/storage/shard-manager.js +51 -3
- package/src/storage/source-artifact-store.d.ts +52 -0
- package/src/storage/source-artifact-store.js +234 -0
- package/src/tooling/command-api-constants.d.ts +9 -0
- package/src/tooling/command-api-constants.js +9 -0
- package/src/tooling/command-api-family-normalizers.d.ts +9 -0
- package/src/tooling/command-api-family-normalizers.js +343 -0
- package/src/tooling/command-api-helpers.d.ts +25 -0
- package/src/tooling/command-api-helpers.js +262 -0
- package/src/tooling/command-api.d.ts +27 -1
- package/src/tooling/command-api.js +26 -473
- package/src/tooling/command-envelope.js +4 -1
- package/src/tooling/command-runner-shared.js +52 -18
- package/src/tooling/lean-execution-contract.js +150 -3
- package/src/tooling/node-browser-command-runner.d.ts +4 -0
- package/src/tooling/node-browser-command-runner.js +218 -273
- package/src/tooling/node-command-runner.js +44 -3
- package/src/tooling/node-converter.js +27 -1
- package/src/tooling/node-source-runtime.d.ts +1 -1
- package/src/tooling/node-source-runtime.js +84 -3
- package/src/tooling/node-webgpu.js +30 -105
- package/src/tooling/opfs-cache.js +21 -4
- package/src/tooling/runtime-input-composition.d.ts +38 -0
- package/src/tooling/runtime-input-composition.js +86 -0
- package/src/tooling/source-runtime-bundle.d.ts +40 -5
- package/src/tooling/source-runtime-bundle.js +261 -34
- package/src/tooling/source-runtime-materializer.d.ts +6 -0
- package/src/tooling/source-runtime-materializer.js +93 -0
- package/src/training/attention-backward.js +32 -17
- package/src/training/autograd.js +80 -52
- package/src/training/checkpoint-watch.d.ts +8 -0
- package/src/training/checkpoint-watch.js +139 -0
- package/src/training/checkpoint.d.ts +6 -1
- package/src/training/checkpoint.js +46 -7
- package/src/training/clip.js +2 -1
- package/src/training/datasets/token-batch.js +20 -8
- package/src/training/distillation/artifacts.d.ts +71 -0
- package/src/training/distillation/artifacts.js +132 -0
- package/src/training/distillation/checkpoint-watch.d.ts +10 -0
- package/src/training/distillation/checkpoint-watch.js +58 -0
- package/src/training/distillation/dataset.d.ts +59 -0
- package/src/training/distillation/dataset.js +337 -0
- package/src/training/distillation/eval.d.ts +34 -0
- package/src/training/distillation/eval.js +310 -0
- package/src/training/distillation/index.d.ts +29 -0
- package/src/training/distillation/index.js +29 -0
- package/src/training/distillation/runtime.d.ts +20 -0
- package/src/training/distillation/runtime.js +121 -0
- package/src/training/distillation/scoreboard.d.ts +6 -0
- package/src/training/distillation/scoreboard.js +8 -0
- package/src/training/distillation/stage-a.d.ts +45 -0
- package/src/training/distillation/stage-a.js +338 -0
- package/src/training/distillation/stage-b.d.ts +24 -0
- package/src/training/distillation/stage-b.js +20 -0
- package/src/training/distillation/student-fixture.d.ts +22 -0
- package/src/training/distillation/student-fixture.js +846 -0
- package/src/training/distillation/suite-data.d.ts +45 -0
- package/src/training/distillation/suite-data.js +189 -0
- package/src/training/index.d.ts +10 -0
- package/src/training/index.js +10 -0
- package/src/training/lora-pipeline.d.ts +40 -0
- package/src/training/lora-pipeline.js +793 -0
- package/src/training/lora.js +26 -12
- package/src/training/loss.js +5 -6
- package/src/training/objectives/cross_entropy.js +2 -5
- package/src/training/objectives/distill_kd.js +4 -8
- package/src/training/objectives/distill_triplet.js +4 -8
- package/src/training/objectives/ul_stage2_base.js +4 -8
- package/src/training/operator-artifacts.d.ts +62 -0
- package/src/training/operator-artifacts.js +140 -0
- package/src/training/operator-command.d.ts +5 -0
- package/src/training/operator-command.js +455 -0
- package/src/training/operator-eval.d.ts +48 -0
- package/src/training/operator-eval.js +230 -0
- package/src/training/operator-scoreboard.d.ts +5 -0
- package/src/training/operator-scoreboard.js +44 -0
- package/src/training/optimizer.js +19 -7
- package/src/training/runner.d.ts +52 -0
- package/src/training/runner.js +31 -5
- package/src/training/suite.d.ts +112 -0
- package/src/training/suite.js +24 -984
- package/src/training/tensor-factory.d.ts +9 -0
- package/src/training/tensor-factory.js +13 -0
- package/src/training/trainer.js +3 -5
- package/src/training/ul_dataset.js +3 -5
- package/src/training/workloads.d.ts +164 -0
- package/src/training/workloads.js +530 -0
- package/src/version.js +1 -1
- package/tools/convert-safetensors-node.js +22 -16
- package/tools/doppler-cli.js +179 -63
|
@@ -0,0 +1,338 @@
|
|
|
1
|
+
import { mkdir } from 'node:fs/promises';
|
|
2
|
+
import { resolve } from 'node:path';
|
|
3
|
+
|
|
4
|
+
import { AdamOptimizer } from '../optimizer.js';
|
|
5
|
+
import { crossEntropyLoss } from '../loss.js';
|
|
6
|
+
import { clipGradients } from '../clip.js';
|
|
7
|
+
import { TrainingRunner } from '../runner.js';
|
|
8
|
+
import {
|
|
9
|
+
createDistillRuntimeContext,
|
|
10
|
+
createDistillStudentRuntimeModelFixture,
|
|
11
|
+
loadDistillDatasetFromJsonl,
|
|
12
|
+
resolveDistillDataScope,
|
|
13
|
+
} from '../suite.js';
|
|
14
|
+
import { loadCanonicalTranslationDataset } from './dataset.js';
|
|
15
|
+
import { evaluateDistillationModel } from './eval.js';
|
|
16
|
+
import {
|
|
17
|
+
buildDistillArtifactBase,
|
|
18
|
+
writeDistillCheckpointComplete,
|
|
19
|
+
writeDistillCheckpointMetadata,
|
|
20
|
+
writeDistillStageManifest,
|
|
21
|
+
} from './artifacts.js';
|
|
22
|
+
import { appendDistillationScoreboardRow } from './scoreboard.js';
|
|
23
|
+
import { buildDistillationTrainingConfigFromWorkload } from './runtime.js';
|
|
24
|
+
|
|
25
|
+
function padStep(step) {
|
|
26
|
+
return String(step).padStart(6, '0');
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
function resolveComparableMetric(report, metric) {
|
|
30
|
+
if (!report || typeof report !== 'object') return null;
|
|
31
|
+
const direct = report[metric];
|
|
32
|
+
if (typeof direct === 'number' && Number.isFinite(direct)) {
|
|
33
|
+
return direct;
|
|
34
|
+
}
|
|
35
|
+
const metrics = report.metrics && typeof report.metrics === 'object' ? report.metrics : null;
|
|
36
|
+
const nested = metrics?.[metric];
|
|
37
|
+
if (typeof nested === 'number' && Number.isFinite(nested)) {
|
|
38
|
+
return nested;
|
|
39
|
+
}
|
|
40
|
+
if (metrics?.[metric]?.score != null && Number.isFinite(metrics[metric].score)) {
|
|
41
|
+
return metrics[metric].score;
|
|
42
|
+
}
|
|
43
|
+
return null;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
function selectBestReport(reports, metric, goal) {
|
|
47
|
+
const normalizedGoal = String(goal || 'max').trim();
|
|
48
|
+
let best = null;
|
|
49
|
+
let bestValue = null;
|
|
50
|
+
for (const report of reports) {
|
|
51
|
+
const value = resolveComparableMetric(report, metric);
|
|
52
|
+
if (!Number.isFinite(value)) continue;
|
|
53
|
+
if (best === null) {
|
|
54
|
+
best = report;
|
|
55
|
+
bestValue = value;
|
|
56
|
+
continue;
|
|
57
|
+
}
|
|
58
|
+
const better = normalizedGoal === 'min'
|
|
59
|
+
? value < bestValue
|
|
60
|
+
: value > bestValue;
|
|
61
|
+
if (better) {
|
|
62
|
+
best = report;
|
|
63
|
+
bestValue = value;
|
|
64
|
+
}
|
|
65
|
+
}
|
|
66
|
+
return best;
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
function shouldEvalOnCheckpoint(stageEntry) {
|
|
70
|
+
const schedule = String(stageEntry?.evalSchedule || 'on_checkpoint').trim();
|
|
71
|
+
return schedule === 'on_checkpoint';
|
|
72
|
+
}
|
|
73
|
+
|
|
74
|
+
function shouldEvalAtEnd(stageEntry) {
|
|
75
|
+
const schedule = String(stageEntry?.evalSchedule || 'on_checkpoint').trim();
|
|
76
|
+
return schedule === 'final';
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
export async function runDistillationStage(options) {
|
|
80
|
+
const loadedWorkload = options.loadedWorkload;
|
|
81
|
+
const workload = loadedWorkload.workload;
|
|
82
|
+
const stageEntry = options.stageEntry;
|
|
83
|
+
const datasetPath = options.datasetPath || workload.datasetPath;
|
|
84
|
+
const legacyArtifactDir = options.legacyArtifactDir || resolve(options.layout.runRoot, 'legacy-stage-artifacts');
|
|
85
|
+
if (workload.training.batchSize !== 1) {
|
|
86
|
+
throw new Error('Distillation stage currently requires training.batchSize=1.');
|
|
87
|
+
}
|
|
88
|
+
if (workload.training.accumSteps !== 1) {
|
|
89
|
+
throw new Error('Distillation stage currently requires training.accumSteps=1.');
|
|
90
|
+
}
|
|
91
|
+
const configBundle = buildDistillationTrainingConfigFromWorkload(loadedWorkload, stageEntry, {
|
|
92
|
+
datasetPath,
|
|
93
|
+
artifactDir: legacyArtifactDir,
|
|
94
|
+
stageAArtifact: options.stageAArtifact || null,
|
|
95
|
+
stageAArtifactHash: options.stageAArtifactHash || null,
|
|
96
|
+
});
|
|
97
|
+
const distillDataScope = resolveDistillDataScope({
|
|
98
|
+
distillSourceLangs: workload.pipeline.sourceLangs,
|
|
99
|
+
distillTargetLangs: workload.pipeline.targetLangs,
|
|
100
|
+
distillPairAllowlist: workload.pipeline.pairAllowlist,
|
|
101
|
+
strictPairContract: workload.pipeline.strictPairContract === true,
|
|
102
|
+
}, configBundle.trainingConfig.training);
|
|
103
|
+
const distillDatasetReport = await loadDistillDatasetFromJsonl(datasetPath, distillDataScope);
|
|
104
|
+
if (!distillDatasetReport) {
|
|
105
|
+
throw new Error(`Unable to resolve distillation dataset "${datasetPath}".`);
|
|
106
|
+
}
|
|
107
|
+
const canonicalDataset = await loadCanonicalTranslationDataset(datasetPath, {
|
|
108
|
+
strictPairContract: workload.pipeline.strictPairContract === true,
|
|
109
|
+
sourceLangs: workload.pipeline.sourceLangs,
|
|
110
|
+
targetLangs: workload.pipeline.targetLangs,
|
|
111
|
+
pairAllowlist: workload.pipeline.pairAllowlist,
|
|
112
|
+
});
|
|
113
|
+
const distillRuntime = await createDistillRuntimeContext({
|
|
114
|
+
teacherModelId: workload.teacherModelId,
|
|
115
|
+
studentModelId: workload.studentModelId,
|
|
116
|
+
trainingStage: configBundle.internalStage,
|
|
117
|
+
studentGraphMode: workload.pipeline.studentGraphMode,
|
|
118
|
+
}, configBundle.trainingConfig.training);
|
|
119
|
+
let fixture = null;
|
|
120
|
+
try {
|
|
121
|
+
fixture = await createDistillStudentRuntimeModelFixture({
|
|
122
|
+
training: configBundle.trainingConfig.training,
|
|
123
|
+
}, {
|
|
124
|
+
distillRuntime,
|
|
125
|
+
studentGraphMode: workload.pipeline.studentGraphMode,
|
|
126
|
+
});
|
|
127
|
+
const stageId = stageEntry.id;
|
|
128
|
+
const stageCheckpointsDir = resolve(options.layout.checkpoints, stageId);
|
|
129
|
+
await mkdir(stageCheckpointsDir, { recursive: true });
|
|
130
|
+
const dataset = distillDatasetReport.createDataset({
|
|
131
|
+
batchSize: workload.training.batchSize,
|
|
132
|
+
shuffle: false,
|
|
133
|
+
seed: workload.seed,
|
|
134
|
+
distillRuntime,
|
|
135
|
+
});
|
|
136
|
+
const checkpointArtifacts = [];
|
|
137
|
+
const evalReports = [];
|
|
138
|
+
const runner = new TrainingRunner(fixture.config, {
|
|
139
|
+
optimizer: new AdamOptimizer(fixture.config),
|
|
140
|
+
crossEntropyLoss,
|
|
141
|
+
clipGradients,
|
|
142
|
+
resolveCheckpointKey({ step }) {
|
|
143
|
+
const checkpointId = `checkpoint-${padStep(step)}`;
|
|
144
|
+
return resolve(stageCheckpointsDir, checkpointId, 'state.json');
|
|
145
|
+
},
|
|
146
|
+
onCheckpoint: async (checkpoint) => {
|
|
147
|
+
const checkpointId = `checkpoint-${padStep(checkpoint.step)}`;
|
|
148
|
+
const checkpointBase = buildDistillArtifactBase(loadedWorkload, {
|
|
149
|
+
prefix: 'dst_ckpt',
|
|
150
|
+
artifactType: 'training_checkpoint',
|
|
151
|
+
datasetPath: distillDatasetReport.absolutePath,
|
|
152
|
+
datasetHash: canonicalDataset.canonicalHash,
|
|
153
|
+
stage: stageId,
|
|
154
|
+
checkpointStep: checkpoint.step,
|
|
155
|
+
configHash: configBundle.trainingConfigHash,
|
|
156
|
+
parentArtifacts: options.parentArtifacts || [],
|
|
157
|
+
});
|
|
158
|
+
const metadataPayload = {
|
|
159
|
+
...checkpointBase,
|
|
160
|
+
checkpointId,
|
|
161
|
+
checkpointPath: checkpoint.path,
|
|
162
|
+
step: checkpoint.step,
|
|
163
|
+
epoch: checkpoint.epoch,
|
|
164
|
+
batch: checkpoint.batch,
|
|
165
|
+
optimizerStatePresent: true,
|
|
166
|
+
schedulerStatePresent: workload.training.optimizer.scheduler.enabled === true,
|
|
167
|
+
stageArtifact: options.stageAArtifact || null,
|
|
168
|
+
resumeLineage: checkpoint.metadata?.lineage || null,
|
|
169
|
+
};
|
|
170
|
+
const metadataFile = await writeDistillCheckpointMetadata(
|
|
171
|
+
options.layout,
|
|
172
|
+
stageId,
|
|
173
|
+
checkpointId,
|
|
174
|
+
metadataPayload
|
|
175
|
+
);
|
|
176
|
+
const completeFile = await writeDistillCheckpointComplete(
|
|
177
|
+
options.layout,
|
|
178
|
+
stageId,
|
|
179
|
+
checkpointId,
|
|
180
|
+
{
|
|
181
|
+
...metadataPayload,
|
|
182
|
+
metadataPath: metadataFile.path,
|
|
183
|
+
finalized: true,
|
|
184
|
+
}
|
|
185
|
+
);
|
|
186
|
+
checkpointArtifacts.push({
|
|
187
|
+
checkpointId,
|
|
188
|
+
checkpointPath: checkpoint.path,
|
|
189
|
+
metadataPath: metadataFile.path,
|
|
190
|
+
completePath: completeFile.path,
|
|
191
|
+
step: checkpoint.step,
|
|
192
|
+
});
|
|
193
|
+
if (!shouldEvalOnCheckpoint(stageEntry)) {
|
|
194
|
+
return;
|
|
195
|
+
}
|
|
196
|
+
const reports = await evaluateDistillationModel({
|
|
197
|
+
loadedWorkload,
|
|
198
|
+
layout: options.layout,
|
|
199
|
+
stageId,
|
|
200
|
+
checkpointId,
|
|
201
|
+
checkpointStep: checkpoint.step,
|
|
202
|
+
checkpointPath: checkpoint.path,
|
|
203
|
+
distillRuntime,
|
|
204
|
+
model: fixture.model,
|
|
205
|
+
configHash: configBundle.trainingConfigHash,
|
|
206
|
+
parentArtifacts: [{
|
|
207
|
+
artifactType: 'training_checkpoint',
|
|
208
|
+
path: metadataFile.path,
|
|
209
|
+
checkpointId,
|
|
210
|
+
}],
|
|
211
|
+
});
|
|
212
|
+
for (const report of reports) {
|
|
213
|
+
evalReports.push(report);
|
|
214
|
+
await appendDistillationScoreboardRow(options.layout, stageId, {
|
|
215
|
+
artifactType: 'training_scoreboard',
|
|
216
|
+
schemaVersion: 1,
|
|
217
|
+
generatedAt: new Date().toISOString(),
|
|
218
|
+
stage: stageId,
|
|
219
|
+
checkpointId,
|
|
220
|
+
checkpointStep: checkpoint.step,
|
|
221
|
+
evalDatasetId: report.evalDatasetId,
|
|
222
|
+
selectionMetric: stageEntry.selectionMetric,
|
|
223
|
+
selectionGoal: stageEntry.selectionGoal,
|
|
224
|
+
primaryMetric: report.primaryMetric,
|
|
225
|
+
primaryScore: report.primaryScore,
|
|
226
|
+
bleu: report.bleu,
|
|
227
|
+
chrf: report.chrf,
|
|
228
|
+
reportPath: report.reportPath || null,
|
|
229
|
+
metrics: {
|
|
230
|
+
bleu: report.bleu,
|
|
231
|
+
chrf: report.chrf,
|
|
232
|
+
primaryScore: report.primaryScore,
|
|
233
|
+
},
|
|
234
|
+
}, {
|
|
235
|
+
selectionMetric: stageEntry.selectionMetric,
|
|
236
|
+
selectionGoal: stageEntry.selectionGoal,
|
|
237
|
+
});
|
|
238
|
+
}
|
|
239
|
+
},
|
|
240
|
+
});
|
|
241
|
+
const metrics = await runner.run(fixture.model, dataset, {
|
|
242
|
+
epochs: 1,
|
|
243
|
+
batchSize: workload.training.batchSize,
|
|
244
|
+
shuffle: false,
|
|
245
|
+
maxSteps: stageEntry.steps,
|
|
246
|
+
checkpointEvery: stageEntry.checkpointEvery,
|
|
247
|
+
modelId: workload.studentModelId,
|
|
248
|
+
trainingStage: configBundle.internalStage,
|
|
249
|
+
runtimePreset: null,
|
|
250
|
+
command: 'distill',
|
|
251
|
+
surface: 'node',
|
|
252
|
+
timestamp: options.timestamp || null,
|
|
253
|
+
distillArtifactDir: legacyArtifactDir,
|
|
254
|
+
stageAArtifact: options.stageAArtifact || null,
|
|
255
|
+
stageAArtifactHash: options.stageAArtifactHash || null,
|
|
256
|
+
teacherModelId: workload.teacherModelId,
|
|
257
|
+
studentModelId: workload.studentModelId,
|
|
258
|
+
distillDatasetId: workload.datasetId,
|
|
259
|
+
distillDatasetPath: distillDatasetReport.absolutePath,
|
|
260
|
+
distillSourceLangs: workload.pipeline.sourceLangs,
|
|
261
|
+
distillTargetLangs: workload.pipeline.targetLangs,
|
|
262
|
+
distillPairAllowlist: workload.pipeline.pairAllowlist,
|
|
263
|
+
strictPairContract: workload.pipeline.strictPairContract === true,
|
|
264
|
+
});
|
|
265
|
+
if (shouldEvalAtEnd(stageEntry) && runner.lastCheckpoint) {
|
|
266
|
+
const finalCheckpointId = `checkpoint-${padStep(runner.lastCheckpoint.step)}`;
|
|
267
|
+
const reports = await evaluateDistillationModel({
|
|
268
|
+
loadedWorkload,
|
|
269
|
+
layout: options.layout,
|
|
270
|
+
stageId,
|
|
271
|
+
checkpointId: finalCheckpointId,
|
|
272
|
+
checkpointStep: runner.lastCheckpoint.step,
|
|
273
|
+
checkpointPath: runner.lastCheckpoint.path || null,
|
|
274
|
+
distillRuntime,
|
|
275
|
+
model: fixture.model,
|
|
276
|
+
configHash: configBundle.trainingConfigHash,
|
|
277
|
+
});
|
|
278
|
+
evalReports.push(...reports);
|
|
279
|
+
}
|
|
280
|
+
const bestReport = selectBestReport(
|
|
281
|
+
evalReports,
|
|
282
|
+
stageEntry.selectionMetric,
|
|
283
|
+
stageEntry.selectionGoal
|
|
284
|
+
);
|
|
285
|
+
const stageManifestPayload = {
|
|
286
|
+
...buildDistillArtifactBase(loadedWorkload, {
|
|
287
|
+
prefix: 'dst_stage',
|
|
288
|
+
artifactType: 'distill_stage_manifest',
|
|
289
|
+
datasetPath: distillDatasetReport.absolutePath,
|
|
290
|
+
datasetHash: canonicalDataset.canonicalHash,
|
|
291
|
+
stage: stageId,
|
|
292
|
+
checkpointStep: runner.lastCheckpoint?.step ?? null,
|
|
293
|
+
configHash: configBundle.trainingConfigHash,
|
|
294
|
+
parentArtifacts: options.parentArtifacts || [],
|
|
295
|
+
}),
|
|
296
|
+
stageId,
|
|
297
|
+
trainingStage: configBundle.internalStage,
|
|
298
|
+
objective: stageEntry.objective,
|
|
299
|
+
stagePlanEntry: stageEntry,
|
|
300
|
+
stepCount: Array.isArray(metrics) ? metrics.length : 0,
|
|
301
|
+
checkpointCount: checkpointArtifacts.length,
|
|
302
|
+
bestCheckpointId: bestReport?.checkpointId || null,
|
|
303
|
+
selectionMetric: stageEntry.selectionMetric,
|
|
304
|
+
selectionGoal: stageEntry.selectionGoal,
|
|
305
|
+
checkpointArtifacts,
|
|
306
|
+
evalReports: evalReports.map((report) => ({
|
|
307
|
+
checkpointId: report.checkpointId,
|
|
308
|
+
evalDatasetId: report.evalDatasetId,
|
|
309
|
+
reportPath: report.reportPath || null,
|
|
310
|
+
primaryMetric: report.primaryMetric,
|
|
311
|
+
primaryScore: report.primaryScore,
|
|
312
|
+
bleu: report.bleu,
|
|
313
|
+
chrf: report.chrf,
|
|
314
|
+
})),
|
|
315
|
+
legacyArtifact: runner.lastArtifact || null,
|
|
316
|
+
lastCheckpoint: runner.lastCheckpoint || null,
|
|
317
|
+
};
|
|
318
|
+
const stageManifest = await writeDistillStageManifest(options.layout, stageManifestPayload);
|
|
319
|
+
return {
|
|
320
|
+
stageId,
|
|
321
|
+
trainingStage: configBundle.internalStage,
|
|
322
|
+
metrics,
|
|
323
|
+
checkpointArtifacts,
|
|
324
|
+
evalReports,
|
|
325
|
+
bestReport,
|
|
326
|
+
stageManifestPath: stageManifest.path,
|
|
327
|
+
legacyArtifact: runner.lastArtifact || null,
|
|
328
|
+
lastCheckpoint: runner.lastCheckpoint || null,
|
|
329
|
+
};
|
|
330
|
+
} finally {
|
|
331
|
+
fixture?.cleanup?.();
|
|
332
|
+
await distillRuntime.cleanup();
|
|
333
|
+
}
|
|
334
|
+
}
|
|
335
|
+
|
|
336
|
+
export async function runDistillationStageA(options) {
|
|
337
|
+
return runDistillationStage(options);
|
|
338
|
+
}
|
|
@@ -0,0 +1,24 @@
|
|
|
1
|
+
import type { LoadedTrainingWorkload, DistillStagePlanEntry } from '../workloads.js';
|
|
2
|
+
|
|
3
|
+
export declare function runDistillationStageB(options: {
|
|
4
|
+
loadedWorkload: LoadedTrainingWorkload;
|
|
5
|
+
stageEntry: DistillStagePlanEntry;
|
|
6
|
+
layout: Record<string, string>;
|
|
7
|
+
datasetPath?: string | null;
|
|
8
|
+
stageAArtifact?: string | null;
|
|
9
|
+
stageAArtifactHash?: string | null;
|
|
10
|
+
legacyArtifactDir?: string | null;
|
|
11
|
+
timestamp?: string | Date | null;
|
|
12
|
+
parentArtifacts?: Array<Record<string, unknown>>;
|
|
13
|
+
priorStageResult?: Record<string, unknown> | null;
|
|
14
|
+
}): Promise<{
|
|
15
|
+
stageId: string;
|
|
16
|
+
trainingStage: 'stage_a' | 'stage_b';
|
|
17
|
+
metrics: Record<string, unknown>[];
|
|
18
|
+
checkpointArtifacts: Array<Record<string, unknown>>;
|
|
19
|
+
evalReports: Array<Record<string, unknown>>;
|
|
20
|
+
bestReport: Record<string, unknown> | null;
|
|
21
|
+
stageManifestPath: string;
|
|
22
|
+
legacyArtifact: Record<string, unknown> | null;
|
|
23
|
+
lastCheckpoint: Record<string, unknown> | null;
|
|
24
|
+
}>;
|
|
@@ -0,0 +1,20 @@
|
|
|
1
|
+
import { resolve } from 'node:path';
|
|
2
|
+
|
|
3
|
+
import { runDistillationStage } from './stage-a.js';
|
|
4
|
+
|
|
5
|
+
export async function runDistillationStageB(options) {
|
|
6
|
+
const priorStage = options.priorStageResult || null;
|
|
7
|
+
const stageAArtifact = options.stageAArtifact
|
|
8
|
+
|| priorStage?.legacyArtifact?.manifestPath
|
|
9
|
+
|| null;
|
|
10
|
+
if (!stageAArtifact) {
|
|
11
|
+
throw new Error('Distillation stage-b requires a Stage A artifact path.');
|
|
12
|
+
}
|
|
13
|
+
return runDistillationStage({
|
|
14
|
+
...options,
|
|
15
|
+
stageAArtifact: resolve(String(stageAArtifact)),
|
|
16
|
+
stageAArtifactHash: options.stageAArtifactHash
|
|
17
|
+
|| priorStage?.legacyArtifact?.manifestHash
|
|
18
|
+
|| null,
|
|
19
|
+
});
|
|
20
|
+
}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
export interface DistillStudentFixture {
|
|
2
|
+
config: Record<string, unknown>;
|
|
3
|
+
model: {
|
|
4
|
+
forward: (input: unknown, tape: unknown) => Promise<unknown>;
|
|
5
|
+
forwardDistill?: (
|
|
6
|
+
batch: unknown,
|
|
7
|
+
tape: unknown,
|
|
8
|
+
options?: Record<string, unknown>
|
|
9
|
+
) => Promise<{ logits: unknown }>;
|
|
10
|
+
cleanupDistillStep?: () => void;
|
|
11
|
+
loraParams?: () => unknown[];
|
|
12
|
+
paramGroups?: () => Record<string, unknown[]>;
|
|
13
|
+
};
|
|
14
|
+
outputDim?: number;
|
|
15
|
+
embeddingDim?: number;
|
|
16
|
+
cleanup(): void;
|
|
17
|
+
}
|
|
18
|
+
|
|
19
|
+
export declare function createDistillStudentRuntimeModelFixture(
|
|
20
|
+
overrides?: Record<string, unknown>,
|
|
21
|
+
options?: Record<string, unknown>
|
|
22
|
+
): Promise<DistillStudentFixture>;
|