mlx-bun 0.4.0 → 1.0.0
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 +1453 -88
- package/THIRD_PARTY_NOTICES.md +882 -0
- package/bin/mlx-bun.convert.mjs +4 -0
- package/bin/mlx-bun.fuse.mjs +4 -0
- package/bin/mlx-bun.generate.mjs +4 -0
- package/bin/mlx-bun.lora.mjs +4 -0
- package/bin/mlx-bun.mjs +13 -21
- package/bin/mlx-bun.server.mjs +4 -0
- package/bin/mlx-bun.upload.mjs +4 -0
- package/dist/web/app.js +9584 -0
- package/package.json +50 -53
- package/src/cli/ane-install.ts +76 -0
- package/src/cli/args.ts +179 -0
- package/src/cli/engine-entry.ts +68 -0
- package/src/cli/inference.ts +214 -0
- package/src/cli/job-entry.ts +97 -0
- package/src/cli/library-host.ts +117 -0
- package/src/cli/main.ts +57 -0
- package/src/cli/memory-engine.ts +163 -0
- package/src/cli/memory.ts +36 -0
- package/src/cli/mlx-lm-aliases.ts +309 -0
- package/src/cli/model-choice.ts +33 -0
- package/src/cli/model-selection.ts +84 -0
- package/src/cli/module-host.ts +11 -0
- package/src/cli/module-verbs.ts +50 -0
- package/src/cli/numerical-policy.ts +12 -0
- package/src/cli/perplexity.ts +89 -0
- package/src/cli/serve-host.ts +423 -0
- package/src/cli/serve-isolated.ts +217 -0
- package/src/cli/serve-options.ts +106 -0
- package/src/cli/serve-state.ts +205 -0
- package/src/cli/serve.ts +500 -0
- package/src/cli/served-model-host.ts +102 -0
- package/src/cli/server-entry.ts +148 -0
- package/src/cli/serving-unit.ts +177 -0
- package/src/{tui.ts → cli/terminal.ts} +13 -8
- package/src/cli/worker-adapters.ts +34 -0
- package/src/cli/worker-entry.ts +272 -0
- package/src/cli/worker-unit.ts +179 -0
- package/src/engine/adapter-operation.ts +26 -0
- package/src/engine/cache-services.ts +114 -0
- package/src/engine/completion.ts +39 -0
- package/src/engine/generation-gateway.ts +358 -0
- package/src/engine/index.ts +59 -0
- package/src/engine/kv-budget.ts +120 -0
- package/src/engine/media-preparation.ts +232 -0
- package/src/engine/model-binding.ts +24 -0
- package/src/engine/model-host.ts +513 -0
- package/src/engine/model-serving.ts +89 -0
- package/src/{serve → engine}/preparation.ts +10 -8
- package/src/{serve → engine}/session-completion-engine.ts +6 -6
- package/src/engine/telemetry.ts +109 -0
- package/src/hub/downloads.ts +116 -0
- package/src/jobs/db.ts +16 -7
- package/src/jobs/events.ts +29 -0
- package/src/jobs/executable.ts +5 -0
- package/src/jobs/host.ts +51 -0
- package/src/jobs/in-process.ts +55 -0
- package/src/{contracts/jobs.ts → jobs/protocol.ts} +7 -9
- package/src/jobs/runner.ts +150 -231
- package/src/jobs/service.ts +75 -0
- package/src/jobs/sse.ts +59 -23
- package/src/jobs/worker-memory.ts +46 -0
- package/src/jobs/worker-process.ts +163 -0
- package/src/jobs/worker-supervisor.ts +258 -0
- package/src/modules.ts +58 -0
- package/src/publishing/catalog-hub.ts +18 -0
- package/src/publishing/credentials.ts +31 -0
- package/src/publishing/upload.ts +37 -0
- package/src/residency/leased-adapters.ts +21 -0
- package/src/residency/local-records.ts +16 -0
- package/src/residency/model-residency.ts +405 -0
- package/src/residency/record-index.ts +28 -0
- package/src/residency/resident-estimate.ts +50 -0
- package/src/{anthropic.ts → server/anthropic.ts} +3 -81
- package/src/server/cache-routes.ts +36 -0
- package/src/{serve → server}/chat-request.ts +12 -10
- package/src/{serve → server}/chat-stage.ts +27 -22
- package/src/{client.ts → server/client.ts} +28 -2
- package/src/{serve → server}/completion-executor.ts +42 -42
- package/src/server/decision-body.ts +55 -0
- package/src/server/decision-routes.ts +134 -0
- package/src/server/decision-wire.ts +228 -0
- package/src/{serve → server}/discovery-routes.ts +42 -83
- package/src/server/generated-token-history.ts +57 -0
- package/src/{serve → server}/http.ts +25 -13
- package/src/{serve → server}/inference-request.ts +5 -5
- package/src/server/job-routes.ts +20 -0
- package/src/server/media-prompt.ts +63 -0
- package/src/server/memory-completion-client.ts +46 -0
- package/src/server/model-routes.ts +192 -0
- package/src/{serve → server}/openai-wire.ts +3 -2
- package/src/{serve → server}/pipeline.ts +1 -1
- package/src/server/prompt-builder.ts +13 -0
- package/src/server/prompt-contracts.ts +17 -0
- package/src/server/proxy-routes.ts +187 -0
- package/src/server/publishing-routes.ts +41 -0
- package/src/{serve → server}/request-plan.ts +8 -8
- package/src/{serve → server}/request-prep.ts +82 -68
- package/src/{serve → server}/responses-client.ts +30 -13
- package/src/{responses.ts → server/responses.ts} +12 -93
- package/src/server/routes.ts +205 -0
- package/src/server/start.ts +72 -0
- package/src/server/status-routes.ts +108 -0
- package/src/{serve → server}/text-completion-stage.ts +25 -4
- package/src/server/text-prompt.ts +12 -0
- package/src/{serve → server}/token-streams.ts +36 -38
- package/src/server/worker-routes.ts +311 -0
- package/src/storage/paths.ts +61 -0
- package/src/web/assets.ts +37 -0
- package/src/web/{src → browser}/api.ts +5 -23
- package/src/web/browser/chat-host.ts +17 -0
- package/src/web/browser/installed-panels.ts +6 -0
- package/src/web/browser/main.ts +55 -0
- package/src/web/browser/markdown.ts +279 -0
- package/src/web/browser/memory-panel.ts +9 -0
- package/src/web/browser/palette.ts +64 -0
- package/src/web/browser/protocol.ts +13 -0
- package/src/web/browser/shell.ts +467 -0
- package/src/web/{src → browser}/status.ts +32 -36
- package/src/web/build.ts +51 -0
- package/src/web/public/app.html +1054 -0
- package/src/web/{sw.js → public/sw.js} +4 -3
- package/src/web/{vendor → public/vendor}/hljs-theme.css +2 -2
- package/src/web/{vendor → public/vendor}/hljs.js +2 -2
- package/THIRD_PARTY_LICENSES.md +0 -51
- package/docs/README.md +0 -94
- package/docs/archive/hlg-sampling.md +0 -350
- package/docs/archive/investigations/audio-input-plan.md +0 -356
- package/docs/archive/investigations/chunk-finetune-experiment.md +0 -84
- package/docs/archive/investigations/colibri-glm52-port.md +0 -1486
- package/docs/archive/investigations/colibri-oracle-pin.md +0 -1284
- package/docs/archive/investigations/curve-bisector-routing.md +0 -123
- package/docs/archive/investigations/decode-roofline-lookagain.md +0 -348
- package/docs/archive/investigations/diffusion-gemma-port.md +0 -602
- package/docs/archive/investigations/dspark-handoff.md +0 -224
- package/docs/archive/investigations/expert-offload-single-user-moe.md +0 -292
- package/docs/archive/investigations/faithful-l1-consolidation.md +0 -310
- package/docs/archive/investigations/hlg-sampling-investigation.md +0 -970
- package/docs/archive/investigations/indexshare-performance-spike.md +0 -357
- package/docs/archive/investigations/kernel-perf-review-2026-07.md +0 -171
- package/docs/archive/investigations/lab-build-journal.md +0 -157
- package/docs/archive/investigations/minicpm5-decode-megakernel.md +0 -489
- package/docs/archive/investigations/openjarvis-fit.md +0 -198
- package/docs/archive/investigations/openwiki-evaluation.md +0 -249
- package/docs/archive/investigations/optimization_plan.md +0 -346
- package/docs/archive/investigations/orpo-base-uf-experiment-and-directions.md +0 -120
- package/docs/archive/investigations/orpo-flash-cce-pin-leak.md +0 -167
- package/docs/archive/investigations/orpo-uf-testing-handoff.md +0 -126
- package/docs/archive/investigations/pareto-specialized-runtime-findings.md +0 -92
- package/docs/archive/investigations/pi-builtin-investigation.md +0 -308
- package/docs/archive/investigations/pre-colibri-stabilization.md +0 -329
- package/docs/archive/investigations/qwen38-27b-campaign-log.md +0 -1092
- package/docs/archive/investigations/segmented-backward-handoff.md +0 -148
- package/docs/archive/investigations/starter-model-port-handoff.md +0 -342
- package/docs/archive/investigations/steel-flash-cce-handoff.md +0 -489
- package/docs/archive/investigations/the-dreaming-handoff.md +0 -141
- package/docs/archive/investigations/trainer-validation-experiment.md +0 -173
- package/docs/archive/planning/chat-ui-vision-memory-first.md +0 -199
- package/docs/archive/planning/chat-ui-vision.md +0 -50
- package/docs/archive/planning/chunk-task-roadmap.md +0 -77
- package/docs/archive/planning/curve-sampler-research-plan.md +0 -331
- package/docs/archive/planning/journal.md +0 -316
- package/docs/archive/planning/memory-docs-and-dag-plan.md +0 -194
- package/docs/archive/planning/pi-integration-review.md +0 -178
- package/docs/archive/planning/release-notes-v0.0.10.md +0 -57
- package/docs/archive/planning/release-notes-v0.0.11.md +0 -239
- package/docs/archive/planning/release-notes-v0.0.12.md +0 -80
- package/docs/archive/planning/release-notes-v0.0.13.md +0 -77
- package/docs/archive/planning/release-notes-v0.0.9.md +0 -122
- package/docs/archive/planning/release-notes-v0.2.0.md +0 -49
- package/docs/archive/planning/release-notes-v0.3.0.md +0 -35
- package/docs/archive/planning/web-ui-pass-plan.md +0 -213
- package/docs/archive/planning/website-readme-pass-plan.md +0 -224
- package/docs/design/batching.md +0 -1841
- package/docs/design/decode-speed-program.md +0 -2349
- package/docs/design/dreaming-nightly-pipeline.md +0 -684
- package/docs/design/generic-model-support.md +0 -705
- package/docs/design/kv-cache.md +0 -824
- package/docs/design/orpo-training.md +0 -680
- package/docs/design/speculative-decoding.md +0 -1990
- package/docs/design/turboquant.md +0 -2212
- package/docs/design/unified-engine-frontier-plan.md +0 -1880
- package/docs/design/web-chat-redesign.md +0 -695
- package/docs/planning/PRODUCT_ROADMAP.md +0 -594
- package/docs/planning/ResearchTopics.md +0 -119
- package/docs/planning/release-notes-v0.4.0.md +0 -45
- package/docs/planning/video-series.md +0 -421
- package/docs/reference/benchmarks.md +0 -3613
- package/docs/reference/cli.md +0 -585
- package/docs/reference/distribution.md +0 -355
- package/docs/reference/environment.md +0 -398
- package/docs/reference/glossary.md +0 -39
- package/docs/reference/library-api.md +0 -674
- package/docs/reference/memory.md +0 -260
- package/docs/reference/models.md +0 -334
- package/docs/reference/server-api.md +0 -1775
- package/docs/reference/server-config.md +0 -846
- package/docs/reference/training.md +0 -462
- package/docs/reference/troubleshooting.md +0 -158
- package/src/admission-plan.ts +0 -18
- package/src/audio/conformer.ts +0 -647
- package/src/audio/decode.ts +0 -133
- package/src/audio/features.ts +0 -202
- package/src/audio/transcode.ts +0 -67
- package/src/backends/mlx/assistant-target.ts +0 -46
- package/src/backends/mlx/autoregressive-draft-rows.ts +0 -120
- package/src/backends/mlx/autoregressive.ts +0 -108
- package/src/backends/mlx/batch-group.ts +0 -1790
- package/src/backends/mlx/cache-attachment.ts +0 -67
- package/src/backends/mlx/cache-layout.ts +0 -31
- package/src/backends/mlx/checkpoint-state.ts +0 -53
- package/src/backends/mlx/continuation-persistence.ts +0 -106
- package/src/backends/mlx/continuation-request.ts +0 -43
- package/src/backends/mlx/continuation.ts +0 -77
- package/src/backends/mlx/diffusion.ts +0 -39
- package/src/backends/mlx/draft-target.ts +0 -32
- package/src/backends/mlx/evaluation.ts +0 -105
- package/src/backends/mlx/fill-append.ts +0 -45
- package/src/backends/mlx/full-kv-row-append.ts +0 -32
- package/src/backends/mlx/full-kv-row-donor.ts +0 -54
- package/src/backends/mlx/full-transitioning-kv-rows.ts +0 -65
- package/src/backends/mlx/gateway-binding.ts +0 -130
- package/src/backends/mlx/graph.ts +0 -38
- package/src/backends/mlx/kv-codec.ts +0 -9
- package/src/backends/mlx/kv-maintenance.ts +0 -133
- package/src/backends/mlx/kv-tensor-rows.ts +0 -212
- package/src/backends/mlx/memory-client.ts +0 -189
- package/src/backends/mlx/methods.ts +0 -90
- package/src/backends/mlx/model-prompt.ts +0 -164
- package/src/backends/mlx/model-serving.ts +0 -79
- package/src/backends/mlx/prefill-cohort.ts +0 -89
- package/src/backends/mlx/prefill-rows.ts +0 -200
- package/src/backends/mlx/prefill.ts +0 -33
- package/src/backends/mlx/projected-context-rows.ts +0 -72
- package/src/backends/mlx/request-policy.ts +0 -20
- package/src/backends/mlx/request-state-policy.ts +0 -41
- package/src/backends/mlx/rollback.ts +0 -89
- package/src/backends/mlx/row-sampling.ts +0 -49
- package/src/backends/mlx/serial-executor.ts +0 -242
- package/src/backends/mlx/speculative-group.ts +0 -359
- package/src/backends/mlx/speculative-round.ts +0 -161
- package/src/backends/mlx/speculative.ts +0 -114
- package/src/backends/mlx/state-rows.ts +0 -104
- package/src/backends/mlx/state-views.ts +0 -46
- package/src/backends/mlx/target-layout-capability.ts +0 -35
- package/src/backends/mlx/text-engine.ts +0 -73
- package/src/backends/mlx/transitioning-kv-rows.ts +0 -165
- package/src/chat-template.ts +0 -382
- package/src/cli.ts +0 -3168
- package/src/config.ts +0 -399
- package/src/contracts/completion.ts +0 -16
- package/src/contracts/execution.ts +0 -60
- package/src/contracts/generation.ts +0 -103
- package/src/contracts/host.ts +0 -7
- package/src/contracts/pi-provider.ts +0 -30
- package/src/contracts/pi.ts +0 -259
- package/src/contracts/prefix-cache.ts +0 -23
- package/src/contracts/resources.ts +0 -25
- package/src/contracts/sampling.ts +0 -23
- package/src/contracts/scheduling.ts +0 -45
- package/src/contracts/tsconfig.json +0 -14
- package/src/dataset/generators.ts +0 -1087
- package/src/dataset/index.ts +0 -12
- package/src/dataset/job.ts +0 -81
- package/src/dataset/llm.ts +0 -71
- package/src/dataset/registry.ts +0 -528
- package/src/diffusion/diffusion-generate.ts +0 -507
- package/src/download.ts +0 -505
- package/src/embed.ts +0 -77
- package/src/engine/admission.ts +0 -59
- package/src/engine/cancellation.ts +0 -38
- package/src/engine/engine.ts +0 -56
- package/src/engine/execution-coordinator.ts +0 -70
- package/src/engine/execution-plan.ts +0 -53
- package/src/engine/resources.ts +0 -53
- package/src/engine/scheduler.ts +0 -72
- package/src/engine/session.ts +0 -247
- package/src/eval/README.md +0 -130
- package/src/eval/ifeval.ts +0 -81
- package/src/eval/kl-prompts.ts +0 -29
- package/src/eval/kl.ts +0 -590
- package/src/eval/perplexity.ts +0 -164
- package/src/eval/quality-db.ts +0 -92
- package/src/eval/reference-logits.ts +0 -207
- package/src/eval/runner.ts +0 -124
- package/src/eval/score.ts +0 -49
- package/src/eval/tasks/bfcl.ts +0 -535
- package/src/eval/tasks/gsm8k.ts +0 -98
- package/src/eval/tasks/hashhop.ts +0 -139
- package/src/eval/tasks/humaneval.ts +0 -275
- package/src/eval/tasks/ifeval.ts +0 -525
- package/src/eval/tasks/mmlu.ts +0 -157
- package/src/eval/ultrafeedback.ts +0 -60
- package/src/evaldb.ts +0 -199
- package/src/expert-io.ts +0 -393
- package/src/expert-offload-build.ts +0 -77
- package/src/expert-offload.ts +0 -76
- package/src/expert-residency.ts +0 -1134
- package/src/expert-usage.ts +0 -415
- package/src/fill/echo-index.ts +0 -188
- package/src/fill/fill-session.ts +0 -452
- package/src/fill/proposal.ts +0 -68
- package/src/fill/schema-rows.ts +0 -506
- package/src/fill/tool-boundary.ts +0 -103
- package/src/fit.ts +0 -249
- package/src/generate.ts +0 -1346
- package/src/grammar.ts +0 -591
- package/src/harness-pi.ts +0 -184
- package/src/hf-push.ts +0 -628
- package/src/hub-rest.ts +0 -242
- package/src/index.ts +0 -63
- package/src/inference/denoising.ts +0 -17
- package/src/inference/draft-acceptance.ts +0 -41
- package/src/inference/draft-output.ts +0 -49
- package/src/inference/graph.ts +0 -33
- package/src/inference/prefill.ts +0 -39
- package/src/inference/rollback.ts +0 -7
- package/src/jobs/index.ts +0 -31
- package/src/jobs/job-entry.ts +0 -107
- package/src/jobs/task-client.ts +0 -19
- package/src/jobs/types.ts +0 -2
- package/src/kv-scheme.ts +0 -257
- package/src/kv-store.ts +0 -1099
- package/src/lab/README.md +0 -18
- package/src/lab/curve/curve-designer.html +0 -303
- package/src/lab/curve/curve-sampler.ts +0 -122
- package/src/lab/expert-trace/expert-trace.ts +0 -69
- package/src/lab/paged-kv/paged-kv-rows.ts +0 -83
- package/src/lab/paged-kv/paged-kv.ts +0 -270
- package/src/library.ts +0 -40
- package/src/lora.ts +0 -408
- package/src/md-modules.d.ts +0 -49
- package/src/media-fetch.ts +0 -442
- package/src/memory/article.ts +0 -558
- package/src/memory/chunk-validation.ts +0 -150
- package/src/memory/chunk.ts +0 -318
- package/src/memory/cluster.ts +0 -351
- package/src/memory/coverage.ts +0 -134
- package/src/memory/crosslink.ts +0 -0
- package/src/memory/db.ts +0 -253
- package/src/memory/entity.ts +0 -203
- package/src/memory/gate.ts +0 -187
- package/src/memory/history.ts +0 -152
- package/src/memory/ingest.ts +0 -335
- package/src/memory/model.ts +0 -183
- package/src/memory/normalize.ts +0 -671
- package/src/memory/parse.ts +0 -22
- package/src/memory/pipeline.ts +0 -275
- package/src/memory/prompts.ts +0 -65
- package/src/memory/query.ts +0 -334
- package/src/memory/reconcile.ts +0 -734
- package/src/memory/redirect.ts +0 -108
- package/src/memory/reindex.ts +0 -325
- package/src/memory/resolve.ts +0 -307
- package/src/memory/rest.ts +0 -245
- package/src/memory/route.ts +0 -272
- package/src/memory/schedule.ts +0 -165
- package/src/memory/stages.ts +0 -777
- package/src/memory/synthesize.ts +0 -1638
- package/src/memory/tools.ts +0 -749
- package/src/memory/vault.ts +0 -1194
- package/src/memory/wikify.ts +0 -1064
- package/src/mlx/array.ts +0 -388
- package/src/mlx/autograd.ts +0 -304
- package/src/mlx/checkpoint.ts +0 -149
- package/src/mlx/compile.ts +0 -168
- package/src/mlx/custom-vjp.ts +0 -188
- package/src/mlx/ffi.ts +0 -512
- package/src/mlx/materialize.ts +0 -23
- package/src/mlx/metal-kernel.ts +0 -159
- package/src/mlx/ops.ts +0 -1021
- package/src/mlx/quantized-rows.ts +0 -36
- package/src/mlx/token-bitmask.ts +0 -52
- package/src/mlx/turboquant-kv-decode.ts +0 -82
- package/src/mlx/turboquant-kv-inverse.ts +0 -122
- package/src/mlx/turboquant-ops.ts +0 -861
- package/src/mlx/turboquant-tables.ts +0 -237
- package/src/mmap.ts +0 -90
- package/src/model/artifact-identity.ts +0 -73
- package/src/model/batched-kv.ts +0 -235
- package/src/model/batched-mask.ts +0 -279
- package/src/model/batched-quant.ts +0 -212
- package/src/model/batched-quantized-kv.ts +0 -144
- package/src/model/batched-rotating-quant.ts +0 -353
- package/src/model/batched-rotating-state.ts +0 -149
- package/src/model/batched-rotating.ts +0 -338
- package/src/model/batched-row-storage.ts +0 -179
- package/src/model/batched-ssm.ts +0 -52
- package/src/model/batched-turboquant-kv.ts +0 -93
- package/src/model/compiled-decode.ts +0 -712
- package/src/model/decoded-kv-donor.ts +0 -12
- package/src/model/delayed-quantized-kv.ts +0 -72
- package/src/model/delayed-rotating-quantized-kv.ts +0 -93
- package/src/model/delayed-turboquant-kv.ts +0 -55
- package/src/model/diffusion-gemma.ts +0 -1026
- package/src/model/factory.ts +0 -189
- package/src/model/fingerprint.ts +0 -33
- package/src/model/flash-attention.ts +0 -694
- package/src/model/full-prefill-padding.ts +0 -56
- package/src/model/full-prefill-row.ts +0 -105
- package/src/model/gemma4-base.ts +0 -2351
- package/src/model/gemma4.ts +0 -1315
- package/src/model/generated/gemma4-12b.ts +0 -429
- package/src/model/generated/gemma4-26b.ts +0 -448
- package/src/model/generated/gemma4-e4b.ts +0 -717
- package/src/model/generated/index.ts +0 -18
- package/src/model/glm52-atlas.ts +0 -511
- package/src/model/glm52-cache.ts +0 -792
- package/src/model/glm52-config.ts +0 -262
- package/src/model/glm52-container.ts +0 -305
- package/src/model/glm52-coupling.ts +0 -388
- package/src/model/glm52-dsa.ts +0 -606
- package/src/model/glm52-expert-layout.ts +0 -253
- package/src/model/glm52-layout.ts +0 -152
- package/src/model/glm52-memory.ts +0 -500
- package/src/model/glm52-mla.ts +0 -792
- package/src/model/glm52-moe.ts +0 -223
- package/src/model/glm52-pilot.ts +0 -537
- package/src/model/glm52-reference.ts +0 -445
- package/src/model/glm52-residency.ts +0 -557
- package/src/model/glm52-resident-weights.ts +0 -135
- package/src/model/glm52-streamed-experts.ts +0 -453
- package/src/model/glm52-streamed-metal.ts +0 -610
- package/src/model/glm52-weights.ts +0 -503
- package/src/model/glm52.ts +0 -1132
- package/src/model/implementation.ts +0 -69
- package/src/model/kv-attention-view.ts +0 -67
- package/src/model/minicpm5.ts +0 -376
- package/src/model/profile.ts +0 -443
- package/src/model/qwen-append.ts +0 -12
- package/src/model/qwen-conv.ts +0 -52
- package/src/model/qwen3-delta-state.ts +0 -74
- package/src/model/qwen3-delta.ts +0 -451
- package/src/model/qwen3-moe.ts +0 -404
- package/src/model/qwen3-mrope.ts +0 -252
- package/src/model/qwen3.ts +0 -277
- package/src/model/qwen3_5-checkpoint.ts +0 -164
- package/src/model/qwen3_5.ts +0 -819
- package/src/model/rotating-kv-donor.ts +0 -35
- package/src/model/rotating-kv-layout.ts +0 -262
- package/src/model/rotating-row-transaction.ts +0 -35
- package/src/model/speculative-rotating-kv.ts +0 -101
- package/src/model/ssm-prefill-padding.ts +0 -39
- package/src/model/support.ts +0 -102
- package/src/model/trellis-balanced-scatter.ts +0 -51
- package/src/model/trellis-linear.ts +0 -672
- package/src/model/trellis-shared-m.ts +0 -151
- package/src/model/trellis-shared-scatter.ts +0 -144
- package/src/model/trellis-splitk-prefill.ts +0 -158
- package/src/model/trellis-tiled-prefill.ts +0 -136
- package/src/model/trellis-vector-expand.ts +0 -62
- package/src/model/trellis-wide-prefill.ts +0 -94
- package/src/model/turboquant-codec.ts +0 -83
- package/src/model/universal/archs.ts +0 -332
- package/src/model/universal/dense.ts +0 -612
- package/src/model/universal/modules.ts +0 -170
- package/src/model/universal/rope.ts +0 -381
- package/src/native/expert_io.c +0 -523
- package/src/native/frame_extract.swift +0 -140
- package/src/native-pack.ts +0 -162
- package/src/pi-provider.ts +0 -53
- package/src/pi-session.ts +0 -81
- package/src/pi-terminal.ts +0 -274
- package/src/pi-web.ts +0 -1932
- package/src/preflight.ts +0 -158
- package/src/prompt-cache.ts +0 -460
- package/src/quantize/allocator.ts +0 -627
- package/src/quantize/atomic-output.ts +0 -38
- package/src/quantize/calibration.ts +0 -181
- package/src/quantize/config-writer.ts +0 -145
- package/src/quantize/index.ts +0 -84
- package/src/quantize/job.ts +0 -162
- package/src/quantize/quantizer.ts +0 -666
- package/src/quantize/rotate.ts +0 -603
- package/src/quantize/safetensors-writer.ts +0 -249
- package/src/quantize/sensitivity.ts +0 -216
- package/src/quantize/trellis.ts +0 -586
- package/src/quantize/weight-transform.ts +0 -292
- package/src/registry.ts +0 -602
- package/src/runtime-config.ts +0 -81
- package/src/safetensors.ts +0 -158
- package/src/sampler.ts +0 -1011
- package/src/serve/admin-routes.ts +0 -184
- package/src/serve/assets/curve-terrain.html +0 -105
- package/src/serve/assets/training-inference-map.html +0 -575
- package/src/serve/aux-routes.ts +0 -132
- package/src/serve/batch-scheduler.ts +0 -8
- package/src/serve/checkpoint-identity.ts +0 -64
- package/src/serve/generation-gateway.ts +0 -496
- package/src/serve/isolate.ts +0 -609
- package/src/serve/kv-budget.ts +0 -14
- package/src/serve/lab-routes.ts +0 -341
- package/src/serve/lane-registry.ts +0 -45
- package/src/serve/model-admin-routes.ts +0 -155
- package/src/serve/model-binding.ts +0 -43
- package/src/serve/model-host.ts +0 -511
- package/src/serve/parent-application.ts +0 -56
- package/src/serve/prompt-response-trace.ts +0 -147
- package/src/serve/response-lease.ts +0 -24
- package/src/serve/session-search.ts +0 -326
- package/src/serve/static-routes.ts +0 -101
- package/src/server.ts +0 -1314
- package/src/spec/assistant-rows.ts +0 -92
- package/src/spec/assistant-source.ts +0 -126
- package/src/spec/cache-identity.ts +0 -22
- package/src/spec/deepspec-rows.ts +0 -29
- package/src/spec/deepspec-source.ts +0 -145
- package/src/spec/dflash-source.ts +0 -149
- package/src/spec/drafter.ts +0 -304
- package/src/spec/dspark/ab-stats.ts +0 -136
- package/src/spec/dspark/calibration.ts +0 -112
- package/src/spec/dspark/data-dflash.ts +0 -156
- package/src/spec/dspark/data.ts +0 -201
- package/src/spec/dspark/deepspec-module.ts +0 -776
- package/src/spec/dspark/generate-dflash.ts +0 -245
- package/src/spec/dspark/generate.ts +0 -290
- package/src/spec/dspark/loader.ts +0 -23
- package/src/spec/dspark/loss.ts +0 -213
- package/src/spec/dspark/module-dflash.ts +0 -560
- package/src/spec/dspark/module.ts +0 -540
- package/src/spec/dspark/quantize-drafter.ts +0 -91
- package/src/spec/dspark/sample.ts +0 -112
- package/src/spec/generate.ts +0 -276
- package/src/spec/glm52-mtp-graph.ts +0 -29
- package/src/spec/glm52-mtp-rows.ts +0 -132
- package/src/spec/glm52-mtp-source.ts +0 -317
- package/src/spec/glm52-mtp-state.ts +0 -33
- package/src/spec/ngram-source.ts +0 -204
- package/src/spec/projected-draft-rows.ts +0 -73
- package/src/spec/qwen-mtp-module.ts +0 -171
- package/src/spec/qwen-mtp-rows.ts +0 -219
- package/src/spec/qwen-mtp-source.ts +0 -383
- package/src/spec/qwen-mtp-state.ts +0 -30
- package/src/spec/serve-loop.ts +0 -585
- package/src/spec/source.ts +0 -215
- package/src/spec/two-model-rows.ts +0 -61
- package/src/spec/two-model.ts +0 -113
- package/src/ssd-cache.ts +0 -372
- package/src/ssd-durability.ts +0 -211
- package/src/tokenizer.ts +0 -335
- package/src/tool-approvals.ts +0 -131
- package/src/tool-call.ts +0 -695
- package/src/train/dataset.ts +0 -484
- package/src/train/diffusion-lora.ts +0 -216
- package/src/train/export.ts +0 -32
- package/src/train/flash-cce.ts +0 -1234
- package/src/train/forward.ts +0 -266
- package/src/train/fuse.ts +0 -243
- package/src/train/index.ts +0 -46
- package/src/train/job.ts +0 -205
- package/src/train/lora-params.ts +0 -254
- package/src/train/loss.ts +0 -1591
- package/src/train/merge.ts +0 -195
- package/src/train/optimizer.ts +0 -183
- package/src/train/prefix-shared.ts +0 -490
- package/src/train/rank.ts +0 -137
- package/src/train/segmented.ts +0 -1828
- package/src/train/steel-qmm-header.ts +0 -1461
- package/src/train/trainer.ts +0 -1221
- package/src/train/watch.ts +0 -542
- package/src/vision/diffusion-vision.ts +0 -620
- package/src/vision/embedder.ts +0 -168
- package/src/vision/preprocess.ts +0 -247
- package/src/vision/prompt.ts +0 -373
- package/src/vision/qwen3vl-preprocess.ts +0 -409
- package/src/vision/qwen3vl-prompt.ts +0 -196
- package/src/vision/qwen3vl-tower.ts +0 -555
- package/src/vision/siglip.ts +0 -599
- package/src/vision/video-frames.ts +0 -148
- package/src/web/app.html +0 -2143
- package/src/web/app.js +0 -5703
- package/src/web/skills/memory/SKILL.md +0 -120
- package/src/web/skills/web-research/SKILL.md +0 -46
- package/src/web/skills.ts +0 -97
- package/src/web/src/adapters-panel.ts +0 -288
- package/src/web/src/assistant.ts +0 -451
- package/src/web/src/chat.ts +0 -1057
- package/src/web/src/composer.ts +0 -959
- package/src/web/src/dataset.ts +0 -132
- package/src/web/src/finetune.ts +0 -267
- package/src/web/src/hub.ts +0 -428
- package/src/web/src/main.ts +0 -98
- package/src/web/src/markdown.ts +0 -0
- package/src/web/src/memory-panel.ts +0 -572
- package/src/web/src/model-picker.ts +0 -159
- package/src/web/src/palette.ts +0 -346
- package/src/web/src/protocol.ts +0 -26
- package/src/web/src/quantize.ts +0 -167
- package/src/web/src/rag.ts +0 -274
- package/src/web/src/sessions.ts +0 -469
- package/src/web/src/shell.ts +0 -804
- package/src/web/src/ui-catalog.ts +0 -78
- package/src/web/vendor/README.md +0 -61
- package/src/web-assets.ts +0 -47
- package/src/web-tools.ts +0 -478
- package/src/weights.ts +0 -255
- /package/src/{serve → server}/completion-sink.ts +0 -0
- /package/src/web/{icon.svg → public/icon.svg} +0 -0
- /package/src/web/{manifest.webmanifest → public/manifest.webmanifest} +0 -0
- /package/src/web/{vendor → public/vendor}/hljs-LICENSE +0 -0
package/src/train/watch.ts
DELETED
|
@@ -1,542 +0,0 @@
|
|
|
1
|
-
// `mlx-bun train-watch <adapter-dir>` — a mactop-style live dashboard for a
|
|
2
|
-
// training run. It tails the trainer's append-only `<adapter>/metrics.jsonl`
|
|
3
|
-
// (written by trainLora) and renders auto-scaled line charts for loss, margin,
|
|
4
|
-
// accuracy (train line + val markers) and memory, plus a progress/ETA header.
|
|
5
|
-
//
|
|
6
|
-
// Decoupled by design: the trainer only writes the stream, this viewer only
|
|
7
|
-
// reads it. So it attaches/detaches freely to a LIVE run, replays a FINISHED
|
|
8
|
-
// one, and never touches training. The renderer is a PURE function
|
|
9
|
-
// (renderFrame) over parsed state, so it can be exercised against a captured
|
|
10
|
-
// fixture with no TTY.
|
|
11
|
-
|
|
12
|
-
import { readFileSync } from "node:fs";
|
|
13
|
-
|
|
14
|
-
// ---------------------------------------------------------------------------
|
|
15
|
-
// Parsed run state
|
|
16
|
-
// ---------------------------------------------------------------------------
|
|
17
|
-
|
|
18
|
-
export interface Pt { step: number; t: number; y: number }
|
|
19
|
-
/** A validation point — sparse (one per eval step). Accuracy carries its raw
|
|
20
|
-
* correct/total so it can be shown as a count, not a bare ratio. */
|
|
21
|
-
export interface ValPt extends Pt { nc?: number; nt?: number }
|
|
22
|
-
export interface RunState {
|
|
23
|
-
meta: {
|
|
24
|
-
method: string; model: string; iters: number; lambda?: number;
|
|
25
|
-
lr?: number; seq?: number; startedAt?: number; adapterPath?: string;
|
|
26
|
-
};
|
|
27
|
-
// Dense per-iter series (a line per metric).
|
|
28
|
-
loss: Pt[];
|
|
29
|
-
margin: Pt[];
|
|
30
|
-
trainAcc: Pt[]; // y ∈ {0,1} at B=1 — render as a rolling fraction, not raw
|
|
31
|
-
mem: Pt[]; // y = active GB (resident — flat = no leak)
|
|
32
|
-
peak: Pt[]; // y = peak GB (cumulative high-water — the climb-to-plateau)
|
|
33
|
-
// Sparse per-eval series (markers, not lines — they only exist at eval steps).
|
|
34
|
-
valAcc: ValPt[];
|
|
35
|
-
valLoss: Pt[];
|
|
36
|
-
peakGb: number; // running high-water
|
|
37
|
-
lastStep: number;
|
|
38
|
-
lastT: number;
|
|
39
|
-
done: boolean;
|
|
40
|
-
}
|
|
41
|
-
|
|
42
|
-
function emptyState(): RunState {
|
|
43
|
-
return {
|
|
44
|
-
meta: { method: "?", model: "", iters: 0 },
|
|
45
|
-
loss: [], margin: [], trainAcc: [], mem: [], peak: [],
|
|
46
|
-
valAcc: [], valLoss: [],
|
|
47
|
-
peakGb: 0, lastStep: 0, lastT: 0, done: false,
|
|
48
|
-
};
|
|
49
|
-
}
|
|
50
|
-
|
|
51
|
-
/** Fold one parsed JSONL record into the state. */
|
|
52
|
-
function ingest(s: RunState, o: Record<string, unknown>): void {
|
|
53
|
-
const type = o.type;
|
|
54
|
-
if (type === "meta") {
|
|
55
|
-
s.meta = {
|
|
56
|
-
method: String(o.method ?? "?"),
|
|
57
|
-
model: String(o.model ?? ""),
|
|
58
|
-
iters: Number(o.iters ?? 0),
|
|
59
|
-
lambda: o.orpo_lambda != null ? Number(o.orpo_lambda) : undefined,
|
|
60
|
-
lr: o.learning_rate != null ? Number(o.learning_rate) : undefined,
|
|
61
|
-
seq: o.max_seq_length != null ? Number(o.max_seq_length) : undefined,
|
|
62
|
-
startedAt: o.t != null ? Number(o.t) : undefined,
|
|
63
|
-
adapterPath: o.adapter_path != null ? String(o.adapter_path) : undefined,
|
|
64
|
-
};
|
|
65
|
-
return;
|
|
66
|
-
}
|
|
67
|
-
if (type !== "metric") return;
|
|
68
|
-
const step = Number(o.step), t = Number(o.t);
|
|
69
|
-
if (!Number.isFinite(step)) return;
|
|
70
|
-
if (o.kind === "train") {
|
|
71
|
-
if (Number.isFinite(Number(o.loss))) s.loss.push({ step, t, y: Number(o.loss) });
|
|
72
|
-
if (Number.isFinite(Number(o.margin))) s.margin.push({ step, t, y: Number(o.margin) });
|
|
73
|
-
if (Number.isFinite(Number(o.accuracy))) s.trainAcc.push({ step, t, y: Number(o.accuracy) });
|
|
74
|
-
if (Number.isFinite(Number(o.active_gb))) s.mem.push({ step, t, y: Number(o.active_gb) });
|
|
75
|
-
if (Number.isFinite(Number(o.peak_gb))) {
|
|
76
|
-
s.peak.push({ step, t, y: Number(o.peak_gb) });
|
|
77
|
-
s.peakGb = Math.max(s.peakGb, Number(o.peak_gb));
|
|
78
|
-
}
|
|
79
|
-
s.lastStep = Math.max(s.lastStep, step);
|
|
80
|
-
if (Number.isFinite(t)) s.lastT = Math.max(s.lastT, t);
|
|
81
|
-
} else if (o.kind === "val") {
|
|
82
|
-
if (Number.isFinite(Number(o.accuracy)))
|
|
83
|
-
s.valAcc.push({ step, t, y: Number(o.accuracy),
|
|
84
|
-
nc: o.n_correct != null ? Number(o.n_correct) : undefined,
|
|
85
|
-
nt: o.n_total != null ? Number(o.n_total) : undefined });
|
|
86
|
-
if (Number.isFinite(Number(o.loss))) s.valLoss.push({ step, t, y: Number(o.loss) });
|
|
87
|
-
}
|
|
88
|
-
}
|
|
89
|
-
|
|
90
|
-
/** Build full state from a metrics.jsonl text blob (replay-from-start). */
|
|
91
|
-
export function parseStream(text: string): RunState {
|
|
92
|
-
const s = emptyState();
|
|
93
|
-
for (const line of text.split("\n")) {
|
|
94
|
-
const t = line.trim();
|
|
95
|
-
if (!t) continue;
|
|
96
|
-
try { ingest(s, JSON.parse(t) as Record<string, unknown>); } catch { /* skip partial/garbled */ }
|
|
97
|
-
}
|
|
98
|
-
if (s.meta.iters > 0 && s.lastStep >= s.meta.iters) s.done = true;
|
|
99
|
-
return s;
|
|
100
|
-
}
|
|
101
|
-
|
|
102
|
-
/** Median step time (seconds) from the last K train timestamps. */
|
|
103
|
-
export function sPerStep(s: RunState, k = 30): number {
|
|
104
|
-
const pts = s.loss.slice(-k - 1);
|
|
105
|
-
const d: number[] = [];
|
|
106
|
-
for (let i = 1; i < pts.length; i++) {
|
|
107
|
-
const dt = (pts[i]!.t - pts[i - 1]!.t) / 1000;
|
|
108
|
-
if (dt > 0 && dt < 600) d.push(dt);
|
|
109
|
-
}
|
|
110
|
-
if (d.length === 0) return 0;
|
|
111
|
-
d.sort((a, b) => a - b);
|
|
112
|
-
return d[Math.floor(d.length / 2)]!;
|
|
113
|
-
}
|
|
114
|
-
|
|
115
|
-
// ---------------------------------------------------------------------------
|
|
116
|
-
// Cell grid compositor — the whole frame is built into a typed grid, then
|
|
117
|
-
// run-length colored on render. Unifies header / charts / axis labels / markers
|
|
118
|
-
// and sidesteps any mid-ANSI string splicing.
|
|
119
|
-
// ---------------------------------------------------------------------------
|
|
120
|
-
|
|
121
|
-
const DEFAULT_FG = -1;
|
|
122
|
-
interface Cell { ch: string; fg: number }
|
|
123
|
-
|
|
124
|
-
// 256-color palette (xterm).
|
|
125
|
-
const C = {
|
|
126
|
-
dim: 240, axis: 244, label: 250, title: 255,
|
|
127
|
-
loss: 203, margin: 78, accT: 45, accV: 207, mem: 221, ok: 78, warn: 203, accent: 81,
|
|
128
|
-
};
|
|
129
|
-
|
|
130
|
-
class Grid {
|
|
131
|
-
rows: number; cols: number; cells: Cell[];
|
|
132
|
-
constructor(rows: number, cols: number) {
|
|
133
|
-
this.rows = rows; this.cols = cols;
|
|
134
|
-
this.cells = new Array(rows * cols);
|
|
135
|
-
for (let i = 0; i < this.cells.length; i++) this.cells[i] = { ch: " ", fg: DEFAULT_FG };
|
|
136
|
-
}
|
|
137
|
-
put(r: number, c: number, ch: string, fg = DEFAULT_FG): void {
|
|
138
|
-
if (r < 0 || r >= this.rows || c < 0 || c >= this.cols) return;
|
|
139
|
-
const cell = this.cells[r * this.cols + c]!;
|
|
140
|
-
cell.ch = ch; cell.fg = fg;
|
|
141
|
-
}
|
|
142
|
-
text(r: number, c: number, str: string, fg = DEFAULT_FG): void {
|
|
143
|
-
for (let i = 0; i < str.length; i++) this.put(r, c + i, str[i]!, fg);
|
|
144
|
-
}
|
|
145
|
-
/** Right-align str ending at column `cEnd`. */
|
|
146
|
-
textRight(r: number, cEnd: number, str: string, fg = DEFAULT_FG): void {
|
|
147
|
-
this.text(r, cEnd - str.length + 1, str, fg);
|
|
148
|
-
}
|
|
149
|
-
render(): string[] {
|
|
150
|
-
const out: string[] = [];
|
|
151
|
-
for (let r = 0; r < this.rows; r++) {
|
|
152
|
-
let line = "", curFg = DEFAULT_FG;
|
|
153
|
-
for (let c = 0; c < this.cols; c++) {
|
|
154
|
-
const cell = this.cells[r * this.cols + c]!;
|
|
155
|
-
if (cell.fg !== curFg) {
|
|
156
|
-
line += cell.fg === DEFAULT_FG ? "\x1b[0m" : `\x1b[38;5;${cell.fg}m`;
|
|
157
|
-
curFg = cell.fg;
|
|
158
|
-
}
|
|
159
|
-
line += cell.ch;
|
|
160
|
-
}
|
|
161
|
-
if (curFg !== DEFAULT_FG) line += "\x1b[0m";
|
|
162
|
-
out.push(line);
|
|
163
|
-
}
|
|
164
|
-
return out;
|
|
165
|
-
}
|
|
166
|
-
}
|
|
167
|
-
|
|
168
|
-
// ---------------------------------------------------------------------------
|
|
169
|
-
// Braille (2×4 dot) canvas — sub-cell line resolution for the charts.
|
|
170
|
-
// ---------------------------------------------------------------------------
|
|
171
|
-
|
|
172
|
-
// dot bit for (col∈{0,1}, row∈{0..3}) within a cell.
|
|
173
|
-
const DOT = [
|
|
174
|
-
[0x01, 0x08],
|
|
175
|
-
[0x02, 0x10],
|
|
176
|
-
[0x04, 0x20],
|
|
177
|
-
[0x40, 0x80],
|
|
178
|
-
];
|
|
179
|
-
|
|
180
|
-
class Braille {
|
|
181
|
-
cols: number; rows: number; data: Uint8Array;
|
|
182
|
-
constructor(cols: number, rows: number) {
|
|
183
|
-
this.cols = cols; this.rows = rows;
|
|
184
|
-
this.data = new Uint8Array(cols * rows);
|
|
185
|
-
}
|
|
186
|
-
setDot(dx: number, dy: number): void {
|
|
187
|
-
const cx = dx >> 1, cy = dy >> 2;
|
|
188
|
-
if (cx < 0 || cx >= this.cols || cy < 0 || cy >= this.rows) return;
|
|
189
|
-
this.data[cy * this.cols + cx]! |= DOT[dy & 3]![dx & 1]!;
|
|
190
|
-
}
|
|
191
|
-
line(x0: number, y0: number, x1: number, y1: number): void {
|
|
192
|
-
let dx = Math.abs(x1 - x0), dy = -Math.abs(y1 - y0);
|
|
193
|
-
const sx = x0 < x1 ? 1 : -1, sy = y0 < y1 ? 1 : -1;
|
|
194
|
-
let err = dx + dy;
|
|
195
|
-
for (;;) {
|
|
196
|
-
this.setDot(x0, y0);
|
|
197
|
-
if (x0 === x1 && y0 === y1) break;
|
|
198
|
-
const e2 = 2 * err;
|
|
199
|
-
if (e2 >= dy) { err += dy; x0 += sx; }
|
|
200
|
-
if (e2 <= dx) { err += dx; y0 += sy; }
|
|
201
|
-
}
|
|
202
|
-
}
|
|
203
|
-
char(cx: number, cy: number): string | null {
|
|
204
|
-
const b = this.data[cy * this.cols + cx]!;
|
|
205
|
-
return b === 0 ? null : String.fromCodePoint(0x2800 | b);
|
|
206
|
-
}
|
|
207
|
-
}
|
|
208
|
-
|
|
209
|
-
// ---------------------------------------------------------------------------
|
|
210
|
-
// One chart panel: title row + y-axis labels + braille plot + x-axis labels.
|
|
211
|
-
// ---------------------------------------------------------------------------
|
|
212
|
-
|
|
213
|
-
interface ChartOpts {
|
|
214
|
-
title: string;
|
|
215
|
-
series: Pt[];
|
|
216
|
-
color: number;
|
|
217
|
-
fixed?: [number, number]; // fixed y-range (e.g. accuracy 0..1)
|
|
218
|
-
fmtY: (v: number) => string;
|
|
219
|
-
headline?: string; // colored value shown after the title
|
|
220
|
-
alert?: string; // warn-colored note after the headline (e.g. "▲ rising")
|
|
221
|
-
zeroLine?: boolean; // dashed y=0 reference
|
|
222
|
-
refLine?: number; // dashed reference rule at an arbitrary y (e.g. 0.5 chance)
|
|
223
|
-
markers?: { pts: Pt[]; color: number }; // discrete sparse overlay (val points)
|
|
224
|
-
}
|
|
225
|
-
|
|
226
|
-
function fmtNum(v: number): string {
|
|
227
|
-
if (!Number.isFinite(v)) return "—";
|
|
228
|
-
const a = Math.abs(v);
|
|
229
|
-
if (a >= 100) return v.toFixed(0);
|
|
230
|
-
if (a >= 10) return v.toFixed(1);
|
|
231
|
-
if (a >= 1) return v.toFixed(2);
|
|
232
|
-
return v.toFixed(3);
|
|
233
|
-
}
|
|
234
|
-
|
|
235
|
-
function compactInt(n: number): string {
|
|
236
|
-
if (n >= 1_000_000) return (n / 1_000_000).toFixed(n % 1_000_000 ? 1 : 0) + "M";
|
|
237
|
-
if (n >= 10_000) return (n / 1000).toFixed(0) + "k";
|
|
238
|
-
return String(n);
|
|
239
|
-
}
|
|
240
|
-
|
|
241
|
-
/** Draw a chart into `grid` at top-left (r0,c0) spanning (h rows, w cols). */
|
|
242
|
-
function drawChart(grid: Grid, r0: number, c0: number, h: number, w: number, o: ChartOpts): void {
|
|
243
|
-
// Title line: title · headline (series color) · alert (warn).
|
|
244
|
-
grid.text(r0, c0, o.title, C.title);
|
|
245
|
-
let hx = c0 + o.title.length + 1;
|
|
246
|
-
if (o.headline) { grid.text(r0, hx, o.headline, o.color); hx += o.headline.length + 2; }
|
|
247
|
-
if (o.alert) grid.text(r0, hx, o.alert, C.warn);
|
|
248
|
-
|
|
249
|
-
const plotTop = r0 + 1;
|
|
250
|
-
const plotRows = h - 2; // minus title + x-axis row
|
|
251
|
-
if (plotRows < 1 || w < 12) return;
|
|
252
|
-
|
|
253
|
-
const pts = o.series;
|
|
254
|
-
// y-range.
|
|
255
|
-
let ymin: number, ymax: number;
|
|
256
|
-
if (o.fixed) { [ymin, ymax] = o.fixed; }
|
|
257
|
-
else if (pts.length === 0) { ymin = 0; ymax = 1; }
|
|
258
|
-
else {
|
|
259
|
-
ymin = Infinity; ymax = -Infinity;
|
|
260
|
-
for (const p of pts) { if (p.y < ymin) ymin = p.y; if (p.y > ymax) ymax = p.y; }
|
|
261
|
-
if (o.markers) for (const p of o.markers.pts) { if (p.y < ymin) ymin = p.y; if (p.y > ymax) ymax = p.y; }
|
|
262
|
-
if (o.zeroLine) { ymin = Math.min(ymin, 0); ymax = Math.max(ymax, 0); }
|
|
263
|
-
if (ymin === ymax) { ymin -= 1; ymax += 1; }
|
|
264
|
-
else { const pad = (ymax - ymin) * 0.08; ymin -= pad; ymax += pad; }
|
|
265
|
-
}
|
|
266
|
-
|
|
267
|
-
// gutter for y labels.
|
|
268
|
-
const gutter = Math.max(o.fmtY(ymin).length, o.fmtY(ymax).length) + 1;
|
|
269
|
-
const plotCols = w - gutter;
|
|
270
|
-
if (plotCols < 4) return;
|
|
271
|
-
|
|
272
|
-
// y-axis labels: top=max, bottom=min, mid where room.
|
|
273
|
-
grid.textRight(plotTop, c0 + gutter - 2, o.fmtY(ymax), C.axis);
|
|
274
|
-
grid.textRight(plotTop + plotRows - 1, c0 + gutter - 2, o.fmtY(ymin), C.axis);
|
|
275
|
-
if (plotRows >= 5) grid.textRight(plotTop + (plotRows >> 1), c0 + gutter - 2, o.fmtY((ymin + ymax) / 2), C.axis);
|
|
276
|
-
// vertical axis rule.
|
|
277
|
-
for (let r = 0; r < plotRows; r++) grid.put(plotTop + r, c0 + gutter - 1, "│", C.dim);
|
|
278
|
-
|
|
279
|
-
// x-range (step).
|
|
280
|
-
const xmin = pts.length ? pts[0]!.step : 0;
|
|
281
|
-
const xmax = pts.length ? pts[pts.length - 1]!.step : 1;
|
|
282
|
-
const xspan = Math.max(1, xmax - xmin);
|
|
283
|
-
|
|
284
|
-
const nDotX = plotCols * 2, nDotY = plotRows * 4;
|
|
285
|
-
const canvas = new Braille(plotCols, plotRows);
|
|
286
|
-
const mapX = (step: number) => Math.round(((step - xmin) / xspan) * (nDotX - 1));
|
|
287
|
-
const mapY = (y: number) => Math.round((1 - (y - ymin) / (ymax - ymin)) * (nDotY - 1));
|
|
288
|
-
|
|
289
|
-
// zero reference: a dim dotted grid rule the data line crosses (drawn first,
|
|
290
|
-
// so the braille blit below overwrites it only where there's data).
|
|
291
|
-
const drawRule = (yval: number) => {
|
|
292
|
-
if (!(yval > ymin && yval < ymax)) return;
|
|
293
|
-
const rRow = plotTop + Math.round((1 - (yval - ymin) / (ymax - ymin)) * (plotRows - 1));
|
|
294
|
-
for (let c = 0; c < plotCols; c++) grid.put(rRow, c0 + gutter + c, "┄", C.dim);
|
|
295
|
-
};
|
|
296
|
-
if (o.zeroLine) drawRule(0);
|
|
297
|
-
if (o.refLine != null) drawRule(o.refLine);
|
|
298
|
-
|
|
299
|
-
// Downsample to keep line cost bounded on long runs.
|
|
300
|
-
if (pts.length > 0) {
|
|
301
|
-
const stride = Math.max(1, Math.floor(pts.length / (nDotX * 2)));
|
|
302
|
-
let prevX = mapX(pts[0]!.step), prevY = mapY(pts[0]!.y);
|
|
303
|
-
canvas.setDot(prevX, prevY);
|
|
304
|
-
for (let i = stride; i < pts.length; i += stride) {
|
|
305
|
-
const cx = mapX(pts[i]!.step), cy = mapY(pts[i]!.y);
|
|
306
|
-
canvas.line(prevX, prevY, cx, cy);
|
|
307
|
-
prevX = cx; prevY = cy;
|
|
308
|
-
}
|
|
309
|
-
// always connect the final point.
|
|
310
|
-
const last = pts[pts.length - 1]!;
|
|
311
|
-
canvas.line(prevX, prevY, mapX(last.step), mapY(last.y));
|
|
312
|
-
}
|
|
313
|
-
|
|
314
|
-
// Blit braille into grid.
|
|
315
|
-
for (let r = 0; r < plotRows; r++)
|
|
316
|
-
for (let c = 0; c < plotCols; c++) {
|
|
317
|
-
const ch = canvas.char(c, r);
|
|
318
|
-
if (ch) grid.put(plotTop + r, c0 + gutter + c, ch, o.color);
|
|
319
|
-
}
|
|
320
|
-
|
|
321
|
-
// Discrete markers (val accuracy) overlaid as ●.
|
|
322
|
-
if (o.markers) {
|
|
323
|
-
for (const p of o.markers.pts) {
|
|
324
|
-
const cx = c0 + gutter + Math.round(((p.step - xmin) / xspan) * (plotCols - 1));
|
|
325
|
-
const cy = plotTop + Math.round((1 - (p.y - ymin) / (ymax - ymin)) * (plotRows - 1));
|
|
326
|
-
grid.put(cy, cx, "●", o.markers.color);
|
|
327
|
-
}
|
|
328
|
-
}
|
|
329
|
-
|
|
330
|
-
// x-axis labels row.
|
|
331
|
-
const axisRow = plotTop + plotRows;
|
|
332
|
-
grid.text(axisRow, c0 + gutter, compactInt(xmin), C.axis);
|
|
333
|
-
grid.textRight(axisRow, c0 + gutter + plotCols - 1, compactInt(xmax), C.axis);
|
|
334
|
-
}
|
|
335
|
-
|
|
336
|
-
// ---------------------------------------------------------------------------
|
|
337
|
-
// Full-frame renderer (pure).
|
|
338
|
-
// ---------------------------------------------------------------------------
|
|
339
|
-
|
|
340
|
-
/** Rolling MEDIAN over a trailing window — robust trend for a metric whose
|
|
341
|
-
* per-step value is per-example (B=1 loss spikes on a long/hard pair are
|
|
342
|
-
* rejected, unlike a mean). This is what makes the loss trend legible. */
|
|
343
|
-
function rollingMedian(pts: Pt[], w: number): Pt[] {
|
|
344
|
-
const out: Pt[] = [];
|
|
345
|
-
for (let i = 0; i < pts.length; i++) {
|
|
346
|
-
const lo = Math.max(0, i - w + 1);
|
|
347
|
-
const win: number[] = [];
|
|
348
|
-
for (let j = lo; j <= i; j++) win.push(pts[j]!.y);
|
|
349
|
-
win.sort((a, b) => a - b);
|
|
350
|
-
out.push({ step: pts[i]!.step, t: pts[i]!.t, y: win[win.length >> 1]! });
|
|
351
|
-
}
|
|
352
|
-
return out;
|
|
353
|
-
}
|
|
354
|
-
|
|
355
|
-
/** Light trailing mean — smooths the piecewise-constant plateaus a rolling
|
|
356
|
-
* median leaves behind (median despikes; this mean removes the "banding" those
|
|
357
|
-
* flat runs make in the braille line). */
|
|
358
|
-
function rollingMean(pts: Pt[], w: number): Pt[] {
|
|
359
|
-
const out: Pt[] = [];
|
|
360
|
-
let sum = 0;
|
|
361
|
-
for (let i = 0; i < pts.length; i++) {
|
|
362
|
-
sum += pts[i]!.y;
|
|
363
|
-
if (i >= w) sum -= pts[i - w]!.y;
|
|
364
|
-
out.push({ step: pts[i]!.step, t: pts[i]!.t, y: sum / Math.min(i + 1, w) });
|
|
365
|
-
}
|
|
366
|
-
return out;
|
|
367
|
-
}
|
|
368
|
-
|
|
369
|
-
function minPt(pts: Pt[]): { y: number; step: number } | null {
|
|
370
|
-
if (!pts.length) return null;
|
|
371
|
-
let y = Infinity, step = pts[0]!.step;
|
|
372
|
-
for (const p of pts) if (p.y < y) { y = p.y; step = p.step; }
|
|
373
|
-
return { y, step };
|
|
374
|
-
}
|
|
375
|
-
|
|
376
|
-
/** Pull a readable model name from an HF cache path
|
|
377
|
-
* (".../models--mlx-community--MiniCPM5-1B-OptiQ-4bit/snapshots/<hash>") or a
|
|
378
|
-
* plain repo id, dropping the quant suffix. */
|
|
379
|
-
function shortModel(p: string): string {
|
|
380
|
-
const m = p.match(/models--[^/]*--([^/]+)/);
|
|
381
|
-
const name = m ? m[1]! : (p.split("/").filter((s) => s && s !== "snapshots" && !/^[0-9a-f]{16,}$/.test(s)).pop() ?? p);
|
|
382
|
-
return name.replace(/-OptiQ-4bit$/i, "");
|
|
383
|
-
}
|
|
384
|
-
|
|
385
|
-
function fmtDuration(sec: number): string {
|
|
386
|
-
if (!Number.isFinite(sec) || sec < 0) return "—";
|
|
387
|
-
const h = Math.floor(sec / 3600), m = Math.floor((sec % 3600) / 60), s = Math.floor(sec % 60);
|
|
388
|
-
if (h > 0) return `${h}h${String(m).padStart(2, "0")}m`;
|
|
389
|
-
if (m > 0) return `${m}m${String(s).padStart(2, "0")}s`;
|
|
390
|
-
return `${s}s`;
|
|
391
|
-
}
|
|
392
|
-
|
|
393
|
-
export function renderFrame(st: RunState, cols: number, rows: number): string[] {
|
|
394
|
-
const grid = new Grid(rows, cols);
|
|
395
|
-
const sps = sPerStep(st);
|
|
396
|
-
const iters = st.meta.iters || st.lastStep || 1;
|
|
397
|
-
const frac = Math.min(1, st.lastStep / iters);
|
|
398
|
-
const eta = sps > 0 ? (iters - st.lastStep) * sps : NaN;
|
|
399
|
-
const elapsed = st.meta.startedAt && st.lastT ? (st.lastT - st.meta.startedAt) / 1000 : NaN;
|
|
400
|
-
|
|
401
|
-
// Header line 1: identity + progress.
|
|
402
|
-
const modelShort = shortModel(st.meta.model);
|
|
403
|
-
const statusDot = st.done ? "✓" : "●";
|
|
404
|
-
const statusFg = st.done ? C.ok : C.accent;
|
|
405
|
-
grid.put(0, 0, statusDot, statusFg);
|
|
406
|
-
grid.text(0, 2, `${st.meta.method} · ${modelShort}`, C.title);
|
|
407
|
-
const hdr = st.done
|
|
408
|
-
? `step ${st.lastStep}/${iters} · complete · ${fmtDuration(elapsed)}`
|
|
409
|
-
: `step ${st.lastStep}/${iters} · ${(frac * 100).toFixed(1)}% · ETA ${fmtDuration(eta)} · ${sps ? sps.toFixed(2) + "s/step" : "—"} · ${fmtDuration(elapsed)}`;
|
|
410
|
-
grid.textRight(0, cols - 1, hdr, C.label);
|
|
411
|
-
|
|
412
|
-
// Progress bar (row 1).
|
|
413
|
-
const barW = cols - 2;
|
|
414
|
-
const filled = Math.round(frac * barW);
|
|
415
|
-
for (let i = 0; i < barW; i++) grid.put(1, 1 + i, i < filled ? "█" : "░", i < filled ? statusFg : C.dim);
|
|
416
|
-
|
|
417
|
-
// Legend (row 2): trend lines are smoothed because B=1 per-step is per-example;
|
|
418
|
-
// ● markers are val (sparse, one per eval step).
|
|
419
|
-
grid.text(2, 1, "loss = orpo/step · margin = confidence · accuracy = correct on val · ● = val @ eval", C.dim);
|
|
420
|
-
|
|
421
|
-
// Chart stack (rows 2..).
|
|
422
|
-
const top = 3;
|
|
423
|
-
const avail = rows - top;
|
|
424
|
-
if (avail < 6) return grid.render();
|
|
425
|
-
|
|
426
|
-
const cur = (a: Pt[]) => (a.length ? a[a.length - 1]!.y : NaN);
|
|
427
|
-
const pct = (v: number) => (Number.isFinite(v) ? `${(v * 100).toFixed(1)}%` : "—");
|
|
428
|
-
const SMOOTH = 15;
|
|
429
|
-
|
|
430
|
-
// Loss: at B=1 the per-step loss is per-EXAMPLE (a long/hard pair spikes it),
|
|
431
|
-
// so the displayed trend is a rolling MEDIAN (robust to those spikes) plus the
|
|
432
|
-
// sparse val-loss markers. The min/rising detector runs on the median — the
|
|
433
|
-
// "loss bottomed out and turned back up" early-stop signal.
|
|
434
|
-
const lossMed = rollingMean(rollingMedian(st.loss, SMOOTH), 5); // median despikes, mean de-bands
|
|
435
|
-
const lm = minPt(lossMed);
|
|
436
|
-
const recentLoss = cur(lossMed);
|
|
437
|
-
const rising = !!(lm && lossMed.length > SMOOTH + 5 && lm.step < st.lastStep - SMOOTH
|
|
438
|
-
&& recentLoss > lm.y * 1.05);
|
|
439
|
-
|
|
440
|
-
const marginMed = rollingMean(rollingMedian(st.margin, SMOOTH), 5); // median despikes, mean de-bands
|
|
441
|
-
|
|
442
|
-
// Val accuracy — the metric that matters: of the held-out val set, how many
|
|
443
|
-
// pairs it correctly prefers (chosen > rejected). Sparse (one per eval step),
|
|
444
|
-
// drawn as a line + markers; 50% is chance.
|
|
445
|
-
const lastVal = st.valAcc[st.valAcc.length - 1];
|
|
446
|
-
const bestVal = st.valAcc.reduce<ValPt | null>((b, p) => (!b || p.y > b.y ? p : b), null);
|
|
447
|
-
|
|
448
|
-
const panels: ChartOpts[] = [
|
|
449
|
-
{
|
|
450
|
-
title: "loss · orpo per step", series: lossMed, color: C.loss, fmtY: fmtNum,
|
|
451
|
-
markers: st.valLoss.length ? { pts: st.valLoss, color: C.accV } : undefined,
|
|
452
|
-
headline: `${fmtNum(recentLoss)}${lm ? ` ↓min ${fmtNum(lm.y)} @ ${lm.step}` : ""}`,
|
|
453
|
-
alert: rising ? `▲ rising +${fmtNum(recentLoss - lm!.y)} since ${lm!.step}` : undefined,
|
|
454
|
-
},
|
|
455
|
-
{
|
|
456
|
-
title: "margin · confidence in the choice", series: marginMed, color: C.margin,
|
|
457
|
-
fmtY: (v) => (v >= 0 ? "+" : "") + fmtNum(v),
|
|
458
|
-
zeroLine: true,
|
|
459
|
-
headline: `${cur(marginMed) >= 0 ? "+" : ""}${fmtNum(cur(marginMed))} ${cur(marginMed) > 0 ? "leans chosen" : "leans rejected"}`,
|
|
460
|
-
},
|
|
461
|
-
{
|
|
462
|
-
title: "accuracy · correct choices on val", series: st.valAcc, color: C.accT, fixed: [0, 1],
|
|
463
|
-
fmtY: (v) => `${Math.round(v * 100)}%`,
|
|
464
|
-
refLine: 0.5,
|
|
465
|
-
markers: { pts: st.valAcc, color: C.accT },
|
|
466
|
-
headline: lastVal
|
|
467
|
-
? `${lastVal.nc ?? "?"}/${lastVal.nt ?? "?"} correct = ${pct(lastVal.y)}`
|
|
468
|
-
+ (bestVal ? ` · best ${pct(bestVal.y)} @ ${bestVal.step}` : "")
|
|
469
|
-
: "awaiting first eval (50% = chance)",
|
|
470
|
-
},
|
|
471
|
-
{
|
|
472
|
-
title: "memory", series: st.peak, color: C.mem, fmtY: (v) => v.toFixed(1) + "G",
|
|
473
|
-
headline: `peak ${st.peakGb.toFixed(2)}GB · active ${fmtNum(cur(st.mem))}GB`,
|
|
474
|
-
},
|
|
475
|
-
];
|
|
476
|
-
|
|
477
|
-
const ph = Math.floor(avail / panels.length);
|
|
478
|
-
if (ph < 4) {
|
|
479
|
-
// Too short: only loss.
|
|
480
|
-
drawChart(grid, top, 1, avail, cols - 2, panels[0]!);
|
|
481
|
-
return grid.render();
|
|
482
|
-
}
|
|
483
|
-
panels.forEach((p, i) => drawChart(grid, top + i * ph, 1, ph, cols - 2, p));
|
|
484
|
-
return grid.render();
|
|
485
|
-
}
|
|
486
|
-
|
|
487
|
-
// ---------------------------------------------------------------------------
|
|
488
|
-
// Live TUI loop.
|
|
489
|
-
// ---------------------------------------------------------------------------
|
|
490
|
-
|
|
491
|
-
const ALT_ON = "\x1b[?1049h\x1b[?25l", ALT_OFF = "\x1b[?25h\x1b[?1049l\x1b[0m";
|
|
492
|
-
const HOME = "\x1b[H", CLEAR = "\x1b[2J";
|
|
493
|
-
|
|
494
|
-
export async function runWatch(adapterDir: string): Promise<void> {
|
|
495
|
-
const path = `${adapterDir}/metrics.jsonl`;
|
|
496
|
-
if (!(await Bun.file(path).exists())) {
|
|
497
|
-
console.error(`no metrics.jsonl in ${adapterDir} — is this an mlx-bun training run dir?`);
|
|
498
|
-
console.error(`(the trainer writes ${path} as it runs; point train-watch at the --adapter dir)`);
|
|
499
|
-
process.exit(1);
|
|
500
|
-
}
|
|
501
|
-
|
|
502
|
-
let lastRender = "";
|
|
503
|
-
const draw = (st: RunState) => {
|
|
504
|
-
const cols = process.stdout.columns ?? 100;
|
|
505
|
-
const rows = process.stdout.rows ?? 30;
|
|
506
|
-
const frame = HOME + renderFrame(st, cols, rows).join("\n");
|
|
507
|
-
if (frame !== lastRender) { process.stdout.write(frame); lastRender = frame; }
|
|
508
|
-
};
|
|
509
|
-
|
|
510
|
-
const restore = () => { process.stdout.write(ALT_OFF); };
|
|
511
|
-
const quit = (code = 0) => {
|
|
512
|
-
if (process.stdin.isTTY) process.stdin.setRawMode(false);
|
|
513
|
-
restore();
|
|
514
|
-
process.exit(code);
|
|
515
|
-
};
|
|
516
|
-
|
|
517
|
-
process.stdout.write(ALT_ON + CLEAR);
|
|
518
|
-
if (process.stdin.isTTY) {
|
|
519
|
-
process.stdin.setRawMode(true);
|
|
520
|
-
process.stdin.resume();
|
|
521
|
-
process.stdin.on("data", (b) => {
|
|
522
|
-
const s = b.toString();
|
|
523
|
-
if (s === "q" || s === "\x03" /* ^C */) quit(0);
|
|
524
|
-
});
|
|
525
|
-
}
|
|
526
|
-
process.on("SIGINT", () => quit(0));
|
|
527
|
-
process.on("SIGTERM", () => quit(0));
|
|
528
|
-
process.stdout.on("resize", () => { lastRender = ""; });
|
|
529
|
-
|
|
530
|
-
// Poll the file; reparse fully each tick (it's small, single-digit MB) — keeps
|
|
531
|
-
// the loop dead simple and robust to truncation/rotation. Stop redrawing once
|
|
532
|
-
// the run is done AND the file has settled.
|
|
533
|
-
let stableTicks = 0;
|
|
534
|
-
for (;;) {
|
|
535
|
-
let st: RunState;
|
|
536
|
-
try { st = parseStream(readFileSync(path, "utf8")); }
|
|
537
|
-
catch { st = emptyState(); }
|
|
538
|
-
draw(st);
|
|
539
|
-
if (st.done) { if (++stableTicks > 8) { /* keep showing final frame, slow poll */ } }
|
|
540
|
-
await Bun.sleep(st.done ? 1000 : 250);
|
|
541
|
-
}
|
|
542
|
-
}
|