mlx-bun 0.0.13 → 0.3.0

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.
Files changed (427) hide show
  1. package/README.md +71 -578
  2. package/bin/mlx-bun.mjs +1 -1
  3. package/docs/README.md +91 -0
  4. package/docs/{design → archive}/hlg-sampling.md +3 -3
  5. package/docs/{design → archive/investigations}/audio-input-plan.md +4 -4
  6. package/docs/{design → archive/investigations}/colibri-glm52-port.md +1 -1
  7. package/docs/{design → archive/investigations}/diffusion-gemma-port.md +3 -3
  8. package/docs/{design → archive/investigations}/faithful-l1-consolidation.md +4 -4
  9. package/docs/{investigations → archive/investigations}/hlg-sampling-investigation.md +1 -1
  10. package/docs/{design → archive/investigations}/minicpm5-decode-megakernel.md +11 -11
  11. package/docs/{design → archive/investigations}/optimization_plan.md +1 -1
  12. package/docs/archive/investigations/pareto-specialized-runtime-findings.md +92 -0
  13. package/docs/{design → archive/investigations}/pre-colibri-stabilization.md +5 -5
  14. package/docs/{design → archive/investigations}/the-dreaming-handoff.md +2 -2
  15. package/docs/{planning → archive/planning}/chat-ui-vision.md +1 -1
  16. package/docs/{planning → archive/planning}/chunk-task-roadmap.md +1 -1
  17. package/docs/{planning → archive/planning}/curve-sampler-research-plan.md +4 -4
  18. package/docs/{planning → archive/planning}/memory-docs-and-dag-plan.md +1 -1
  19. package/docs/{planning → archive/planning}/release-notes-v0.0.9.md +1 -1
  20. package/docs/archive/planning/release-notes-v0.2.0.md +49 -0
  21. package/docs/{planning → archive/planning}/website-readme-pass-plan.md +1 -1
  22. package/docs/design/batching.md +569 -0
  23. package/docs/design/decode-speed-program.md +3178 -169
  24. package/docs/design/dreaming-nightly-pipeline.md +675 -282
  25. package/docs/design/generic-model-support.md +579 -334
  26. package/docs/design/kv-cache.md +680 -0
  27. package/docs/design/orpo-training.md +673 -1239
  28. package/docs/design/speculative-decoding.md +1627 -0
  29. package/docs/design/turboquant.md +2102 -0
  30. package/docs/design/unified-engine-frontier-plan.md +1831 -727
  31. package/docs/design/web-chat-redesign.md +666 -1083
  32. package/docs/planning/PRODUCT_ROADMAP.md +34 -1
  33. package/docs/planning/release-notes-v0.3.0.md +35 -0
  34. package/docs/planning/video-series.md +421 -0
  35. package/docs/reference/benchmarks.md +1162 -0
  36. package/docs/reference/cli.md +395 -149
  37. package/docs/reference/distribution.md +321 -181
  38. package/docs/reference/environment.md +348 -0
  39. package/docs/reference/glossary.md +39 -0
  40. package/docs/reference/library-api.md +580 -142
  41. package/docs/reference/memory.md +227 -64
  42. package/docs/reference/models.md +137 -6
  43. package/docs/reference/server-api.md +938 -422
  44. package/docs/reference/server-config.md +652 -320
  45. package/docs/reference/training.md +393 -254
  46. package/docs/reference/troubleshooting.md +166 -0
  47. package/package.json +10 -5
  48. package/src/admission-plan.ts +18 -0
  49. package/src/anthropic.ts +92 -3
  50. package/src/audio/decode.ts +2 -2
  51. package/src/audio/features.ts +2 -2
  52. package/src/backends/mlx/autoregressive.ts +108 -0
  53. package/src/backends/mlx/batch-group.ts +1646 -0
  54. package/src/backends/mlx/diffusion.ts +39 -0
  55. package/src/backends/mlx/draft-target.ts +47 -0
  56. package/src/backends/mlx/evaluation.ts +105 -0
  57. package/src/backends/mlx/fill-append.ts +45 -0
  58. package/src/backends/mlx/gateway-binding.ts +71 -0
  59. package/src/backends/mlx/graph.ts +38 -0
  60. package/src/backends/mlx/memory-client.ts +189 -0
  61. package/src/backends/mlx/memory-guard.ts +40 -0
  62. package/src/backends/mlx/methods.ts +90 -0
  63. package/src/backends/mlx/model-prompt.ts +164 -0
  64. package/src/backends/mlx/model-serving.ts +79 -0
  65. package/src/backends/mlx/prefill.ts +33 -0
  66. package/src/backends/mlx/request-policy.ts +20 -0
  67. package/src/backends/mlx/rollback.ts +51 -0
  68. package/src/backends/mlx/serial-executor.ts +273 -0
  69. package/src/backends/mlx/speculative.ts +114 -0
  70. package/src/backends/mlx/state-views.ts +26 -0
  71. package/src/backends/mlx/text-engine.ts +73 -0
  72. package/src/cli.ts +189 -61
  73. package/src/client.ts +66 -0
  74. package/src/config.ts +21 -4
  75. package/src/contracts/completion.ts +16 -0
  76. package/src/contracts/execution.ts +49 -0
  77. package/src/contracts/generation.ts +103 -0
  78. package/src/contracts/host.ts +7 -0
  79. package/src/contracts/jobs.ts +66 -0
  80. package/src/contracts/pi-provider.ts +30 -0
  81. package/src/contracts/pi.ts +259 -0
  82. package/src/contracts/resources.ts +18 -0
  83. package/src/contracts/scheduling.ts +27 -0
  84. package/src/contracts/tsconfig.json +14 -0
  85. package/src/diffusion/diffusion-generate.ts +218 -153
  86. package/src/embed.ts +1 -1
  87. package/src/engine/admission.ts +59 -0
  88. package/src/engine/cancellation.ts +38 -0
  89. package/src/engine/engine.ts +56 -0
  90. package/src/engine/execution-coordinator.ts +70 -0
  91. package/src/engine/execution-plan.ts +51 -0
  92. package/src/engine/resources.ts +42 -0
  93. package/src/engine/scheduler.ts +67 -0
  94. package/src/engine/session.ts +247 -0
  95. package/src/eval/README.md +44 -1
  96. package/src/eval/kl.ts +263 -23
  97. package/src/eval/reference-logits.ts +207 -0
  98. package/src/eval/runner.ts +10 -79
  99. package/src/eval/tasks/bfcl.ts +2 -1
  100. package/src/eval/tasks/gsm8k.ts +2 -1
  101. package/src/eval/tasks/hashhop.ts +2 -1
  102. package/src/eval/tasks/humaneval.ts +5 -4
  103. package/src/eval/tasks/ifeval.ts +2 -1
  104. package/src/eval/tasks/mmlu.ts +2 -1
  105. package/src/expert-io.ts +3 -1
  106. package/src/expert-offload.ts +3 -1
  107. package/src/fill/echo-index.ts +188 -0
  108. package/src/fill/fill-session.ts +452 -0
  109. package/src/fill/proposal.ts +68 -0
  110. package/src/fill/schema-rows.ts +506 -0
  111. package/src/fill/tool-boundary.ts +103 -0
  112. package/src/fit.ts +24 -43
  113. package/src/generate.ts +740 -337
  114. package/src/grammar.ts +14 -8
  115. package/src/harness-pi.ts +15 -34
  116. package/src/hub-rest.ts +4 -4
  117. package/src/index.ts +27 -0
  118. package/src/inference/denoising.ts +17 -0
  119. package/src/inference/graph.ts +33 -0
  120. package/src/inference/prefill.ts +39 -0
  121. package/src/inference/rollback.ts +7 -0
  122. package/src/jobs/db.ts +4 -0
  123. package/src/jobs/index.ts +2 -0
  124. package/src/jobs/job-entry.ts +15 -14
  125. package/src/jobs/runner.ts +98 -13
  126. package/src/jobs/task-client.ts +19 -0
  127. package/src/jobs/types.ts +2 -58
  128. package/src/kv-scheme.ts +199 -0
  129. package/src/kv-store.ts +459 -234
  130. package/src/lab/README.md +18 -0
  131. package/src/{assets → lab/curve}/curve-designer.html +2 -2
  132. package/src/{curve-sampler.ts → lab/curve/curve-sampler.ts} +2 -2
  133. package/src/{expert-trace.ts → lab/expert-trace/expert-trace.ts} +7 -5
  134. package/src/{model → lab/paged-kv}/paged-kv.ts +9 -6
  135. package/src/library.ts +40 -0
  136. package/src/lora.ts +18 -0
  137. package/src/media-fetch.ts +20 -4
  138. package/src/memory/article.ts +1 -1
  139. package/src/memory/chunk.ts +1 -1
  140. package/src/memory/model.ts +32 -170
  141. package/src/memory/prompts.ts +1 -1
  142. package/src/memory/rest.ts +1 -1
  143. package/src/memory/schedule.ts +1 -1
  144. package/src/memory/stages.ts +4 -1
  145. package/src/memory/vault.ts +3 -4
  146. package/src/mlx/array.ts +9 -10
  147. package/src/mlx/compile.ts +12 -2
  148. package/src/mlx/ffi.ts +66 -22
  149. package/src/mlx/materialize.ts +23 -0
  150. package/src/mlx/metal-kernel.ts +10 -11
  151. package/src/mlx/ops.ts +64 -19
  152. package/src/mlx/quantized-rows.ts +36 -0
  153. package/src/mlx/token-bitmask.ts +52 -0
  154. package/src/mlx/turboquant-kv-decode.ts +82 -0
  155. package/src/mlx/turboquant-kv-inverse.ts +122 -0
  156. package/src/mlx/turboquant-ops.ts +33 -3
  157. package/src/mlx/turboquant-tables.ts +2 -2
  158. package/src/model/batched-mask.ts +9 -3
  159. package/src/model/batched-quant.ts +2 -2
  160. package/src/model/batched-rotating-quant.ts +45 -99
  161. package/src/model/batched-rotating-state.ts +73 -0
  162. package/src/model/batched-rotating.ts +57 -106
  163. package/src/model/batched-row-storage.ts +127 -0
  164. package/src/model/compiled-decode.ts +4 -3
  165. package/src/model/diffusion-gemma.ts +8 -8
  166. package/src/model/factory.ts +87 -49
  167. package/src/model/fingerprint.ts +1 -1
  168. package/src/model/flash-attention.ts +3 -3
  169. package/src/model/gemma4-base.ts +187 -17
  170. package/src/model/gemma4.ts +6 -6
  171. package/src/model/generated/gemma4-12b.ts +52 -148
  172. package/src/model/generated/gemma4-26b.ts +34 -94
  173. package/src/model/generated/gemma4-e4b.ts +28 -76
  174. package/src/model/glm52-cache.ts +19 -9
  175. package/src/model/glm52-dsa.ts +13 -9
  176. package/src/model/glm52-memory.ts +28 -1
  177. package/src/model/implementation.ts +69 -0
  178. package/src/model/minicpm5.ts +7 -8
  179. package/src/model/profile.ts +443 -0
  180. package/src/model/qwen-append.ts +12 -0
  181. package/src/model/qwen-conv.ts +52 -0
  182. package/src/model/qwen3-delta.ts +98 -2
  183. package/src/model/qwen3-mrope.ts +252 -0
  184. package/src/model/qwen3_5-checkpoint.ts +164 -0
  185. package/src/model/qwen3_5.ts +336 -60
  186. package/src/model/support.ts +1 -1
  187. package/src/model/trellis-balanced-scatter.ts +51 -0
  188. package/src/model/trellis-linear.ts +672 -0
  189. package/src/model/trellis-shared-m.ts +151 -0
  190. package/src/model/trellis-shared-scatter.ts +144 -0
  191. package/src/model/trellis-splitk-prefill.ts +158 -0
  192. package/src/model/trellis-tiled-prefill.ts +136 -0
  193. package/src/model/trellis-vector-expand.ts +62 -0
  194. package/src/model/trellis-wide-prefill.ts +94 -0
  195. package/src/model/universal/dense.ts +1 -1
  196. package/src/model/universal/modules.ts +8 -1
  197. package/src/native/frame_extract.swift +140 -0
  198. package/src/native-pack.ts +15 -7
  199. package/src/pi-provider.ts +6 -64
  200. package/src/pi-web.ts +14 -320
  201. package/src/prompt-cache.ts +97 -54
  202. package/src/quantize/allocator.ts +1 -1
  203. package/src/quantize/atomic-output.ts +38 -0
  204. package/src/quantize/calibration.ts +2 -1
  205. package/src/quantize/config-writer.ts +16 -3
  206. package/src/quantize/index.ts +28 -1
  207. package/src/quantize/job.ts +11 -0
  208. package/src/quantize/quantizer.ts +265 -100
  209. package/src/quantize/rotate.ts +603 -0
  210. package/src/quantize/safetensors-writer.ts +79 -1
  211. package/src/quantize/sensitivity.ts +41 -59
  212. package/src/quantize/trellis.ts +586 -0
  213. package/src/quantize/weight-transform.ts +292 -0
  214. package/src/responses.ts +130 -3
  215. package/src/runtime-config.ts +81 -0
  216. package/src/sampler.ts +276 -4
  217. package/src/serve/admin-routes.ts +184 -0
  218. package/src/serve/aux-routes.ts +132 -0
  219. package/src/serve/batch-scheduler.ts +8 -1558
  220. package/src/serve/chat-request.ts +391 -0
  221. package/src/serve/chat-stage.ts +195 -0
  222. package/src/serve/checkpoint-identity.ts +64 -0
  223. package/src/serve/completion-executor.ts +354 -0
  224. package/src/serve/completion-sink.ts +157 -0
  225. package/src/serve/discovery-routes.ts +204 -0
  226. package/src/serve/generation-gateway.ts +211 -243
  227. package/src/serve/http.ts +159 -0
  228. package/src/serve/inference-request.ts +79 -0
  229. package/src/serve/isolate.ts +297 -105
  230. package/src/serve/kv-budget.ts +14 -0
  231. package/src/serve/lab-routes.ts +341 -0
  232. package/src/serve/lane-registry.ts +2 -1
  233. package/src/serve/model-admin-routes.ts +155 -0
  234. package/src/serve/model-binding.ts +43 -0
  235. package/src/serve/model-host.ts +512 -0
  236. package/src/serve/openai-wire.ts +103 -0
  237. package/src/serve/parent-application.ts +56 -0
  238. package/src/serve/pipeline.ts +33 -0
  239. package/src/serve/preparation.ts +33 -0
  240. package/src/serve/prompt-response-trace.ts +147 -0
  241. package/src/serve/request-plan.ts +162 -0
  242. package/src/serve/request-prep.ts +346 -0
  243. package/src/serve/response-lease.ts +24 -0
  244. package/src/serve/responses-client.ts +69 -0
  245. package/src/serve/session-completion-engine.ts +110 -0
  246. package/src/serve/session-search.ts +1 -1
  247. package/src/serve/static-routes.ts +101 -0
  248. package/src/serve/text-completion-stage.ts +104 -0
  249. package/src/serve/token-streams.ts +519 -0
  250. package/src/server.ts +523 -3551
  251. package/src/spec/assistant-source.ts +10 -44
  252. package/src/spec/deepspec-source.ts +6 -6
  253. package/src/spec/dflash-source.ts +9 -8
  254. package/src/spec/dspark/ab-stats.ts +3 -3
  255. package/src/spec/dspark/calibration.ts +2 -2
  256. package/src/spec/dspark/deepspec-module.ts +2 -2
  257. package/src/spec/dspark/generate-dflash.ts +2 -2
  258. package/src/spec/dspark/generate.ts +2 -2
  259. package/src/spec/dspark/loader.ts +1 -1
  260. package/src/spec/dspark/module-dflash.ts +5 -4
  261. package/src/spec/dspark/module.ts +3 -3
  262. package/src/spec/dspark/quantize-drafter.ts +1 -1
  263. package/src/spec/generate.ts +1 -1
  264. package/src/spec/glm52-mtp-source.ts +9 -1
  265. package/src/spec/ngram-source.ts +1 -1
  266. package/src/spec/prefix-state.ts +45 -0
  267. package/src/spec/qwen-mtp-source.ts +528 -0
  268. package/src/spec/serve-loop.ts +179 -213
  269. package/src/spec/source.ts +42 -10
  270. package/src/spec/two-model.ts +1 -1
  271. package/src/ssd-cache.ts +95 -24
  272. package/src/ssd-durability.ts +207 -0
  273. package/src/tool-call.ts +51 -14
  274. package/src/train/flash-cce.ts +14 -13
  275. package/src/train/forward.ts +38 -1
  276. package/src/train/loss.ts +3 -2
  277. package/src/train/prefix-shared.ts +5 -1
  278. package/src/train/segmented.ts +9 -6
  279. package/src/train/trainer.ts +9 -8
  280. package/src/vision/preprocess.ts +1 -1
  281. package/src/vision/prompt.ts +48 -2
  282. package/src/vision/qwen3vl-preprocess.ts +409 -0
  283. package/src/vision/qwen3vl-prompt.ts +196 -0
  284. package/src/vision/qwen3vl-tower.ts +555 -0
  285. package/src/vision/video-frames.ts +148 -0
  286. package/src/web/app.html +1 -1
  287. package/src/web/app.js +15 -6
  288. package/src/web/src/adapters-panel.ts +3 -3
  289. package/src/web/src/assistant.ts +2 -36
  290. package/src/web/src/chat.ts +16 -9
  291. package/src/web/src/composer.ts +4 -4
  292. package/src/web/src/hub.ts +1 -1
  293. package/src/web/src/markdown.ts +0 -0
  294. package/src/web/src/memory-panel.ts +1 -1
  295. package/src/web/src/model-picker.ts +1 -1
  296. package/src/web/src/palette.ts +1 -1
  297. package/src/web/src/protocol.ts +3 -39
  298. package/src/web/src/sessions.ts +1 -1
  299. package/src/web/src/ui-catalog.ts +4 -18
  300. package/src/web-assets.ts +47 -0
  301. package/src/weights.ts +134 -33
  302. package/docs/archive/mlx-bun-lab-report.html +0 -275
  303. package/docs/archive/wwdc-mlx-bun.html +0 -848
  304. package/docs/design/adapters-end-to-end.md +0 -205
  305. package/docs/design/batching-perf-path.md +0 -255
  306. package/docs/design/batching-v2-plan.md +0 -529
  307. package/docs/design/bucketing-stage.md +0 -197
  308. package/docs/design/compat-cli-surface-design.md +0 -230
  309. package/docs/design/docs-reorg-plan.md +0 -165
  310. package/docs/design/dspark-serving-program.md +0 -371
  311. package/docs/design/dspark-speculative-decoding.md +0 -268
  312. package/docs/design/grammar-spec-batching-integration.md +0 -301
  313. package/docs/design/memory-inference-path.md +0 -422
  314. package/docs/design/memory-synthesis.md +0 -287
  315. package/docs/design/memory-system.md +0 -237
  316. package/docs/design/mlx-lm-tool-parity-plan.md +0 -545
  317. package/docs/design/omlx-adoption-map.md +0 -98
  318. package/docs/design/orpo-dynamic-lambda.md +0 -220
  319. package/docs/design/orpo-future-enhancements.md +0 -340
  320. package/docs/design/paged-kv-cache.md +0 -214
  321. package/docs/design/parallel-slots.md +0 -567
  322. package/docs/design/parity-tier-dag.md +0 -161
  323. package/docs/design/repo-cleanup-plan.md +0 -189
  324. package/docs/design/runtime-isolation.md +0 -263
  325. package/docs/design/segmented-backward-training.md +0 -543
  326. package/docs/design/spec-decode-larger-targets.md +0 -229
  327. package/docs/design/ssd-kv-cold-tier.md +0 -288
  328. package/docs/design/structured-output.md +0 -484
  329. package/docs/design/tauri-desktop-app.md +0 -127
  330. package/docs/design/the-dreaming-master-plan.md +0 -445
  331. package/docs/design/turboquant-kv.md +0 -235
  332. package/docs/design/w4a16-compute-precision-spike.md +0 -232
  333. package/docs/design/web-chat-beat-matrix.md +0 -315
  334. package/docs/design/write-pipeline-entity.md +0 -468
  335. package/docs/investigations/curve-runs/bisector-route.json +0 -98
  336. package/docs/investigations/curve-runs/bisector-trace.json +0 -1393
  337. package/docs/investigations/curve-runs/curve-terrain.json +0 -1179
  338. package/docs/investigations/curve-runs/curve-terrain.svg +0 -42
  339. package/docs/investigations/hlg-ab.html +0 -34
  340. package/docs/investigations/hlg-explorer.html +0 -238
  341. package/docs/investigations/hlg-figs/cliff.svg +0 -1
  342. package/docs/investigations/hlg-figs/cross.svg +0 -1
  343. package/docs/investigations/hlg-figs/edges.svg +0 -1
  344. package/docs/investigations/hlg-figs/k20-grids.svg +0 -1
  345. package/docs/investigations/hlg-figs/k20-peaks.svg +0 -1
  346. package/docs/investigations/hlg-figs/lw-curve.svg +0 -1
  347. package/docs/investigations/hlg-figs/terrain.svg +0 -1
  348. package/docs/investigations/hlg-report.html +0 -262
  349. package/docs/investigations/hlg-runs/hlg-ab.json +0 -149
  350. package/docs/investigations/hlg-runs/hlg-blind-prompts.md +0 -87
  351. package/docs/investigations/hlg-runs/hlg-blind.json +0 -107
  352. package/docs/investigations/hlg-runs/hlg-cliff.json +0 -197
  353. package/docs/investigations/hlg-runs/hlg-cross.json +0 -505
  354. package/docs/investigations/hlg-runs/hlg-edge.json +0 -388
  355. package/docs/investigations/hlg-runs/hlg-eval1.json +0 -25
  356. package/docs/investigations/hlg-runs/hlg-frontier.json +0 -86
  357. package/docs/investigations/hlg-runs/hlg-grid.json +0 -242
  358. package/docs/investigations/hlg-runs/hlg-grid2.json +0 -343
  359. package/docs/investigations/hlg-runs/hlg-lw-coarse.json +0 -110
  360. package/docs/investigations/hlg-runs/hlg-map.json +0 -383
  361. package/docs/investigations/hlg-runs/run0-original.log +0 -159
  362. package/docs/investigations/hlg-runs/runA-recommended.log +0 -131
  363. package/docs/investigations/hlg-runs/runAA-cliff-cpm5.log +0 -26
  364. package/docs/investigations/hlg-runs/runAB-grid2-e4b.log +0 -18
  365. package/docs/investigations/hlg-runs/runAC-peak-e4b-partial.log +0 -25
  366. package/docs/investigations/hlg-runs/runAD-ab-samples.log +0 -8
  367. package/docs/investigations/hlg-runs/runAE-edge-e4b.log +0 -51
  368. package/docs/investigations/hlg-runs/runAF-logit-calibration.log +0 -10
  369. package/docs/investigations/hlg-runs/runAG-eval-config2.log +0 -16
  370. package/docs/investigations/hlg-runs/runB-gain-variants.log +0 -192
  371. package/docs/investigations/hlg-runs/runC-pivots.log +0 -162
  372. package/docs/investigations/hlg-runs/runD-oetf.log +0 -132
  373. package/docs/investigations/hlg-runs/runE-eotf.log +0 -162
  374. package/docs/investigations/hlg-runs/runF-shaper-working.log +0 -86
  375. package/docs/investigations/hlg-runs/runG-shaper-sweep.log +0 -101
  376. package/docs/investigations/hlg-runs/runH-wsweep.log +0 -176
  377. package/docs/investigations/hlg-runs/runI-asweep.log +0 -176
  378. package/docs/investigations/hlg-runs/runJ-knobmap.log +0 -49
  379. package/docs/investigations/hlg-runs/runK-frontier.log +0 -19
  380. package/docs/investigations/hlg-runs/runL-widemap.log +0 -65
  381. package/docs/investigations/hlg-runs/runM-frontier-emb.log +0 -15
  382. package/docs/investigations/hlg-runs/runN-verify.log +0 -116
  383. package/docs/investigations/hlg-runs/runO-frontier-seed5000.log +0 -15
  384. package/docs/investigations/hlg-runs/runP-grid-AxSm.log +0 -15
  385. package/docs/investigations/hlg-runs/runQ-grid-tgW.log +0 -15
  386. package/docs/investigations/hlg-runs/runR-blind-transcript.log +0 -79
  387. package/docs/investigations/hlg-runs/runS-cross-coherence.log +0 -33
  388. package/docs/investigations/hlg-runs/runT-cross-cpm5.log +0 -33
  389. package/docs/investigations/hlg-runs/runU-cross-12b.log +0 -33
  390. package/docs/investigations/hlg-runs/runV-lw-e4b.log +0 -22
  391. package/docs/investigations/hlg-runs/runW-lw-12b.log +0 -22
  392. package/docs/investigations/hlg-runs/runX-lw-cpm5.log +0 -22
  393. package/docs/investigations/hlg-runs/runY-cliff-e4b.log +0 -26
  394. package/docs/investigations/hlg-runs/runZ-cliff-12b.log +0 -26
  395. package/docs/planning/IDEAS.md +0 -69
  396. package/docs/reference/embedding.md +0 -197
  397. package/docs/reference/features-matrix.md +0 -100
  398. package/docs/reference/orpo-quickstart.md +0 -135
  399. package/src/flags.ts +0 -13
  400. /package/docs/{investigations → archive/investigations}/chunk-finetune-experiment.md +0 -0
  401. /package/docs/{investigations → archive/investigations}/colibri-oracle-pin.md +0 -0
  402. /package/docs/{investigations → archive/investigations}/curve-bisector-routing.md +0 -0
  403. /package/docs/{investigations → archive/investigations}/decode-roofline-lookagain.md +0 -0
  404. /package/docs/{investigations → archive/investigations}/dspark-handoff.md +0 -0
  405. /package/docs/{investigations → archive/investigations}/expert-offload-single-user-moe.md +0 -0
  406. /package/docs/{investigations → archive/investigations}/indexshare-performance-spike.md +0 -0
  407. /package/docs/{investigations → archive/investigations}/kernel-perf-review-2026-07.md +0 -0
  408. /package/docs/{investigations → archive/investigations}/lab-build-journal.md +0 -0
  409. /package/docs/{investigations → archive/investigations}/openwiki-evaluation.md +0 -0
  410. /package/docs/{investigations → archive/investigations}/orpo-base-uf-experiment-and-directions.md +0 -0
  411. /package/docs/{investigations → archive/investigations}/orpo-flash-cce-pin-leak.md +0 -0
  412. /package/docs/{investigations → archive/investigations}/orpo-uf-testing-handoff.md +0 -0
  413. /package/docs/{investigations → archive/investigations}/pi-builtin-investigation.md +0 -0
  414. /package/docs/{investigations → archive/investigations}/segmented-backward-handoff.md +0 -0
  415. /package/docs/{investigations → archive/investigations}/starter-model-port-handoff.md +0 -0
  416. /package/docs/{investigations → archive/investigations}/steel-flash-cce-handoff.md +0 -0
  417. /package/docs/{investigations → archive/investigations}/trainer-validation-experiment.md +0 -0
  418. /package/docs/{planning → archive/planning}/chat-ui-vision-memory-first.md +0 -0
  419. /package/docs/{planning → archive/planning}/journal.md +0 -0
  420. /package/docs/{planning → archive/planning}/pi-integration-review.md +0 -0
  421. /package/docs/{planning → archive/planning}/release-notes-v0.0.10.md +0 -0
  422. /package/docs/{planning → archive/planning}/release-notes-v0.0.11.md +0 -0
  423. /package/docs/{planning → archive/planning}/release-notes-v0.0.12.md +0 -0
  424. /package/docs/{planning → archive/planning}/release-notes-v0.0.13.md +0 -0
  425. /package/docs/{planning → archive/planning}/web-ui-pass-plan.md +0 -0
  426. /package/{docs/investigations → src/serve/assets}/curve-terrain.html +0 -0
  427. /package/{docs/dag → src/serve/assets}/training-inference-map.html +0 -0
package/docs/investigations/hlg-runs/hlg-cross.json DELETED
@@ -1,505 +0,0 @@
1
- {
2
- "model": "gemma-4-12B-it-OptiQ-4bit",
3
- "base": {
4
- "window": 6,
5
- "shoulderA": 0.35,
6
- "sM": 0.7,
7
- "targetGap": 15,
8
- "refGap": 4,
9
- "lw": 1200
10
- },
11
- "vals": {
12
- "shoulderA": [
13
- 0.5,
14
- 5,
15
- 30,
16
- 100
17
- ],
18
- "sM": [
19
- 0.2,
20
- 0.7,
21
- 2,
22
- 4
23
- ],
24
- "targetGap": [
25
- 8,
26
- 12,
27
- 20,
28
- 50
29
- ],
30
- "window": [
31
- 3,
32
- 5,
33
- 10,
34
- 30
35
- ]
36
- },
37
- "planes": [
38
- {
39
- "y": "shoulderA",
40
- "x": "targetGap"
41
- },
42
- {
43
- "y": "shoulderA",
44
- "x": "window"
45
- },
46
- {
47
- "y": "sM",
48
- "x": "targetGap"
49
- },
50
- {
51
- "y": "sM",
52
- "x": "window"
53
- }
54
- ],
55
- "grid": [
56
- {
57
- "plane": "shoulderA×targetGap",
58
- "shoulderA": 0.5,
59
- "targetGap": 8,
60
- "junk": 0.11678832116788321,
61
- "ok": false
62
- },
63
- {
64
- "plane": "shoulderA×targetGap",
65
- "shoulderA": 0.5,
66
- "targetGap": 12,
67
- "junk": 0,
68
- "ok": true
69
- },
70
- {
71
- "plane": "shoulderA×targetGap",
72
- "shoulderA": 0.5,
73
- "targetGap": 20,
74
- "junk": 0,
75
- "ok": true
76
- },
77
- {
78
- "plane": "shoulderA×targetGap",
79
- "shoulderA": 0.5,
80
- "targetGap": 50,
81
- "junk": 0,
82
- "ok": true
83
- },
84
- {
85
- "plane": "shoulderA×targetGap",
86
- "shoulderA": 5,
87
- "targetGap": 8,
88
- "junk": 0.17679558011049723,
89
- "ok": false
90
- },
91
- {
92
- "plane": "shoulderA×targetGap",
93
- "shoulderA": 5,
94
- "targetGap": 12,
95
- "junk": 0,
96
- "ok": true
97
- },
98
- {
99
- "plane": "shoulderA×targetGap",
100
- "shoulderA": 5,
101
- "targetGap": 20,
102
- "junk": 0,
103
- "ok": true
104
- },
105
- {
106
- "plane": "shoulderA×targetGap",
107
- "shoulderA": 5,
108
- "targetGap": 50,
109
- "junk": 0,
110
- "ok": true
111
- },
112
- {
113
- "plane": "shoulderA×targetGap",
114
- "shoulderA": 30,
115
- "targetGap": 8,
116
- "junk": 0.25,
117
- "ok": false
118
- },
119
- {
120
- "plane": "shoulderA×targetGap",
121
- "shoulderA": 30,
122
- "targetGap": 12,
123
- "junk": 0,
124
- "ok": true
125
- },
126
- {
127
- "plane": "shoulderA×targetGap",
128
- "shoulderA": 30,
129
- "targetGap": 20,
130
- "junk": 0,
131
- "ok": true
132
- },
133
- {
134
- "plane": "shoulderA×targetGap",
135
- "shoulderA": 30,
136
- "targetGap": 50,
137
- "junk": 0,
138
- "ok": true
139
- },
140
- {
141
- "plane": "shoulderA×targetGap",
142
- "shoulderA": 100,
143
- "targetGap": 8,
144
- "junk": 0.2721518987341772,
145
- "ok": false
146
- },
147
- {
148
- "plane": "shoulderA×targetGap",
149
- "shoulderA": 100,
150
- "targetGap": 12,
151
- "junk": 0.21768707482993196,
152
- "ok": false
153
- },
154
- {
155
- "plane": "shoulderA×targetGap",
156
- "shoulderA": 100,
157
- "targetGap": 20,
158
- "junk": 0,
159
- "ok": true
160
- },
161
- {
162
- "plane": "shoulderA×targetGap",
163
- "shoulderA": 100,
164
- "targetGap": 50,
165
- "junk": 0,
166
- "ok": true
167
- },
168
- {
169
- "plane": "shoulderA×window",
170
- "shoulderA": 0.5,
171
- "window": 3,
172
- "junk": 0.02142857142857143,
173
- "ok": false
174
- },
175
- {
176
- "plane": "shoulderA×window",
177
- "shoulderA": 0.5,
178
- "window": 5,
179
- "junk": 0,
180
- "ok": true
181
- },
182
- {
183
- "plane": "shoulderA×window",
184
- "shoulderA": 0.5,
185
- "window": 10,
186
- "junk": 0,
187
- "ok": true
188
- },
189
- {
190
- "plane": "shoulderA×window",
191
- "shoulderA": 0.5,
192
- "window": 30,
193
- "junk": 0,
194
- "ok": true
195
- },
196
- {
197
- "plane": "shoulderA×window",
198
- "shoulderA": 5,
199
- "window": 3,
200
- "junk": 0.02142857142857143,
201
- "ok": false
202
- },
203
- {
204
- "plane": "shoulderA×window",
205
- "shoulderA": 5,
206
- "window": 5,
207
- "junk": 0,
208
- "ok": true
209
- },
210
- {
211
- "plane": "shoulderA×window",
212
- "shoulderA": 5,
213
- "window": 10,
214
- "junk": 0,
215
- "ok": true
216
- },
217
- {
218
- "plane": "shoulderA×window",
219
- "shoulderA": 5,
220
- "window": 30,
221
- "junk": 0,
222
- "ok": true
223
- },
224
- {
225
- "plane": "shoulderA×window",
226
- "shoulderA": 30,
227
- "window": 3,
228
- "junk": 0.09333333333333334,
229
- "ok": false
230
- },
231
- {
232
- "plane": "shoulderA×window",
233
- "shoulderA": 30,
234
- "window": 5,
235
- "junk": 0.02054794520547945,
236
- "ok": false
237
- },
238
- {
239
- "plane": "shoulderA×window",
240
- "shoulderA": 30,
241
- "window": 10,
242
- "junk": 0,
243
- "ok": true
244
- },
245
- {
246
- "plane": "shoulderA×window",
247
- "shoulderA": 30,
248
- "window": 30,
249
- "junk": 0,
250
- "ok": true
251
- },
252
- {
253
- "plane": "shoulderA×window",
254
- "shoulderA": 100,
255
- "window": 3,
256
- "junk": 0.27388535031847133,
257
- "ok": false
258
- },
259
- {
260
- "plane": "shoulderA×window",
261
- "shoulderA": 100,
262
- "window": 5,
263
- "junk": 0.2857142857142857,
264
- "ok": false
265
- },
266
- {
267
- "plane": "shoulderA×window",
268
- "shoulderA": 100,
269
- "window": 10,
270
- "junk": 0,
271
- "ok": true
272
- },
273
- {
274
- "plane": "shoulderA×window",
275
- "shoulderA": 100,
276
- "window": 30,
277
- "junk": 0.049019607843137254,
278
- "ok": false
279
- },
280
- {
281
- "plane": "sM×targetGap",
282
- "sM": 0.2,
283
- "targetGap": 8,
284
- "junk": 0,
285
- "ok": true
286
- },
287
- {
288
- "plane": "sM×targetGap",
289
- "sM": 0.2,
290
- "targetGap": 12,
291
- "junk": 0,
292
- "ok": true
293
- },
294
- {
295
- "plane": "sM×targetGap",
296
- "sM": 0.2,
297
- "targetGap": 20,
298
- "junk": 0,
299
- "ok": true
300
- },
301
- {
302
- "plane": "sM×targetGap",
303
- "sM": 0.2,
304
- "targetGap": 50,
305
- "junk": 0,
306
- "ok": true
307
- },
308
- {
309
- "plane": "sM×targetGap",
310
- "sM": 0.7,
311
- "targetGap": 8,
312
- "junk": 0.19117647058823528,
313
- "ok": false
314
- },
315
- {
316
- "plane": "sM×targetGap",
317
- "sM": 0.7,
318
- "targetGap": 12,
319
- "junk": 0,
320
- "ok": true
321
- },
322
- {
323
- "plane": "sM×targetGap",
324
- "sM": 0.7,
325
- "targetGap": 20,
326
- "junk": 0,
327
- "ok": true
328
- },
329
- {
330
- "plane": "sM×targetGap",
331
- "sM": 0.7,
332
- "targetGap": 50,
333
- "junk": 0,
334
- "ok": true
335
- },
336
- {
337
- "plane": "sM×targetGap",
338
- "sM": 2,
339
- "targetGap": 8,
340
- "junk": 0.27631578947368424,
341
- "ok": false
342
- },
343
- {
344
- "plane": "sM×targetGap",
345
- "sM": 2,
346
- "targetGap": 12,
347
- "junk": 0.08264462809917356,
348
- "ok": false
349
- },
350
- {
351
- "plane": "sM×targetGap",
352
- "sM": 2,
353
- "targetGap": 20,
354
- "junk": 0,
355
- "ok": true
356
- },
357
- {
358
- "plane": "sM×targetGap",
359
- "sM": 2,
360
- "targetGap": 50,
361
- "junk": 0,
362
- "ok": true
363
- },
364
- {
365
- "plane": "sM×targetGap",
366
- "sM": 4,
367
- "targetGap": 8,
368
- "junk": 0.27388535031847133,
369
- "ok": false
370
- },
371
- {
372
- "plane": "sM×targetGap",
373
- "sM": 4,
374
- "targetGap": 12,
375
- "junk": 0.15104166666666666,
376
- "ok": false
377
- },
378
- {
379
- "plane": "sM×targetGap",
380
- "sM": 4,
381
- "targetGap": 20,
382
- "junk": 0,
383
- "ok": true
384
- },
385
- {
386
- "plane": "sM×targetGap",
387
- "sM": 4,
388
- "targetGap": 50,
389
- "junk": 0,
390
- "ok": true
391
- },
392
- {
393
- "plane": "sM×window",
394
- "sM": 0.2,
395
- "window": 3,
396
- "junk": 0.02142857142857143,
397
- "ok": false
398
- },
399
- {
400
- "plane": "sM×window",
401
- "sM": 0.2,
402
- "window": 5,
403
- "junk": 0,
404
- "ok": true
405
- },
406
- {
407
- "plane": "sM×window",
408
- "sM": 0.2,
409
- "window": 10,
410
- "junk": 0,
411
- "ok": true
412
- },
413
- {
414
- "plane": "sM×window",
415
- "sM": 0.2,
416
- "window": 30,
417
- "junk": 0,
418
- "ok": true
419
- },
420
- {
421
- "plane": "sM×window",
422
- "sM": 0.7,
423
- "window": 3,
424
- "junk": 0.02142857142857143,
425
- "ok": false
426
- },
427
- {
428
- "plane": "sM×window",
429
- "sM": 0.7,
430
- "window": 5,
431
- "junk": 0,
432
- "ok": true
433
- },
434
- {
435
- "plane": "sM×window",
436
- "sM": 0.7,
437
- "window": 10,
438
- "junk": 0,
439
- "ok": true
440
- },
441
- {
442
- "plane": "sM×window",
443
- "sM": 0.7,
444
- "window": 30,
445
- "junk": 0,
446
- "ok": true
447
- },
448
- {
449
- "plane": "sM×window",
450
- "sM": 2,
451
- "window": 3,
452
- "junk": 0.02142857142857143,
453
- "ok": false
454
- },
455
- {
456
- "plane": "sM×window",
457
- "sM": 2,
458
- "window": 5,
459
- "junk": 0.020134228187919462,
460
- "ok": false
461
- },
462
- {
463
- "plane": "sM×window",
464
- "sM": 2,
465
- "window": 10,
466
- "junk": 0,
467
- "ok": true
468
- },
469
- {
470
- "plane": "sM×window",
471
- "sM": 2,
472
- "window": 30,
473
- "junk": 0,
474
- "ok": true
475
- },
476
- {
477
- "plane": "sM×window",
478
- "sM": 4,
479
- "window": 3,
480
- "junk": 0.02142857142857143,
481
- "ok": false
482
- },
483
- {
484
- "plane": "sM×window",
485
- "sM": 4,
486
- "window": 5,
487
- "junk": 0.020134228187919462,
488
- "ok": false
489
- },
490
- {
491
- "plane": "sM×window",
492
- "sM": 4,
493
- "window": 10,
494
- "junk": 0,
495
- "ok": true
496
- },
497
- {
498
- "plane": "sM×window",
499
- "sM": 4,
500
- "window": 30,
501
- "junk": 0,
502
- "ok": true
503
- }
504
- ]
505
- }