mlx-bun 0.5.0 → 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.
Files changed (640) hide show
  1. package/README.md +1449 -109
  2. package/THIRD_PARTY_NOTICES.md +882 -0
  3. package/bin/mlx-bun.convert.mjs +4 -0
  4. package/bin/mlx-bun.fuse.mjs +4 -0
  5. package/bin/mlx-bun.generate.mjs +4 -0
  6. package/bin/mlx-bun.lora.mjs +4 -0
  7. package/bin/mlx-bun.mjs +13 -21
  8. package/bin/mlx-bun.server.mjs +4 -0
  9. package/bin/mlx-bun.upload.mjs +4 -0
  10. package/dist/web/app.js +9584 -0
  11. package/package.json +50 -53
  12. package/src/cli/ane-install.ts +76 -0
  13. package/src/cli/args.ts +179 -0
  14. package/src/cli/engine-entry.ts +68 -0
  15. package/src/cli/inference.ts +214 -0
  16. package/src/cli/job-entry.ts +97 -0
  17. package/src/cli/library-host.ts +117 -0
  18. package/src/cli/main.ts +57 -0
  19. package/src/cli/memory-engine.ts +163 -0
  20. package/src/cli/memory.ts +36 -0
  21. package/src/cli/mlx-lm-aliases.ts +309 -0
  22. package/src/cli/model-choice.ts +33 -0
  23. package/src/cli/model-selection.ts +84 -0
  24. package/src/cli/module-host.ts +11 -0
  25. package/src/cli/module-verbs.ts +50 -0
  26. package/src/cli/numerical-policy.ts +12 -0
  27. package/src/cli/perplexity.ts +89 -0
  28. package/src/cli/serve-host.ts +423 -0
  29. package/src/cli/serve-isolated.ts +217 -0
  30. package/src/cli/serve-options.ts +106 -0
  31. package/src/cli/serve-state.ts +205 -0
  32. package/src/cli/serve.ts +500 -0
  33. package/src/cli/served-model-host.ts +102 -0
  34. package/src/cli/server-entry.ts +148 -0
  35. package/src/cli/serving-unit.ts +177 -0
  36. package/src/{tui.ts → cli/terminal.ts} +13 -8
  37. package/src/cli/worker-adapters.ts +34 -0
  38. package/src/cli/worker-entry.ts +272 -0
  39. package/src/cli/worker-unit.ts +179 -0
  40. package/src/engine/adapter-operation.ts +26 -0
  41. package/src/engine/cache-services.ts +114 -0
  42. package/src/engine/completion.ts +39 -0
  43. package/src/engine/generation-gateway.ts +358 -0
  44. package/src/engine/index.ts +59 -0
  45. package/src/engine/kv-budget.ts +120 -0
  46. package/src/engine/media-preparation.ts +232 -0
  47. package/src/engine/model-binding.ts +24 -0
  48. package/src/engine/model-host.ts +513 -0
  49. package/src/engine/model-serving.ts +89 -0
  50. package/src/{serve → engine}/preparation.ts +7 -5
  51. package/src/{serve → engine}/session-completion-engine.ts +6 -6
  52. package/src/engine/telemetry.ts +109 -0
  53. package/src/hub/downloads.ts +116 -0
  54. package/src/jobs/db.ts +16 -7
  55. package/src/jobs/events.ts +29 -0
  56. package/src/jobs/executable.ts +5 -0
  57. package/src/jobs/host.ts +51 -0
  58. package/src/jobs/in-process.ts +55 -0
  59. package/src/{contracts/jobs.ts → jobs/protocol.ts} +7 -9
  60. package/src/jobs/runner.ts +150 -231
  61. package/src/jobs/service.ts +75 -0
  62. package/src/jobs/sse.ts +59 -23
  63. package/src/jobs/worker-memory.ts +46 -0
  64. package/src/jobs/worker-process.ts +163 -0
  65. package/src/jobs/worker-supervisor.ts +258 -0
  66. package/src/modules.ts +58 -0
  67. package/src/publishing/catalog-hub.ts +18 -0
  68. package/src/publishing/credentials.ts +31 -0
  69. package/src/publishing/upload.ts +37 -0
  70. package/src/residency/leased-adapters.ts +21 -0
  71. package/src/residency/local-records.ts +16 -0
  72. package/src/residency/model-residency.ts +405 -0
  73. package/src/residency/record-index.ts +28 -0
  74. package/src/residency/resident-estimate.ts +50 -0
  75. package/src/{anthropic.ts → server/anthropic.ts} +3 -81
  76. package/src/server/cache-routes.ts +36 -0
  77. package/src/{serve → server}/chat-request.ts +10 -10
  78. package/src/{serve → server}/chat-stage.ts +14 -14
  79. package/src/{client.ts → server/client.ts} +28 -2
  80. package/src/{serve → server}/completion-executor.ts +33 -41
  81. package/src/server/decision-body.ts +55 -0
  82. package/src/server/decision-routes.ts +134 -0
  83. package/src/server/decision-wire.ts +228 -0
  84. package/src/{serve → server}/discovery-routes.ts +25 -82
  85. package/src/{serve → server}/http.ts +25 -13
  86. package/src/{serve → server}/inference-request.ts +5 -5
  87. package/src/server/job-routes.ts +20 -0
  88. package/src/server/media-prompt.ts +63 -0
  89. package/src/server/memory-completion-client.ts +46 -0
  90. package/src/server/model-routes.ts +192 -0
  91. package/src/{serve → server}/openai-wire.ts +3 -2
  92. package/src/{serve → server}/pipeline.ts +1 -1
  93. package/src/server/prompt-builder.ts +13 -0
  94. package/src/server/prompt-contracts.ts +17 -0
  95. package/src/server/proxy-routes.ts +187 -0
  96. package/src/server/publishing-routes.ts +41 -0
  97. package/src/{serve → server}/request-plan.ts +6 -8
  98. package/src/{serve → server}/request-prep.ts +47 -50
  99. package/src/{serve → server}/responses-client.ts +30 -13
  100. package/src/{responses.ts → server/responses.ts} +12 -93
  101. package/src/server/routes.ts +205 -0
  102. package/src/server/start.ts +72 -0
  103. package/src/server/status-routes.ts +108 -0
  104. package/src/{serve → server}/text-completion-stage.ts +25 -4
  105. package/src/server/text-prompt.ts +12 -0
  106. package/src/{serve → server}/token-streams.ts +36 -38
  107. package/src/server/worker-routes.ts +311 -0
  108. package/src/storage/paths.ts +61 -0
  109. package/src/web/assets.ts +37 -0
  110. package/src/web/{src → browser}/api.ts +5 -23
  111. package/src/web/browser/chat-host.ts +17 -0
  112. package/src/web/browser/installed-panels.ts +6 -0
  113. package/src/web/browser/main.ts +55 -0
  114. package/src/web/browser/markdown.ts +279 -0
  115. package/src/web/browser/memory-panel.ts +9 -0
  116. package/src/web/browser/palette.ts +64 -0
  117. package/src/web/browser/protocol.ts +13 -0
  118. package/src/web/browser/shell.ts +467 -0
  119. package/src/web/{src → browser}/status.ts +32 -36
  120. package/src/web/build.ts +51 -0
  121. package/src/web/public/app.html +1054 -0
  122. package/src/web/{sw.js → public/sw.js} +4 -3
  123. package/src/web/{vendor → public/vendor}/hljs-theme.css +2 -2
  124. package/src/web/{vendor → public/vendor}/hljs.js +2 -2
  125. package/THIRD_PARTY_LICENSES.md +0 -51
  126. package/docs/README.md +0 -107
  127. package/docs/archive/hlg-sampling.md +0 -350
  128. package/docs/archive/investigations/audio-input-plan.md +0 -356
  129. package/docs/archive/investigations/benchmark-ledger-through-v0.4.0.md +0 -3617
  130. package/docs/archive/investigations/chunk-finetune-experiment.md +0 -84
  131. package/docs/archive/investigations/colibri-glm52-port.md +0 -1486
  132. package/docs/archive/investigations/colibri-oracle-pin.md +0 -1284
  133. package/docs/archive/investigations/curve-bisector-routing.md +0 -123
  134. package/docs/archive/investigations/decode-roofline-lookagain.md +0 -348
  135. package/docs/archive/investigations/diffusion-gemma-port.md +0 -602
  136. package/docs/archive/investigations/dspark-handoff.md +0 -224
  137. package/docs/archive/investigations/expert-offload-single-user-moe.md +0 -292
  138. package/docs/archive/investigations/faithful-l1-consolidation.md +0 -310
  139. package/docs/archive/investigations/hlg-sampling-investigation.md +0 -970
  140. package/docs/archive/investigations/indexshare-performance-spike.md +0 -357
  141. package/docs/archive/investigations/kernel-perf-review-2026-07.md +0 -171
  142. package/docs/archive/investigations/lab-build-journal.md +0 -157
  143. package/docs/archive/investigations/minicpm5-decode-megakernel.md +0 -489
  144. package/docs/archive/investigations/openjarvis-fit.md +0 -198
  145. package/docs/archive/investigations/openwiki-evaluation.md +0 -249
  146. package/docs/archive/investigations/optimization_plan.md +0 -346
  147. package/docs/archive/investigations/orpo-base-uf-experiment-and-directions.md +0 -120
  148. package/docs/archive/investigations/orpo-flash-cce-pin-leak.md +0 -167
  149. package/docs/archive/investigations/orpo-uf-testing-handoff.md +0 -126
  150. package/docs/archive/investigations/pareto-specialized-runtime-findings.md +0 -92
  151. package/docs/archive/investigations/pi-builtin-investigation.md +0 -308
  152. package/docs/archive/investigations/pre-colibri-stabilization.md +0 -329
  153. package/docs/archive/investigations/qwen38-27b-campaign-log.md +0 -1092
  154. package/docs/archive/investigations/segmented-backward-handoff.md +0 -148
  155. package/docs/archive/investigations/starter-model-port-handoff.md +0 -342
  156. package/docs/archive/investigations/steel-flash-cce-handoff.md +0 -489
  157. package/docs/archive/investigations/the-dreaming-handoff.md +0 -141
  158. package/docs/archive/investigations/trainer-validation-experiment.md +0 -173
  159. package/docs/archive/planning/chat-ui-vision-memory-first.md +0 -199
  160. package/docs/archive/planning/chat-ui-vision.md +0 -50
  161. package/docs/archive/planning/chunk-task-roadmap.md +0 -77
  162. package/docs/archive/planning/curve-sampler-research-plan.md +0 -331
  163. package/docs/archive/planning/journal.md +0 -316
  164. package/docs/archive/planning/memory-docs-and-dag-plan.md +0 -194
  165. package/docs/archive/planning/pi-integration-review.md +0 -178
  166. package/docs/archive/planning/release-notes-v0.0.10.md +0 -57
  167. package/docs/archive/planning/release-notes-v0.0.11.md +0 -239
  168. package/docs/archive/planning/release-notes-v0.0.12.md +0 -80
  169. package/docs/archive/planning/release-notes-v0.0.13.md +0 -77
  170. package/docs/archive/planning/release-notes-v0.0.9.md +0 -122
  171. package/docs/archive/planning/release-notes-v0.2.0.md +0 -49
  172. package/docs/archive/planning/release-notes-v0.3.0.md +0 -35
  173. package/docs/archive/planning/release-notes-v0.4.0.md +0 -45
  174. package/docs/archive/planning/release-notes-v0.5.0.md +0 -37
  175. package/docs/archive/planning/web-ui-pass-plan.md +0 -213
  176. package/docs/archive/planning/website-readme-pass-plan.md +0 -224
  177. package/docs/design/batching.md +0 -2054
  178. package/docs/design/decode-speed-program.md +0 -2404
  179. package/docs/design/dreaming-nightly-pipeline.md +0 -684
  180. package/docs/design/generic-model-support.md +0 -821
  181. package/docs/design/kv-cache.md +0 -993
  182. package/docs/design/orpo-training.md +0 -680
  183. package/docs/design/speculative-decoding.md +0 -2129
  184. package/docs/design/turboquant.md +0 -2225
  185. package/docs/design/unified-engine-frontier-plan.md +0 -1987
  186. package/docs/design/web-chat-redesign.md +0 -695
  187. package/docs/planning/PRODUCT_ROADMAP.md +0 -593
  188. package/docs/planning/ResearchTopics.md +0 -119
  189. package/docs/planning/public-readiness-handoff.md +0 -244
  190. package/docs/planning/public-readiness.md +0 -149
  191. package/docs/planning/video-series.md +0 -421
  192. package/docs/reference/benchmarks.md +0 -3587
  193. package/docs/reference/cli.md +0 -646
  194. package/docs/reference/distribution.md +0 -380
  195. package/docs/reference/environment.md +0 -494
  196. package/docs/reference/glossary.md +0 -39
  197. package/docs/reference/library-api.md +0 -688
  198. package/docs/reference/memory.md +0 -260
  199. package/docs/reference/models.md +0 -346
  200. package/docs/reference/server-api.md +0 -1945
  201. package/docs/reference/server-config.md +0 -973
  202. package/docs/reference/training.md +0 -462
  203. package/docs/reference/troubleshooting.md +0 -158
  204. package/src/admission-plan.ts +0 -18
  205. package/src/audio/audiotoolbox.ts +0 -112
  206. package/src/audio/conformer.ts +0 -655
  207. package/src/audio/decode.ts +0 -133
  208. package/src/audio/features.ts +0 -202
  209. package/src/audio/mic-capture.ts +0 -98
  210. package/src/audio/silero-vad.ts +0 -379
  211. package/src/audio/transcode.ts +0 -67
  212. package/src/audio/whisper-decode.ts +0 -774
  213. package/src/audio/whisper-fast.ts +0 -732
  214. package/src/audio/whisper-format.ts +0 -73
  215. package/src/audio/whisper-mel.ts +0 -185
  216. package/src/audio/whisper-timing.ts +0 -274
  217. package/src/audio/whisper-tokenizer.ts +0 -239
  218. package/src/audio/whisper-transcribe.ts +0 -410
  219. package/src/backends/mlx/assistant-target.ts +0 -46
  220. package/src/backends/mlx/autoregressive-draft-rows.ts +0 -120
  221. package/src/backends/mlx/autoregressive.ts +0 -129
  222. package/src/backends/mlx/batch-group.ts +0 -1874
  223. package/src/backends/mlx/cache-attachment.ts +0 -67
  224. package/src/backends/mlx/cache-layout.ts +0 -31
  225. package/src/backends/mlx/checkpoint-state.ts +0 -53
  226. package/src/backends/mlx/continuation-persistence.ts +0 -107
  227. package/src/backends/mlx/continuation-request.ts +0 -43
  228. package/src/backends/mlx/continuation.ts +0 -77
  229. package/src/backends/mlx/diffusion.ts +0 -39
  230. package/src/backends/mlx/draft-target.ts +0 -32
  231. package/src/backends/mlx/encoder-cache.ts +0 -25
  232. package/src/backends/mlx/evaluation.ts +0 -105
  233. package/src/backends/mlx/fill-append.ts +0 -58
  234. package/src/backends/mlx/fill-group.ts +0 -497
  235. package/src/backends/mlx/full-kv-row-append.ts +0 -32
  236. package/src/backends/mlx/full-kv-row-donor.ts +0 -54
  237. package/src/backends/mlx/full-transitioning-kv-rows.ts +0 -65
  238. package/src/backends/mlx/gateway-binding.ts +0 -174
  239. package/src/backends/mlx/graph.ts +0 -38
  240. package/src/backends/mlx/kv-codec.ts +0 -9
  241. package/src/backends/mlx/kv-maintenance.ts +0 -161
  242. package/src/backends/mlx/kv-tensor-rows.ts +0 -212
  243. package/src/backends/mlx/memory-client.ts +0 -189
  244. package/src/backends/mlx/methods.ts +0 -90
  245. package/src/backends/mlx/mixed-iteration.ts +0 -62
  246. package/src/backends/mlx/model-prompt.ts +0 -178
  247. package/src/backends/mlx/model-serving.ts +0 -80
  248. package/src/backends/mlx/paged-cache-codec.ts +0 -27
  249. package/src/backends/mlx/prefill-cohort.ts +0 -100
  250. package/src/backends/mlx/prefill-policy.ts +0 -26
  251. package/src/backends/mlx/prefill-rows.ts +0 -244
  252. package/src/backends/mlx/prefill.ts +0 -33
  253. package/src/backends/mlx/projected-context-rows.ts +0 -72
  254. package/src/backends/mlx/prompt-input.ts +0 -29
  255. package/src/backends/mlx/qwen-prompt-input.ts +0 -40
  256. package/src/backends/mlx/request-policy.ts +0 -20
  257. package/src/backends/mlx/request-state-policy.ts +0 -54
  258. package/src/backends/mlx/rollback.ts +0 -92
  259. package/src/backends/mlx/row-sampling.ts +0 -49
  260. package/src/backends/mlx/serial-executor.ts +0 -251
  261. package/src/backends/mlx/speculative-group.ts +0 -438
  262. package/src/backends/mlx/speculative-round.ts +0 -161
  263. package/src/backends/mlx/speculative.ts +0 -131
  264. package/src/backends/mlx/state-rows.ts +0 -104
  265. package/src/backends/mlx/state-views.ts +0 -46
  266. package/src/backends/mlx/target-layout-capability.ts +0 -35
  267. package/src/backends/mlx/text-engine.ts +0 -73
  268. package/src/backends/mlx/transitioning-kv-rows.ts +0 -173
  269. package/src/chat-template.ts +0 -382
  270. package/src/cli.ts +0 -3490
  271. package/src/config.ts +0 -399
  272. package/src/contracts/completion.ts +0 -16
  273. package/src/contracts/execution.ts +0 -70
  274. package/src/contracts/generation.ts +0 -103
  275. package/src/contracts/host.ts +0 -7
  276. package/src/contracts/object-cache.ts +0 -8
  277. package/src/contracts/pi-provider.ts +0 -30
  278. package/src/contracts/pi.ts +0 -261
  279. package/src/contracts/prefix-cache.ts +0 -33
  280. package/src/contracts/resources.ts +0 -25
  281. package/src/contracts/sampling.ts +0 -23
  282. package/src/contracts/scheduling.ts +0 -54
  283. package/src/contracts/tsconfig.json +0 -14
  284. package/src/dataset/generators.ts +0 -1087
  285. package/src/dataset/index.ts +0 -12
  286. package/src/dataset/job.ts +0 -81
  287. package/src/dataset/llm.ts +0 -71
  288. package/src/dataset/registry.ts +0 -528
  289. package/src/diffusion/diffusion-generate.ts +0 -507
  290. package/src/download.ts +0 -505
  291. package/src/embed.ts +0 -77
  292. package/src/engine/admission.ts +0 -59
  293. package/src/engine/cancellation.ts +0 -38
  294. package/src/engine/engine.ts +0 -56
  295. package/src/engine/execution-coordinator.ts +0 -70
  296. package/src/engine/execution-plan.ts +0 -64
  297. package/src/engine/execution-tasks.ts +0 -44
  298. package/src/engine/namespaced-cache.ts +0 -13
  299. package/src/engine/resources.ts +0 -53
  300. package/src/engine/scheduler.ts +0 -92
  301. package/src/engine/session.ts +0 -247
  302. package/src/eval/README.md +0 -130
  303. package/src/eval/ifeval.ts +0 -81
  304. package/src/eval/kl-prompts.ts +0 -29
  305. package/src/eval/kl.ts +0 -590
  306. package/src/eval/perplexity.ts +0 -164
  307. package/src/eval/quality-db.ts +0 -92
  308. package/src/eval/reference-logits.ts +0 -207
  309. package/src/eval/runner.ts +0 -124
  310. package/src/eval/score.ts +0 -49
  311. package/src/eval/tasks/bfcl.ts +0 -535
  312. package/src/eval/tasks/gsm8k.ts +0 -98
  313. package/src/eval/tasks/hashhop.ts +0 -139
  314. package/src/eval/tasks/humaneval.ts +0 -275
  315. package/src/eval/tasks/ifeval.ts +0 -525
  316. package/src/eval/tasks/mmlu.ts +0 -157
  317. package/src/eval/ultrafeedback.ts +0 -60
  318. package/src/evaldb.ts +0 -199
  319. package/src/expert-io.ts +0 -393
  320. package/src/expert-offload-build.ts +0 -77
  321. package/src/expert-offload.ts +0 -76
  322. package/src/expert-residency.ts +0 -1134
  323. package/src/expert-usage.ts +0 -415
  324. package/src/fill/echo-index.ts +0 -188
  325. package/src/fill/fill-session.ts +0 -463
  326. package/src/fill/proposal.ts +0 -68
  327. package/src/fill/schema-rows.ts +0 -506
  328. package/src/fill/tool-boundary.ts +0 -103
  329. package/src/fit.ts +0 -249
  330. package/src/generate.ts +0 -1352
  331. package/src/grammar.ts +0 -604
  332. package/src/harness-pi.ts +0 -184
  333. package/src/hf-push.ts +0 -628
  334. package/src/hub-rest.ts +0 -242
  335. package/src/index.ts +0 -63
  336. package/src/inference/denoising.ts +0 -17
  337. package/src/inference/draft-acceptance.ts +0 -41
  338. package/src/inference/draft-output.ts +0 -49
  339. package/src/inference/graph.ts +0 -33
  340. package/src/inference/prefill.ts +0 -45
  341. package/src/inference/rollback.ts +0 -7
  342. package/src/jobs/index.ts +0 -31
  343. package/src/jobs/job-entry.ts +0 -107
  344. package/src/jobs/task-client.ts +0 -19
  345. package/src/jobs/types.ts +0 -2
  346. package/src/kv-scheme.ts +0 -279
  347. package/src/kv-store.ts +0 -1135
  348. package/src/lab/README.md +0 -22
  349. package/src/lab/curve/curve-designer.html +0 -303
  350. package/src/lab/curve/curve-sampler.ts +0 -122
  351. package/src/lab/expert-trace/expert-trace.ts +0 -69
  352. package/src/lab/paged-kv/paged-attention.ts +0 -117
  353. package/src/lab/paged-kv/paged-kv-rows.ts +0 -111
  354. package/src/lab/paged-kv/paged-kv.ts +0 -336
  355. package/src/library.ts +0 -40
  356. package/src/lora.ts +0 -408
  357. package/src/md-modules.d.ts +0 -54
  358. package/src/media-fetch.ts +0 -442
  359. package/src/memory/article.ts +0 -558
  360. package/src/memory/chunk-validation.ts +0 -150
  361. package/src/memory/chunk.ts +0 -318
  362. package/src/memory/cluster.ts +0 -351
  363. package/src/memory/coverage.ts +0 -134
  364. package/src/memory/crosslink.ts +0 -0
  365. package/src/memory/db.ts +0 -253
  366. package/src/memory/entity.ts +0 -203
  367. package/src/memory/gate.ts +0 -187
  368. package/src/memory/history.ts +0 -152
  369. package/src/memory/ingest.ts +0 -335
  370. package/src/memory/model.ts +0 -183
  371. package/src/memory/normalize.ts +0 -671
  372. package/src/memory/parse.ts +0 -22
  373. package/src/memory/pipeline.ts +0 -275
  374. package/src/memory/prompts.ts +0 -65
  375. package/src/memory/query.ts +0 -334
  376. package/src/memory/reconcile.ts +0 -734
  377. package/src/memory/redirect.ts +0 -108
  378. package/src/memory/reindex.ts +0 -325
  379. package/src/memory/resolve.ts +0 -307
  380. package/src/memory/rest.ts +0 -245
  381. package/src/memory/route.ts +0 -272
  382. package/src/memory/schedule.ts +0 -165
  383. package/src/memory/stages.ts +0 -777
  384. package/src/memory/synthesize.ts +0 -1638
  385. package/src/memory/tools.ts +0 -749
  386. package/src/memory/vault.ts +0 -1194
  387. package/src/memory/wikify.ts +0 -1064
  388. package/src/mlx/array.ts +0 -425
  389. package/src/mlx/autograd.ts +0 -304
  390. package/src/mlx/checkpoint.ts +0 -149
  391. package/src/mlx/compile.ts +0 -168
  392. package/src/mlx/custom-vjp.ts +0 -188
  393. package/src/mlx/ffi.ts +0 -531
  394. package/src/mlx/materialize.ts +0 -23
  395. package/src/mlx/metal-kernel.ts +0 -159
  396. package/src/mlx/normalized-argmax.ts +0 -116
  397. package/src/mlx/ops.ts +0 -1108
  398. package/src/mlx/quantized-rows.ts +0 -36
  399. package/src/mlx/token-bitmask.ts +0 -52
  400. package/src/mlx/turboquant-kv-decode.ts +0 -82
  401. package/src/mlx/turboquant-kv-inverse.ts +0 -122
  402. package/src/mlx/turboquant-ops.ts +0 -861
  403. package/src/mlx/turboquant-tables.ts +0 -237
  404. package/src/mmap.ts +0 -90
  405. package/src/model/artifact-identity.ts +0 -73
  406. package/src/model/batched-kv.ts +0 -235
  407. package/src/model/batched-mask.ts +0 -279
  408. package/src/model/batched-quant.ts +0 -212
  409. package/src/model/batched-quantized-kv.ts +0 -144
  410. package/src/model/batched-rotating-quant.ts +0 -353
  411. package/src/model/batched-rotating-state.ts +0 -149
  412. package/src/model/batched-rotating.ts +0 -338
  413. package/src/model/batched-row-storage.ts +0 -179
  414. package/src/model/batched-ssm.ts +0 -52
  415. package/src/model/batched-turboquant-kv.ts +0 -94
  416. package/src/model/compiled-decode.ts +0 -712
  417. package/src/model/decoded-kv-donor.ts +0 -12
  418. package/src/model/delayed-quantized-kv.ts +0 -72
  419. package/src/model/delayed-rotating-quantized-kv.ts +0 -93
  420. package/src/model/delayed-turboquant-kv.ts +0 -57
  421. package/src/model/diffusion-gemma.ts +0 -1026
  422. package/src/model/factory.ts +0 -215
  423. package/src/model/fingerprint.ts +0 -45
  424. package/src/model/flash-attention.ts +0 -694
  425. package/src/model/full-prefill-padding.ts +0 -56
  426. package/src/model/full-prefill-row.ts +0 -105
  427. package/src/model/gemma4-base.ts +0 -2370
  428. package/src/model/gemma4.ts +0 -1395
  429. package/src/model/generated/gemma4-12b.ts +0 -429
  430. package/src/model/generated/gemma4-26b.ts +0 -448
  431. package/src/model/generated/gemma4-e4b.ts +0 -717
  432. package/src/model/generated/index.ts +0 -18
  433. package/src/model/glm52-atlas.ts +0 -511
  434. package/src/model/glm52-cache.ts +0 -792
  435. package/src/model/glm52-config.ts +0 -262
  436. package/src/model/glm52-container.ts +0 -305
  437. package/src/model/glm52-coupling.ts +0 -388
  438. package/src/model/glm52-dsa.ts +0 -606
  439. package/src/model/glm52-expert-layout.ts +0 -253
  440. package/src/model/glm52-layout.ts +0 -152
  441. package/src/model/glm52-memory.ts +0 -500
  442. package/src/model/glm52-mla.ts +0 -792
  443. package/src/model/glm52-moe.ts +0 -223
  444. package/src/model/glm52-pilot.ts +0 -537
  445. package/src/model/glm52-reference.ts +0 -445
  446. package/src/model/glm52-residency.ts +0 -557
  447. package/src/model/glm52-resident-weights.ts +0 -135
  448. package/src/model/glm52-streamed-experts.ts +0 -453
  449. package/src/model/glm52-streamed-metal.ts +0 -610
  450. package/src/model/glm52-weights.ts +0 -503
  451. package/src/model/glm52.ts +0 -1132
  452. package/src/model/implementation.ts +0 -69
  453. package/src/model/kv-attention-view.ts +0 -70
  454. package/src/model/minicpm5.ts +0 -376
  455. package/src/model/profile.ts +0 -468
  456. package/src/model/quantized-append-attention.ts +0 -31
  457. package/src/model/qwen-append.ts +0 -12
  458. package/src/model/qwen-conv.ts +0 -52
  459. package/src/model/qwen3-delta-state.ts +0 -74
  460. package/src/model/qwen3-delta.ts +0 -451
  461. package/src/model/qwen3-moe.ts +0 -404
  462. package/src/model/qwen3-mrope.ts +0 -250
  463. package/src/model/qwen3.ts +0 -277
  464. package/src/model/qwen3_5-checkpoint.ts +0 -164
  465. package/src/model/qwen3_5.ts +0 -881
  466. package/src/model/rotating-kv-donor.ts +0 -35
  467. package/src/model/rotating-kv-layout.ts +0 -262
  468. package/src/model/rotating-row-transaction.ts +0 -35
  469. package/src/model/speculative-rotating-kv.ts +0 -101
  470. package/src/model/ssm-prefill-padding.ts +0 -39
  471. package/src/model/support.ts +0 -110
  472. package/src/model/token-groups.ts +0 -66
  473. package/src/model/trellis-balanced-scatter.ts +0 -51
  474. package/src/model/trellis-linear.ts +0 -672
  475. package/src/model/trellis-shared-m.ts +0 -151
  476. package/src/model/trellis-shared-scatter.ts +0 -144
  477. package/src/model/trellis-splitk-prefill.ts +0 -158
  478. package/src/model/trellis-tiled-prefill.ts +0 -136
  479. package/src/model/trellis-vector-expand.ts +0 -62
  480. package/src/model/trellis-wide-prefill.ts +0 -94
  481. package/src/model/turboquant-codec.ts +0 -89
  482. package/src/model/universal/archs.ts +0 -332
  483. package/src/model/universal/dense.ts +0 -612
  484. package/src/model/universal/modules.ts +0 -170
  485. package/src/model/universal/rope.ts +0 -381
  486. package/src/model/whisper.ts +0 -435
  487. package/src/native/expert_io.c +0 -523
  488. package/src/native/frame_extract.swift +0 -140
  489. package/src/native/mic_capture.swift +0 -92
  490. package/src/native-pack.ts +0 -163
  491. package/src/pi-provider.ts +0 -53
  492. package/src/pi-session.ts +0 -81
  493. package/src/pi-terminal.ts +0 -274
  494. package/src/pi-web.ts +0 -1940
  495. package/src/preflight.ts +0 -158
  496. package/src/prompt-cache.ts +0 -682
  497. package/src/quantize/allocator.ts +0 -627
  498. package/src/quantize/atomic-output.ts +0 -38
  499. package/src/quantize/calibration.ts +0 -181
  500. package/src/quantize/config-writer.ts +0 -145
  501. package/src/quantize/index.ts +0 -84
  502. package/src/quantize/job.ts +0 -162
  503. package/src/quantize/quantizer.ts +0 -666
  504. package/src/quantize/rotate.ts +0 -603
  505. package/src/quantize/safetensors-writer.ts +0 -249
  506. package/src/quantize/sensitivity.ts +0 -216
  507. package/src/quantize/trellis.ts +0 -586
  508. package/src/quantize/weight-transform.ts +0 -292
  509. package/src/registry.ts +0 -602
  510. package/src/runtime-config.ts +0 -77
  511. package/src/safetensors.ts +0 -165
  512. package/src/sampler.ts +0 -1012
  513. package/src/serve/admin-routes.ts +0 -184
  514. package/src/serve/assets/curve-terrain.html +0 -105
  515. package/src/serve/assets/training-inference-map.html +0 -575
  516. package/src/serve/audio-routes.ts +0 -323
  517. package/src/serve/aux-routes.ts +0 -132
  518. package/src/serve/batch-scheduler.ts +0 -8
  519. package/src/serve/checkpoint-identity.ts +0 -64
  520. package/src/serve/generation-gateway.ts +0 -537
  521. package/src/serve/isolate.ts +0 -609
  522. package/src/serve/kv-budget.ts +0 -14
  523. package/src/serve/lab-routes.ts +0 -341
  524. package/src/serve/lane-registry.ts +0 -45
  525. package/src/serve/model-admin-routes.ts +0 -155
  526. package/src/serve/model-binding.ts +0 -49
  527. package/src/serve/model-host.ts +0 -511
  528. package/src/serve/parent-application.ts +0 -56
  529. package/src/serve/prompt-response-trace.ts +0 -161
  530. package/src/serve/response-lease.ts +0 -24
  531. package/src/serve/session-search.ts +0 -326
  532. package/src/serve/static-routes.ts +0 -101
  533. package/src/serve/transcription-server.ts +0 -64
  534. package/src/serve/transcription-service.ts +0 -502
  535. package/src/server.ts +0 -1282
  536. package/src/spec/assistant-rows.ts +0 -93
  537. package/src/spec/assistant-source.ts +0 -126
  538. package/src/spec/cache-identity.ts +0 -22
  539. package/src/spec/deepspec-rows.ts +0 -29
  540. package/src/spec/deepspec-source.ts +0 -145
  541. package/src/spec/dflash-source.ts +0 -149
  542. package/src/spec/drafter.ts +0 -304
  543. package/src/spec/dspark/ab-stats.ts +0 -136
  544. package/src/spec/dspark/calibration.ts +0 -112
  545. package/src/spec/dspark/data-dflash.ts +0 -156
  546. package/src/spec/dspark/data.ts +0 -201
  547. package/src/spec/dspark/deepspec-module.ts +0 -776
  548. package/src/spec/dspark/generate-dflash.ts +0 -245
  549. package/src/spec/dspark/generate.ts +0 -290
  550. package/src/spec/dspark/loader.ts +0 -23
  551. package/src/spec/dspark/loss.ts +0 -213
  552. package/src/spec/dspark/module-dflash.ts +0 -560
  553. package/src/spec/dspark/module.ts +0 -540
  554. package/src/spec/dspark/quantize-drafter.ts +0 -91
  555. package/src/spec/dspark/sample.ts +0 -112
  556. package/src/spec/generate.ts +0 -276
  557. package/src/spec/glm52-mtp-graph.ts +0 -29
  558. package/src/spec/glm52-mtp-rows.ts +0 -132
  559. package/src/spec/glm52-mtp-source.ts +0 -318
  560. package/src/spec/glm52-mtp-state.ts +0 -33
  561. package/src/spec/ngram-source.ts +0 -226
  562. package/src/spec/projected-draft-rows.ts +0 -73
  563. package/src/spec/qwen-mtp-module.ts +0 -171
  564. package/src/spec/qwen-mtp-rows.ts +0 -230
  565. package/src/spec/qwen-mtp-source.ts +0 -386
  566. package/src/spec/qwen-mtp-state.ts +0 -30
  567. package/src/spec/serve-loop.ts +0 -587
  568. package/src/spec/source.ts +0 -228
  569. package/src/spec/two-model-rows.ts +0 -62
  570. package/src/spec/two-model.ts +0 -113
  571. package/src/ssd-cache.ts +0 -431
  572. package/src/ssd-durability.ts +0 -209
  573. package/src/storage/host-buffer.ts +0 -24
  574. package/src/storage/kv-writer.ts +0 -83
  575. package/src/storage/kv-writer.worker.js +0 -264
  576. package/src/storage/retention-policy.ts +0 -37
  577. package/src/tiered-prompt-cache.ts +0 -45
  578. package/src/tokenizer.ts +0 -335
  579. package/src/tool-approvals.ts +0 -131
  580. package/src/tool-call.ts +0 -695
  581. package/src/train/dataset.ts +0 -484
  582. package/src/train/diffusion-lora.ts +0 -216
  583. package/src/train/export.ts +0 -32
  584. package/src/train/flash-cce.ts +0 -1234
  585. package/src/train/forward.ts +0 -266
  586. package/src/train/fuse.ts +0 -243
  587. package/src/train/index.ts +0 -46
  588. package/src/train/job.ts +0 -205
  589. package/src/train/lora-params.ts +0 -254
  590. package/src/train/loss.ts +0 -1591
  591. package/src/train/merge.ts +0 -195
  592. package/src/train/optimizer.ts +0 -183
  593. package/src/train/prefix-shared.ts +0 -490
  594. package/src/train/rank.ts +0 -137
  595. package/src/train/segmented.ts +0 -1828
  596. package/src/train/steel-qmm-header.ts +0 -1461
  597. package/src/train/trainer.ts +0 -1221
  598. package/src/train/watch.ts +0 -542
  599. package/src/vision/diffusion-vision.ts +0 -620
  600. package/src/vision/embedder.ts +0 -176
  601. package/src/vision/preprocess.ts +0 -247
  602. package/src/vision/prompt.ts +0 -409
  603. package/src/vision/qwen3vl-preprocess.ts +0 -409
  604. package/src/vision/qwen3vl-prompt.ts +0 -233
  605. package/src/vision/qwen3vl-tower.ts +0 -566
  606. package/src/vision/siglip.ts +0 -607
  607. package/src/vision/video-frames.ts +0 -148
  608. package/src/web/app.html +0 -2155
  609. package/src/web/app.js +0 -5863
  610. package/src/web/skills/memory/SKILL.md +0 -120
  611. package/src/web/skills/web-research/SKILL.md +0 -46
  612. package/src/web/skills.ts +0 -97
  613. package/src/web/src/adapters-panel.ts +0 -288
  614. package/src/web/src/assistant.ts +0 -451
  615. package/src/web/src/chat.ts +0 -1060
  616. package/src/web/src/composer.ts +0 -962
  617. package/src/web/src/dataset.ts +0 -132
  618. package/src/web/src/finetune.ts +0 -267
  619. package/src/web/src/hub.ts +0 -428
  620. package/src/web/src/main.ts +0 -98
  621. package/src/web/src/markdown.ts +0 -0
  622. package/src/web/src/memory-panel.ts +0 -572
  623. package/src/web/src/model-picker.ts +0 -159
  624. package/src/web/src/palette.ts +0 -346
  625. package/src/web/src/protocol.ts +0 -26
  626. package/src/web/src/quantize.ts +0 -167
  627. package/src/web/src/rag.ts +0 -274
  628. package/src/web/src/sessions.ts +0 -469
  629. package/src/web/src/shell.ts +0 -804
  630. package/src/web/src/ui-catalog.ts +0 -78
  631. package/src/web/src/voice.ts +0 -141
  632. package/src/web/vendor/README.md +0 -61
  633. package/src/web-assets.ts +0 -47
  634. package/src/web-tools.ts +0 -478
  635. package/src/weights.ts +0 -255
  636. /package/src/{serve → server}/completion-sink.ts +0 -0
  637. /package/src/{serve → server}/generated-token-history.ts +0 -0
  638. /package/src/web/{icon.svg → public/icon.svg} +0 -0
  639. /package/src/web/{manifest.webmanifest → public/manifest.webmanifest} +0 -0
  640. /package/src/web/{vendor → public/vendor}/hljs-LICENSE +0 -0
@@ -1,428 +0,0 @@
1
- // GENERATED-ADJACENT source module — part of the src/web/src/* split (plan
2
- // §7/§9 Phase 3). Built into src/web/app.js by scripts/build-web.ts.
3
- //
4
- // Model Hub panel (docs/design/web-chat-redesign.md §9 Phase 3, beat-matrix
5
- // Axis 3 "Hub" row). Same overlay language/chrome as the Memory panel
6
- // (mem-overlay/mem-panel/mem-head/mem-body): fixed right-side panel,
7
- // scrimmed, focus-trapped, Escape closes, opened from the model picker
8
- // popover's "Browse models…" action (and exported for palette use later).
9
- //
10
- // Three sections talking to src/hub-rest.ts's REST wrappers — no new WS
11
- // frames:
12
- // - Downloaded <- GET /api/hub/local (size/quant/capability chips, the
13
- // /fit verdict as a green/yellow/red dot + predicted tok/s, a "serve"
14
- // action).
15
- // - Search Hugging Face <- debounced GET /api/hub/search (id/downloads/
16
- // size, a Download action -> POST /api/hub/download, then live progress
17
- // via polling GET /downloads while the panel stays open).
18
- // - A quiet offline footer note when search/local fail to reach the
19
- // network (never conflated with "no results").
20
- //
21
- // NEVER auto-downloads, NEVER auto-serves — every action is an explicit
22
- // click. esc() every interpolation site: repo ids/model types are
23
- // server-reported but ultimately HF-namespace strings, same hazard class
24
- // model-picker.ts already treats as untrusted.
25
-
26
- import { $, toast, trapFocus, setHubPanelClose, setOpenHubFromModelPicker, type FocusTrap } from "./shell";
27
- import { api } from "./api";
28
- import type { ApiEnvelope } from "./protocol";
29
- import { esc } from "./markdown";
30
-
31
- /* ────────────────────────────────────────────────────────────────────
32
- REST response shapes (mirrors src/hub-rest.ts's jsonOk() bodies).
33
- ──────────────────────────────────────────────────────────────────── */
34
-
35
- export interface HubLocalRow {
36
- repo_id: string;
37
- model_type: string;
38
- size_bytes: number;
39
- quant_bits: number | null;
40
- quant_group_size: number | null;
41
- vision: boolean;
42
- supported: boolean;
43
- support_tier: "targeted" | "generic" | null;
44
- assessment: { fits: boolean; max_safe_context: number; predicted_decode_tps: number } | null;
45
- }
46
- type LocalResp = ApiEnvelope & { models?: HubLocalRow[] };
47
-
48
- export interface HubSearchRow {
49
- id: string;
50
- downloads: number;
51
- likes: number;
52
- size_estimate: number | null;
53
- }
54
- type SearchResp = ApiEnvelope & { offline?: boolean; results?: HubSearchRow[] };
55
-
56
- interface DownloadInfo {
57
- repoId: string;
58
- state: "active" | "done" | "error";
59
- currentFile: string | null;
60
- receivedBytes: number;
61
- totalBytes: number;
62
- bytesPerSec?: number;
63
- error?: string;
64
- }
65
- type DownloadsResp = { downloads?: DownloadInfo[] };
66
-
67
- /* ────────────────────────────────────────────────────────────────────
68
- Shared formatting (local copies — hub.ts has no import from
69
- model-picker.ts to avoid a cross-popover coupling neither side needs).
70
- ──────────────────────────────────────────────────────────────────── */
71
-
72
- function gb(n: number): string { return (n / 2 ** 30).toFixed(1) + " GB"; }
73
-
74
- /** Same three-tier convention as model-picker.ts's fitVerdict: green (fits
75
- * with headroom, >=10 tok/s), yellow (fits but slow), red (doesn't fit),
76
- * null (no assessment). Kept as a local copy rather than a shared import
77
- * since the two modules' row shapes differ slightly (Hub's assessment is
78
- * non-optional-null but otherwise identical) — duplicating one small pure
79
- * function is cheaper than adding a cross-import for four lines of logic. */
80
- function fitVerdict(a: HubLocalRow["assessment"]): "green" | "yellow" | "red" | null {
81
- if (!a) return null;
82
- if (!a.fits) return "red";
83
- return a.predicted_decode_tps >= 10 ? "green" : "yellow";
84
- }
85
-
86
- function fitDotHtml(verdict: "green" | "yellow" | "red" | null): string {
87
- return verdict
88
- ? '<span class="hub-fit-dot ' + verdict + '" aria-hidden="true"></span>'
89
- : '<span class="hub-fit-dot" aria-hidden="true" style="background:var(--dimmer)"></span>';
90
- }
91
-
92
- /* ────────────────────────────────────────────────────────────────────
93
- Downloaded section
94
- ──────────────────────────────────────────────────────────────────── */
95
-
96
- function localRowHtml(m: HubLocalRow): string {
97
- const verdict = fitVerdict(m.assessment);
98
- const name = m.repo_id.split("/").pop() || m.repo_id;
99
- const quant = m.quant_bits ? m.quant_bits + "-bit" : "unquantized";
100
- const metaBits = [gb(m.size_bytes), quant];
101
- if (m.vision) metaBits.push("vision");
102
- if (!m.supported) metaBits.push("unsupported model family");
103
- const tpsBit = m.assessment && m.assessment.fits
104
- ? m.assessment.predicted_decode_tps.toFixed(0) + " tok/s predicted"
105
- : m.assessment ? "doesn't fit this Mac's memory" : "fit unknown";
106
- const serveBtn = m.supported
107
- ? '<button type="button" class="hub-serve-btn" data-repo="' + esc(m.repo_id) + '">Serve</button>'
108
- : "";
109
- return (
110
- '<div class="hub-row">' +
111
- fitDotHtml(verdict) +
112
- '<div class="hub-row-main">' +
113
- '<div class="hub-row-name" title="' + esc(m.repo_id) + '">' + esc(name) + "</div>" +
114
- '<div class="hub-row-meta">' + esc(metaBits.join(" · ")) + " · " + esc(tpsBit) + "</div>" +
115
- "</div>" +
116
- '<div class="hub-row-actions">' + serveBtn + "</div>" +
117
- "</div>"
118
- );
119
- }
120
-
121
- /** Pure render, exported for esc()-discipline + empty-state tests
122
- * (tests/using/web-app.test.ts, mirroring renderModelPopBodyHtml's pattern). */
123
- export function renderHubLocalHtml(models: HubLocalRow[]): string {
124
- if (!models.length) {
125
- return '<div class="hub-empty">No models downloaded yet — search Hugging Face below to get one.</div>';
126
- }
127
- return [...models]
128
- .sort((a, b) => a.repo_id.localeCompare(b.repo_id))
129
- .map(localRowHtml)
130
- .join("");
131
- }
132
-
133
- async function loadLocal(): Promise<void> {
134
- const body = $("hub-local-body");
135
- if (!body) return;
136
- body.innerHTML = '<div class="hub-empty"><span class="shimmer">loading…</span></div>';
137
- try {
138
- const d = await api<LocalResp>("/api/hub/local");
139
- body.innerHTML = renderHubLocalHtml(d.models || []);
140
- wireServeButtons(body);
141
- } catch {
142
- body.innerHTML = '<div class="hub-empty">Could not reach the server.</div>';
143
- }
144
- }
145
-
146
- /* ────────────────────────────────────────────────────────────────────
147
- Serve action — handles BOTH the live-swap outcome (if a future server
148
- ever returns ok:true) and today's honest restart_required outcome,
149
- rendered as a copy-able command. NEVER auto-serves — this is only
150
- reachable from an explicit button click.
151
- ──────────────────────────────────────────────────────────────────── */
152
-
153
- function wireServeButtons(container: HTMLElement): void {
154
- container.querySelectorAll<HTMLButtonElement>(".hub-serve-btn").forEach((btn) => {
155
- btn.onclick = () => serveModel(btn.dataset.repo || "", btn);
156
- });
157
- }
158
-
159
- type ServeResp = ApiEnvelope & { restart_required?: boolean; command?: string };
160
-
161
- async function serveModel(repo: string, btn: HTMLButtonElement): Promise<void> {
162
- if (!repo) return;
163
- btn.disabled = true;
164
- const prevText = btn.textContent;
165
- btn.textContent = "Checking…";
166
- const d = await api<ServeResp>("/api/hub/serve", { method: "POST", body: { model: repo } })
167
- .catch((): ServeResp => ({ ok: false, error: "request failed" }));
168
- btn.disabled = false;
169
- btn.textContent = prevText;
170
- if (d.ok) {
171
- // Honest today-doesn't-happen branch, kept real rather than dead code:
172
- // if the server ever reports a completed live swap, reflect it plainly.
173
- toast("Now serving " + repo, "ok");
174
- return;
175
- }
176
- if (d.restart_required && d.command) {
177
- showRestartCommand(repo, d.command);
178
- return;
179
- }
180
- toast("Couldn't switch models: " + (d.error || "unknown error"), "err");
181
- }
182
-
183
- /** Renders the restart command inline in the footer strip (not a toast —
184
- * it needs to stay visible long enough to copy) with an honest note:
185
- * restarting takes ~seconds and sessions are preserved on disk (chat
186
- * history lives in local JSON regardless of which model process serves
187
- * it — restarting the model doesn't touch it). */
188
- function showRestartCommand(repo: string, command: string): void {
189
- const strip = $("hub-restart-strip");
190
- if (!strip) return;
191
- strip.innerHTML =
192
- '<div class="hub-restart-note">Switching to <strong>' + esc(repo) + '</strong> needs a restart — ' +
193
- "there's no live in-process model swap on this path yet. Restarting takes about a second " +
194
- "and your sessions are preserved on disk, exactly as they are now.</div>" +
195
- '<div class="hub-cmd"><code>' + esc(command) + '</code>' +
196
- '<button type="button" class="hub-copy-btn" data-cmd="' + esc(command) + '">Copy</button></div>';
197
- strip.classList.add("show");
198
- const copyBtn = strip.querySelector(".hub-copy-btn") as HTMLButtonElement | null;
199
- if (copyBtn) {
200
- copyBtn.onclick = () => {
201
- const cmd = copyBtn.dataset.cmd || "";
202
- if (navigator.clipboard) {
203
- navigator.clipboard.writeText(cmd).then(() => {
204
- const prev = copyBtn.textContent; copyBtn.textContent = "Copied";
205
- setTimeout(() => { copyBtn.textContent = prev; }, 1200);
206
- }).catch(() => {});
207
- }
208
- };
209
- }
210
- }
211
-
212
- function hideRestartStrip(): void {
213
- const strip = $("hub-restart-strip");
214
- if (strip) { strip.classList.remove("show"); strip.innerHTML = ""; }
215
- }
216
-
217
- /* ────────────────────────────────────────────────────────────────────
218
- Search Hugging Face section — debounced, offline-aware.
219
- ──────────────────────────────────────────────────────────────────── */
220
-
221
- function searchRowHtml(r: HubSearchRow, downloading: boolean): string {
222
- const name = r.id.split("/").pop() || r.id;
223
- const metaBits = [r.downloads.toLocaleString() + " downloads"];
224
- if (r.size_estimate != null) metaBits.push(gb(r.size_estimate));
225
- const action = downloading
226
- ? '<span class="hub-dl-tag">downloading…</span>'
227
- : '<button type="button" class="hub-download-btn" data-repo="' + esc(r.id) + '">Download</button>';
228
- return (
229
- '<div class="hub-row" data-search-repo="' + esc(r.id) + '">' +
230
- '<div class="hub-row-main">' +
231
- '<div class="hub-row-name" title="' + esc(r.id) + '">' + esc(name) + "</div>" +
232
- '<div class="hub-row-meta">' + esc(r.id) + " · " + esc(metaBits.join(" · ")) + "</div>" +
233
- "</div>" +
234
- '<div class="hub-row-actions">' + action + "</div>" +
235
- "</div>"
236
- );
237
- }
238
-
239
- /** Pure render, exported for tests. `downloading` is the set of repo ids
240
- * with an active download right now (from GET /downloads polling). */
241
- export function renderHubSearchHtml(results: HubSearchRow[], offline: boolean, downloading: Set<string>): string {
242
- if (offline) {
243
- return '<div class="hub-empty">Can\'t reach Hugging Face right now — search needs a network connection.</div>';
244
- }
245
- if (!results.length) {
246
- return '<div class="hub-empty">No matches.</div>';
247
- }
248
- return results.map((r) => searchRowHtml(r, downloading.has(r.id))).join("");
249
- }
250
-
251
- let searchDebounce: ReturnType<typeof setTimeout> | undefined;
252
- let lastSearchQuery = "";
253
- let inFlightDownloads = new Set<string>();
254
-
255
- async function runSearch(query: string): Promise<void> {
256
- const body = $("hub-search-body");
257
- if (!body) return;
258
- lastSearchQuery = query;
259
- if (!query) { body.innerHTML = '<div class="hub-empty">Type to search Hugging Face for MLX models.</div>'; return; }
260
- body.innerHTML = '<div class="hub-empty"><span class="shimmer">searching…</span></div>';
261
- const d = await api<SearchResp>("/api/hub/search?q=" + encodeURIComponent(query))
262
- .catch((): SearchResp => ({ ok: false, offline: true, results: [] }));
263
- if (lastSearchQuery !== query) return; // a newer keystroke already superseded this response
264
- const offline = !!d.offline;
265
- body.innerHTML = renderHubSearchHtml(d.results || [], offline, inFlightDownloads);
266
- wireDownloadButtons(body);
267
- setOfflineFooter(offline);
268
- }
269
-
270
- function wireDownloadButtons(container: HTMLElement): void {
271
- container.querySelectorAll<HTMLButtonElement>(".hub-download-btn").forEach((btn) => {
272
- btn.onclick = () => startDownload(btn.dataset.repo || "", btn);
273
- });
274
- }
275
-
276
- type DownloadStartResp = ApiEnvelope & { repo?: string; started?: boolean };
277
-
278
- async function startDownload(repo: string, btn: HTMLButtonElement): Promise<void> {
279
- if (!repo) return;
280
- btn.disabled = true;
281
- btn.textContent = "Starting…";
282
- const d = await api<DownloadStartResp>("/api/hub/download", { method: "POST", body: { repo } })
283
- .catch((): DownloadStartResp => ({ ok: false, error: "request failed" }));
284
- if (!d.ok) {
285
- btn.disabled = false;
286
- btn.textContent = "Download";
287
- toast("Couldn't start download: " + (d.error || "unknown error"), "err");
288
- return;
289
- }
290
- inFlightDownloads.add(repo);
291
- toast("Downloading " + repo, "ok");
292
- const row = btn.closest(".hub-row");
293
- const actions = row ? row.querySelector(".hub-row-actions") : null;
294
- if (actions) actions.innerHTML = '<span class="hub-dl-tag">downloading…</span>';
295
- ensureDownloadPolling();
296
- }
297
-
298
- /* ────────────────────────────────────────────────────────────────────
299
- Live download progress — reuses the existing GET /downloads endpoint
300
- (same source the nav download pill polls) so the Hub panel's own
301
- in-flight rows show real progress without a second tracker. Only polls
302
- while the panel is open (started on open, stopped on close).
303
- ──────────────────────────────────────────────────────────────────── */
304
-
305
- let downloadPollTimer: ReturnType<typeof setInterval> | undefined;
306
-
307
- function ensureDownloadPolling(): void {
308
- if (downloadPollTimer) return;
309
- downloadPollTimer = setInterval(pollDownloads, 1500);
310
- }
311
-
312
- function stopDownloadPolling(): void {
313
- if (downloadPollTimer) { clearInterval(downloadPollTimer); downloadPollTimer = undefined; }
314
- }
315
-
316
- async function pollDownloads(): Promise<void> {
317
- let downloads: DownloadInfo[] = [];
318
- try {
319
- const r = await fetch("/downloads");
320
- const d = (await r.json()) as DownloadsResp;
321
- downloads = d.downloads || [];
322
- } catch { return; }
323
-
324
- let anyActive = false;
325
- for (const dl of downloads) {
326
- if (dl.state === "active") { anyActive = true; inFlightDownloads.add(dl.repoId); }
327
- const row = document.querySelector<HTMLElement>('.hub-row[data-search-repo="' + cssEscape(dl.repoId) + '"]');
328
- if (!row) continue;
329
- const actions = row.querySelector(".hub-row-actions");
330
- if (!actions) continue;
331
- if (dl.state === "active") {
332
- const pct = dl.totalBytes ? Math.floor((dl.receivedBytes / dl.totalBytes) * 100) : 0;
333
- actions.innerHTML = '<span class="hub-dl-tag">' + pct + "%</span>";
334
- } else if (dl.state === "done") {
335
- inFlightDownloads.delete(dl.repoId);
336
- actions.innerHTML = '<span class="hub-dl-tag done">done — reload to serve</span>';
337
- } else if (dl.state === "error") {
338
- inFlightDownloads.delete(dl.repoId);
339
- actions.innerHTML = '<span class="hub-dl-tag error">' + esc(dl.error || "download failed") + "</span>";
340
- }
341
- }
342
- // A download that finished/errored and dropped out of the tracker's
343
- // rolling window entirely still needs its stale "downloading…" tag
344
- // cleared — treat "not present anymore" the same as "not active".
345
- for (const repo of [...inFlightDownloads]) {
346
- if (!downloads.some((d) => d.repoId === repo && d.state === "active")) inFlightDownloads.delete(repo);
347
- }
348
- if (!anyActive) stopDownloadPolling();
349
- }
350
-
351
- /** Minimal CSS.escape polyfill-free helper for attribute-selector lookups —
352
- * repo ids are `org/name` (slash + alnum/hyphen/underscore/dot), so a full
353
- * CSS.escape isn't needed; this only guards the one character (`"`) that
354
- * would break the quoted attribute selector above. HF repo ids can't
355
- * contain quotes, but this stays defensive rather than assuming it. */
356
- function cssEscape(s: string): string {
357
- return s.replace(/"/g, '\\"');
358
- }
359
-
360
- /* ────────────────────────────────────────────────────────────────────
361
- Panel chrome: open/close, focus trap, section wiring.
362
- ──────────────────────────────────────────────────────────────────── */
363
-
364
- let panelTrap: FocusTrap;
365
-
366
- export function isHubPanelOpen(): boolean {
367
- const ov = $("hub-overlay");
368
- return !!ov && ov.classList.contains("open");
369
- }
370
-
371
- export function closeHubPanel(): void {
372
- $("hub-overlay").classList.remove("open");
373
- stopDownloadPolling();
374
- panelTrap.restore();
375
- }
376
-
377
- export async function openHubPanel(): Promise<void> {
378
- panelTrap.capture();
379
- $("hub-overlay").classList.add("open");
380
- hideRestartStrip();
381
- const search = $("hub-search-input") as HTMLInputElement | null;
382
- if (search) search.value = "";
383
- const searchBody = $("hub-search-body");
384
- if (searchBody) searchBody.innerHTML = '<div class="hub-empty">Type to search Hugging Face for MLX models.</div>';
385
- setOfflineFooter(false);
386
- await loadLocal();
387
- if (inFlightDownloads.size) ensureDownloadPolling();
388
- setTimeout(() => { const el2 = $("hub-search-input") as HTMLInputElement | null; if (el2) el2.focus(); }, 30);
389
- }
390
-
391
- function setOfflineFooter(offline: boolean): void {
392
- const note = $("hub-offline-note");
393
- if (!note) return;
394
- note.style.display = offline ? "" : "none";
395
- }
396
-
397
- /** Wires the panel chrome + registers the open-from-model-picker callback.
398
- * Call once at chat controller init, same lifecycle as initMemoryPanel().
399
- * The model picker's "Browse models…" button is re-rendered via innerHTML
400
- * on every popover open (model-picker.ts's refreshModelPop()), so it can't
401
- * hold a handler bound once here — model-picker.ts calls
402
- * openHubFromModelPicker() (shell.ts's registered-callback pattern)
403
- * instead of a direct element lookup. */
404
- export function initHubPanel(): void {
405
- panelTrap = trapFocus($("hub-overlay"), isHubPanelOpen);
406
- setHubPanelClose(closeHubPanel); // registers with shell.ts's global Escape sweep
407
- setOpenHubFromModelPicker(() => { openHubPanel(); });
408
-
409
- const closeBtn = $("hub-close") as HTMLButtonElement | null;
410
- if (closeBtn) closeBtn.onclick = closeHubPanel;
411
- const overlay = $("hub-overlay");
412
- if (overlay) overlay.addEventListener("click", (e) => { if (e.target === overlay) closeHubPanel(); });
413
-
414
- const search = $("hub-search-input") as HTMLInputElement | null;
415
- if (search) {
416
- search.addEventListener("input", () => {
417
- clearTimeout(searchDebounce);
418
- const q = search.value.trim();
419
- searchDebounce = setTimeout(() => runSearch(q), 300);
420
- });
421
- }
422
- }
423
-
424
- // Exposed for tests that need to reset module-level state between cases.
425
- export function _resetHubStateForTests(): void {
426
- inFlightDownloads = new Set();
427
- stopDownloadPolling();
428
- }
@@ -1,98 +0,0 @@
1
- // GENERATED-ADJACENT source module — part of the src/web/src/* split (plan
2
- // §7/§9 Phase 2). Built by scripts/build-web.ts into src/web/app.js, which
3
- // server.ts serves at GET /assets/app.js and app.html loads via
4
- // <script defer src="/assets/app.js"></script>.
5
- //
6
- // Entrypoint: imports and initializes every module in the EXACT order the
7
- // original monolithic inline <script> in app.html ran its top-level
8
- // statements. This ordering is load-bearing — see the design doc
9
- // (docs/design/web-chat-redesign.md §7) and the original file's own
10
- // section comments for why (controllers object populated before router()
11
- // can dispatch to it; updateTabFades() needs to run once at parse time so
12
- // a tab row that's already overflowing on load gets its fade mask
13
- // immediately, not just after the first scroll/resize event).
14
-
15
- import {
16
- controllers, initDeveloperToggle, initDrawer, initGlobalKeydown, initHfSettings,
17
- initRouter, initRoutesProbe, initServiceWorker, initShortcutSheet, initTheme, pollIdentity, router,
18
- } from "./shell";
19
- import { createChatController } from "./chat";
20
- import { createQuantizeController } from "./quantize";
21
- import { createFinetuneController } from "./finetune";
22
- import { createDatasetController } from "./dataset";
23
- import { createStatusController } from "./status";
24
- import { initModelPicker } from "./model-picker";
25
- import { initHubPanel } from "./hub";
26
- import { initPalette } from "./palette";
27
-
28
- // — Hugging Face nav gear + push-to-hub modal (original: an IIFE right
29
- // after mdToHtml/markdown helpers were declared) —
30
- initHfSettings();
31
-
32
- // — Theme: auto/dark/light, prefers-color-scheme —
33
- initTheme();
34
-
35
- // — PWA service worker (plan §9 Phase 3): shell-only cache-first, guarded
36
- // to secure contexts / localhost. Registration is fire-and-forget and
37
- // never blocks boot —
38
- initServiceWorker();
39
-
40
- // — Keyboard shortcut sheet (Cmd/Ctrl+/) —
41
- initShortcutSheet();
42
-
43
- // — Mobile drawer (chat sidebar slide-over) —
44
- initDrawer();
45
-
46
- // — Model picker (plan §5.6/§9 Phase 2): #nav-model click -> /library popover —
47
- initModelPicker();
48
-
49
- // — Model Hub (plan §9 Phase 3): panel chrome + the model picker's "Browse
50
- // models…" entry. Registers its open callback before initModelPicker's
51
- // popover is ever opened, so the first click already has a live target —
52
- // order here doesn't actually matter (both are registered-callback
53
- // indirections, not direct DOM lookups), but this keeps the two model-
54
- // surface inits visually adjacent in the boot sequence. —
55
- initHubPanel();
56
-
57
- // — Command palette (plan §9 Phase 3): Cmd/Ctrl+K, registers its
58
- // open/close/isOpen callbacks (shell.ts's registered-callback pattern)
59
- // before initGlobalKeydown()'s binding below can ever fire —
60
- initPalette();
61
-
62
- // — Global Escape/shortcut keydown sweep (closeTopOverlay + shortcuts) —
63
- initGlobalKeydown();
64
-
65
- // — Router: hashchange listener + tab-fade scroll/resize listeners, and
66
- // run updateTabFades() once immediately (original ran it at parse time,
67
- // not just on first scroll/resize) —
68
- initRouter();
69
-
70
- // — Developer toggle (plan §8/§9 Phase 2): apply saved/first-run state to
71
- // the tab row before router() below reads it for the deep-link check —
72
- initDeveloperToggle();
73
-
74
- // — Controller registration, in the original declaration order: chat,
75
- // quantize, finetune, dataset, status. Each controller's own init() is
76
- // still lazy (invoked by router() on first visit to that route) — only
77
- // the outer factory call (equivalent to the original IIFE body running)
78
- // happens here, matching the original's controllers.chat = (() => {...})()
79
- // eager-closure-body / lazy-init() split exactly. —
80
- controllers.chat = createChatController();
81
- controllers.quantize = createQuantizeController();
82
- controllers.finetune = createFinetuneController();
83
- controllers.dataset = createDatasetController();
84
- controllers.status = createStatusController();
85
-
86
- /* ════════════════════════════════════════════════════════════════════
87
- BOOT
88
- ════════════════════════════════════════════════════════════════════ */
89
- if (!location.hash) location.replace("#/chat");
90
- router();
91
- pollIdentity();
92
- setInterval(pollIdentity, 4000);
93
-
94
- // — Routes tab feature-detection (web-ui-pass-plan.md #17): probe /dag once;
95
- // hides the tab (and bounces off #/routes if already there) on 404. Runs
96
- // after the initial router() so a direct deep link to #/routes has
97
- // already rendered its section before this decides whether to keep it. —
98
- initRoutesProbe();
Binary file