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.
- package/README.md +1449 -109
- 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 +7 -5
- 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 +10 -10
- package/src/{serve → server}/chat-stage.ts +14 -14
- package/src/{client.ts → server/client.ts} +28 -2
- package/src/{serve → server}/completion-executor.ts +33 -41
- 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 +25 -82
- 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 +6 -8
- package/src/{serve → server}/request-prep.ts +47 -50
- 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 -107
- package/docs/archive/hlg-sampling.md +0 -350
- package/docs/archive/investigations/audio-input-plan.md +0 -356
- package/docs/archive/investigations/benchmark-ledger-through-v0.4.0.md +0 -3617
- 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/release-notes-v0.4.0.md +0 -45
- package/docs/archive/planning/release-notes-v0.5.0.md +0 -37
- 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 -2054
- package/docs/design/decode-speed-program.md +0 -2404
- package/docs/design/dreaming-nightly-pipeline.md +0 -684
- package/docs/design/generic-model-support.md +0 -821
- package/docs/design/kv-cache.md +0 -993
- package/docs/design/orpo-training.md +0 -680
- package/docs/design/speculative-decoding.md +0 -2129
- package/docs/design/turboquant.md +0 -2225
- package/docs/design/unified-engine-frontier-plan.md +0 -1987
- package/docs/design/web-chat-redesign.md +0 -695
- package/docs/planning/PRODUCT_ROADMAP.md +0 -593
- package/docs/planning/ResearchTopics.md +0 -119
- package/docs/planning/public-readiness-handoff.md +0 -244
- package/docs/planning/public-readiness.md +0 -149
- package/docs/planning/video-series.md +0 -421
- package/docs/reference/benchmarks.md +0 -3587
- package/docs/reference/cli.md +0 -646
- package/docs/reference/distribution.md +0 -380
- package/docs/reference/environment.md +0 -494
- package/docs/reference/glossary.md +0 -39
- package/docs/reference/library-api.md +0 -688
- package/docs/reference/memory.md +0 -260
- package/docs/reference/models.md +0 -346
- package/docs/reference/server-api.md +0 -1945
- package/docs/reference/server-config.md +0 -973
- package/docs/reference/training.md +0 -462
- package/docs/reference/troubleshooting.md +0 -158
- package/src/admission-plan.ts +0 -18
- package/src/audio/audiotoolbox.ts +0 -112
- package/src/audio/conformer.ts +0 -655
- package/src/audio/decode.ts +0 -133
- package/src/audio/features.ts +0 -202
- package/src/audio/mic-capture.ts +0 -98
- package/src/audio/silero-vad.ts +0 -379
- package/src/audio/transcode.ts +0 -67
- package/src/audio/whisper-decode.ts +0 -774
- package/src/audio/whisper-fast.ts +0 -732
- package/src/audio/whisper-format.ts +0 -73
- package/src/audio/whisper-mel.ts +0 -185
- package/src/audio/whisper-timing.ts +0 -274
- package/src/audio/whisper-tokenizer.ts +0 -239
- package/src/audio/whisper-transcribe.ts +0 -410
- 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 -129
- package/src/backends/mlx/batch-group.ts +0 -1874
- 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 -107
- 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/encoder-cache.ts +0 -25
- package/src/backends/mlx/evaluation.ts +0 -105
- package/src/backends/mlx/fill-append.ts +0 -58
- package/src/backends/mlx/fill-group.ts +0 -497
- 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 -174
- 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 -161
- 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/mixed-iteration.ts +0 -62
- package/src/backends/mlx/model-prompt.ts +0 -178
- package/src/backends/mlx/model-serving.ts +0 -80
- package/src/backends/mlx/paged-cache-codec.ts +0 -27
- package/src/backends/mlx/prefill-cohort.ts +0 -100
- package/src/backends/mlx/prefill-policy.ts +0 -26
- package/src/backends/mlx/prefill-rows.ts +0 -244
- package/src/backends/mlx/prefill.ts +0 -33
- package/src/backends/mlx/projected-context-rows.ts +0 -72
- package/src/backends/mlx/prompt-input.ts +0 -29
- package/src/backends/mlx/qwen-prompt-input.ts +0 -40
- package/src/backends/mlx/request-policy.ts +0 -20
- package/src/backends/mlx/request-state-policy.ts +0 -54
- package/src/backends/mlx/rollback.ts +0 -92
- package/src/backends/mlx/row-sampling.ts +0 -49
- package/src/backends/mlx/serial-executor.ts +0 -251
- package/src/backends/mlx/speculative-group.ts +0 -438
- package/src/backends/mlx/speculative-round.ts +0 -161
- package/src/backends/mlx/speculative.ts +0 -131
- 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 -173
- package/src/chat-template.ts +0 -382
- package/src/cli.ts +0 -3490
- package/src/config.ts +0 -399
- package/src/contracts/completion.ts +0 -16
- package/src/contracts/execution.ts +0 -70
- package/src/contracts/generation.ts +0 -103
- package/src/contracts/host.ts +0 -7
- package/src/contracts/object-cache.ts +0 -8
- package/src/contracts/pi-provider.ts +0 -30
- package/src/contracts/pi.ts +0 -261
- package/src/contracts/prefix-cache.ts +0 -33
- package/src/contracts/resources.ts +0 -25
- package/src/contracts/sampling.ts +0 -23
- package/src/contracts/scheduling.ts +0 -54
- 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 -64
- package/src/engine/execution-tasks.ts +0 -44
- package/src/engine/namespaced-cache.ts +0 -13
- package/src/engine/resources.ts +0 -53
- package/src/engine/scheduler.ts +0 -92
- 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 -463
- 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 -1352
- package/src/grammar.ts +0 -604
- 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 -45
- 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 -279
- package/src/kv-store.ts +0 -1135
- package/src/lab/README.md +0 -22
- 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-attention.ts +0 -117
- package/src/lab/paged-kv/paged-kv-rows.ts +0 -111
- package/src/lab/paged-kv/paged-kv.ts +0 -336
- package/src/library.ts +0 -40
- package/src/lora.ts +0 -408
- package/src/md-modules.d.ts +0 -54
- 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 -425
- 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 -531
- package/src/mlx/materialize.ts +0 -23
- package/src/mlx/metal-kernel.ts +0 -159
- package/src/mlx/normalized-argmax.ts +0 -116
- package/src/mlx/ops.ts +0 -1108
- 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 -94
- 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 -57
- package/src/model/diffusion-gemma.ts +0 -1026
- package/src/model/factory.ts +0 -215
- package/src/model/fingerprint.ts +0 -45
- 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 -2370
- package/src/model/gemma4.ts +0 -1395
- 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 -70
- package/src/model/minicpm5.ts +0 -376
- package/src/model/profile.ts +0 -468
- package/src/model/quantized-append-attention.ts +0 -31
- 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 -250
- package/src/model/qwen3.ts +0 -277
- package/src/model/qwen3_5-checkpoint.ts +0 -164
- package/src/model/qwen3_5.ts +0 -881
- 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 -110
- package/src/model/token-groups.ts +0 -66
- 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 -89
- 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/model/whisper.ts +0 -435
- package/src/native/expert_io.c +0 -523
- package/src/native/frame_extract.swift +0 -140
- package/src/native/mic_capture.swift +0 -92
- package/src/native-pack.ts +0 -163
- 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 -1940
- package/src/preflight.ts +0 -158
- package/src/prompt-cache.ts +0 -682
- 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 -77
- package/src/safetensors.ts +0 -165
- package/src/sampler.ts +0 -1012
- 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/audio-routes.ts +0 -323
- 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 -537
- 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 -49
- package/src/serve/model-host.ts +0 -511
- package/src/serve/parent-application.ts +0 -56
- package/src/serve/prompt-response-trace.ts +0 -161
- 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/serve/transcription-server.ts +0 -64
- package/src/serve/transcription-service.ts +0 -502
- package/src/server.ts +0 -1282
- package/src/spec/assistant-rows.ts +0 -93
- 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 -318
- package/src/spec/glm52-mtp-state.ts +0 -33
- package/src/spec/ngram-source.ts +0 -226
- 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 -230
- package/src/spec/qwen-mtp-source.ts +0 -386
- package/src/spec/qwen-mtp-state.ts +0 -30
- package/src/spec/serve-loop.ts +0 -587
- package/src/spec/source.ts +0 -228
- package/src/spec/two-model-rows.ts +0 -62
- package/src/spec/two-model.ts +0 -113
- package/src/ssd-cache.ts +0 -431
- package/src/ssd-durability.ts +0 -209
- package/src/storage/host-buffer.ts +0 -24
- package/src/storage/kv-writer.ts +0 -83
- package/src/storage/kv-writer.worker.js +0 -264
- package/src/storage/retention-policy.ts +0 -37
- package/src/tiered-prompt-cache.ts +0 -45
- 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 -176
- package/src/vision/preprocess.ts +0 -247
- package/src/vision/prompt.ts +0 -409
- package/src/vision/qwen3vl-preprocess.ts +0 -409
- package/src/vision/qwen3vl-prompt.ts +0 -233
- package/src/vision/qwen3vl-tower.ts +0 -566
- package/src/vision/siglip.ts +0 -607
- package/src/vision/video-frames.ts +0 -148
- package/src/web/app.html +0 -2155
- package/src/web/app.js +0 -5863
- 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 -1060
- package/src/web/src/composer.ts +0 -962
- 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/src/voice.ts +0 -141
- 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/{serve → server}/generated-token-history.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
|
@@ -1,112 +0,0 @@
|
|
|
1
|
-
// In-process audio decoding through macOS AudioToolbox (ExtAudioFile) over
|
|
2
|
-
// bun:ffi — every container/codec CoreAudio reads (mp3, m4a/aac, flac, ogg,
|
|
3
|
-
// aiff, caf, wav, mov/mp4 audio tracks…) straight to 16 kHz mono float32,
|
|
4
|
-
// with CoreAudio's own resampler. Replaces the `afconvert` subprocess of
|
|
5
|
-
// transcode.ts (no process spawn, no WAV re-parse); the bytes still touch a
|
|
6
|
-
// temp file because ExtAudioFile opens URLs.
|
|
7
|
-
|
|
8
|
-
import { dlopen, FFIType, ptr, read } from "bun:ffi";
|
|
9
|
-
import { randomUUID } from "node:crypto";
|
|
10
|
-
import { unlinkSync, writeFileSync } from "node:fs";
|
|
11
|
-
import { tmpdir } from "node:os";
|
|
12
|
-
import { join } from "node:path";
|
|
13
|
-
|
|
14
|
-
const { ptr: P, i32, u32, u64, i64 } = FFIType;
|
|
15
|
-
|
|
16
|
-
function openLibs() {
|
|
17
|
-
const at = dlopen("/System/Library/Frameworks/AudioToolbox.framework/AudioToolbox", {
|
|
18
|
-
ExtAudioFileOpenURL: { args: [u64, P], returns: i32 },
|
|
19
|
-
ExtAudioFileSetProperty: { args: [u64, u32, u32, P], returns: i32 },
|
|
20
|
-
ExtAudioFileGetProperty: { args: [u64, u32, P, P], returns: i32 },
|
|
21
|
-
ExtAudioFileRead: { args: [u64, P, P], returns: i32 },
|
|
22
|
-
ExtAudioFileDispose: { args: [u64], returns: i32 },
|
|
23
|
-
}).symbols;
|
|
24
|
-
const cf = dlopen("/System/Library/Frameworks/CoreFoundation.framework/CoreFoundation", {
|
|
25
|
-
CFURLCreateFromFileSystemRepresentation: { args: [u64, P, i64, FFIType.bool], returns: u64 },
|
|
26
|
-
CFRelease: { args: [u64], returns: i32 },
|
|
27
|
-
}).symbols;
|
|
28
|
-
return { at, cf };
|
|
29
|
-
}
|
|
30
|
-
|
|
31
|
-
let libs: ReturnType<typeof openLibs> | null = null;
|
|
32
|
-
const load = () => (libs ??= openLibs());
|
|
33
|
-
|
|
34
|
-
const fourcc = (s: string) => (s.charCodeAt(0) << 24 | s.charCodeAt(1) << 16 | s.charCodeAt(2) << 8 | s.charCodeAt(3)) >>> 0;
|
|
35
|
-
const kExtAudioFileProperty_ClientDataFormat = fourcc("cfmt");
|
|
36
|
-
const kExtAudioFileProperty_FileLengthFrames = fourcc("#frm");
|
|
37
|
-
const kAudioFormatLinearPCM = fourcc("lpcm");
|
|
38
|
-
const kAudioFormatFlagIsFloat = 1;
|
|
39
|
-
const kAudioFormatFlagIsPacked = 8;
|
|
40
|
-
|
|
41
|
-
export const AUDIOTOOLBOX_SAMPLE_RATE = 16_000;
|
|
42
|
-
|
|
43
|
-
/** Decode a file to 16 kHz mono float32 with CoreAudio. Throws on undecodable input. */
|
|
44
|
-
export function decodeFileWithAudioToolbox(path: string, sampleRate = AUDIOTOOLBOX_SAMPLE_RATE): Float32Array {
|
|
45
|
-
const { at, cf } = load();
|
|
46
|
-
const pathBuf = Buffer.from(path, "utf8");
|
|
47
|
-
const url = cf.CFURLCreateFromFileSystemRepresentation(0n, ptr(pathBuf), BigInt(pathBuf.length), false) as bigint;
|
|
48
|
-
if (!url) throw new Error("AudioToolbox: cannot build file URL");
|
|
49
|
-
const refSlot = new BigUint64Array(1);
|
|
50
|
-
try {
|
|
51
|
-
let st = at.ExtAudioFileOpenURL(url, ptr(refSlot)) as number;
|
|
52
|
-
if (st !== 0) throw new Error(`AudioToolbox: unrecognized or undecodable audio (ExtAudioFileOpenURL ${st})`);
|
|
53
|
-
const ref = read.u64(ptr(refSlot), 0);
|
|
54
|
-
try {
|
|
55
|
-
// AudioStreamBasicDescription: 40 bytes
|
|
56
|
-
const asbd = Buffer.alloc(40);
|
|
57
|
-
asbd.writeDoubleLE(sampleRate, 0);
|
|
58
|
-
asbd.writeUInt32LE(kAudioFormatLinearPCM, 8);
|
|
59
|
-
asbd.writeUInt32LE(kAudioFormatFlagIsFloat | kAudioFormatFlagIsPacked, 12);
|
|
60
|
-
asbd.writeUInt32LE(4, 16); // bytes per packet
|
|
61
|
-
asbd.writeUInt32LE(1, 20); // frames per packet
|
|
62
|
-
asbd.writeUInt32LE(4, 24); // bytes per frame
|
|
63
|
-
asbd.writeUInt32LE(1, 28); // channels
|
|
64
|
-
asbd.writeUInt32LE(32, 32); // bits per channel
|
|
65
|
-
st = at.ExtAudioFileSetProperty(ref, kExtAudioFileProperty_ClientDataFormat, 40, ptr(asbd)) as number;
|
|
66
|
-
if (st !== 0) throw new Error(`AudioToolbox: cannot set client format (${st})`);
|
|
67
|
-
// frame count hint (source rate), used only to size the first buffer
|
|
68
|
-
const lenSlot = new BigInt64Array(1);
|
|
69
|
-
const lenSize = new Uint32Array([8]);
|
|
70
|
-
at.ExtAudioFileGetProperty(ref, kExtAudioFileProperty_FileLengthFrames, ptr(lenSize), ptr(lenSlot));
|
|
71
|
-
const chunkFrames = 1 << 16;
|
|
72
|
-
const chunk = new Float32Array(chunkFrames);
|
|
73
|
-
// AudioBufferList { UInt32 mNumberBuffers; [pad 4]; AudioBuffer { UInt32 mNumberChannels; UInt32 mDataByteSize; void* mData } }
|
|
74
|
-
const abl = Buffer.alloc(24);
|
|
75
|
-
const frames = new Uint32Array(1);
|
|
76
|
-
const parts: Float32Array[] = [];
|
|
77
|
-
let total = 0;
|
|
78
|
-
for (;;) {
|
|
79
|
-
abl.writeUInt32LE(1, 0);
|
|
80
|
-
abl.writeUInt32LE(1, 8);
|
|
81
|
-
abl.writeUInt32LE(chunkFrames * 4, 12);
|
|
82
|
-
abl.writeBigUInt64LE(BigInt(ptr(chunk)), 16);
|
|
83
|
-
frames[0] = chunkFrames;
|
|
84
|
-
st = at.ExtAudioFileRead(ref, ptr(frames), ptr(abl)) as number;
|
|
85
|
-
if (st !== 0) throw new Error(`AudioToolbox: decode failed (ExtAudioFileRead ${st})`);
|
|
86
|
-
const n = frames[0]!;
|
|
87
|
-
if (n === 0) break;
|
|
88
|
-
parts.push(chunk.slice(0, n));
|
|
89
|
-
total += n;
|
|
90
|
-
}
|
|
91
|
-
const out = new Float32Array(total);
|
|
92
|
-
let o = 0;
|
|
93
|
-
for (const p of parts) { out.set(p, o); o += p.length; }
|
|
94
|
-
return out;
|
|
95
|
-
} finally {
|
|
96
|
-
at.ExtAudioFileDispose(ref);
|
|
97
|
-
}
|
|
98
|
-
} finally {
|
|
99
|
-
cf.CFRelease(url);
|
|
100
|
-
}
|
|
101
|
-
}
|
|
102
|
-
|
|
103
|
-
/** Decode in-memory bytes (any CoreAudio-readable container) via a temp file. */
|
|
104
|
-
export function decodeBytesWithAudioToolbox(bytes: Uint8Array, sampleRate = AUDIOTOOLBOX_SAMPLE_RATE): Float32Array {
|
|
105
|
-
const path = join(tmpdir(), `mlx-bun-audio-${randomUUID()}`);
|
|
106
|
-
writeFileSync(path, bytes);
|
|
107
|
-
try {
|
|
108
|
-
return decodeFileWithAudioToolbox(path, sampleRate);
|
|
109
|
-
} finally {
|
|
110
|
-
try { unlinkSync(path); } catch { /* already gone */ }
|
|
111
|
-
}
|
|
112
|
-
}
|