mlx-bun 0.4.0 → 1.0.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +1453 -88
- package/THIRD_PARTY_NOTICES.md +882 -0
- package/bin/mlx-bun.convert.mjs +4 -0
- package/bin/mlx-bun.fuse.mjs +4 -0
- package/bin/mlx-bun.generate.mjs +4 -0
- package/bin/mlx-bun.lora.mjs +4 -0
- package/bin/mlx-bun.mjs +13 -21
- package/bin/mlx-bun.server.mjs +4 -0
- package/bin/mlx-bun.upload.mjs +4 -0
- package/dist/web/app.js +9584 -0
- package/package.json +50 -53
- package/src/cli/ane-install.ts +76 -0
- package/src/cli/args.ts +179 -0
- package/src/cli/engine-entry.ts +68 -0
- package/src/cli/inference.ts +214 -0
- package/src/cli/job-entry.ts +97 -0
- package/src/cli/library-host.ts +117 -0
- package/src/cli/main.ts +57 -0
- package/src/cli/memory-engine.ts +163 -0
- package/src/cli/memory.ts +36 -0
- package/src/cli/mlx-lm-aliases.ts +309 -0
- package/src/cli/model-choice.ts +33 -0
- package/src/cli/model-selection.ts +84 -0
- package/src/cli/module-host.ts +11 -0
- package/src/cli/module-verbs.ts +50 -0
- package/src/cli/numerical-policy.ts +12 -0
- package/src/cli/perplexity.ts +89 -0
- package/src/cli/serve-host.ts +423 -0
- package/src/cli/serve-isolated.ts +217 -0
- package/src/cli/serve-options.ts +106 -0
- package/src/cli/serve-state.ts +205 -0
- package/src/cli/serve.ts +500 -0
- package/src/cli/served-model-host.ts +102 -0
- package/src/cli/server-entry.ts +148 -0
- package/src/cli/serving-unit.ts +177 -0
- package/src/{tui.ts → cli/terminal.ts} +13 -8
- package/src/cli/worker-adapters.ts +34 -0
- package/src/cli/worker-entry.ts +272 -0
- package/src/cli/worker-unit.ts +179 -0
- package/src/engine/adapter-operation.ts +26 -0
- package/src/engine/cache-services.ts +114 -0
- package/src/engine/completion.ts +39 -0
- package/src/engine/generation-gateway.ts +358 -0
- package/src/engine/index.ts +59 -0
- package/src/engine/kv-budget.ts +120 -0
- package/src/engine/media-preparation.ts +232 -0
- package/src/engine/model-binding.ts +24 -0
- package/src/engine/model-host.ts +513 -0
- package/src/engine/model-serving.ts +89 -0
- package/src/{serve → engine}/preparation.ts +10 -8
- package/src/{serve → engine}/session-completion-engine.ts +6 -6
- package/src/engine/telemetry.ts +109 -0
- package/src/hub/downloads.ts +116 -0
- package/src/jobs/db.ts +16 -7
- package/src/jobs/events.ts +29 -0
- package/src/jobs/executable.ts +5 -0
- package/src/jobs/host.ts +51 -0
- package/src/jobs/in-process.ts +55 -0
- package/src/{contracts/jobs.ts → jobs/protocol.ts} +7 -9
- package/src/jobs/runner.ts +150 -231
- package/src/jobs/service.ts +75 -0
- package/src/jobs/sse.ts +59 -23
- package/src/jobs/worker-memory.ts +46 -0
- package/src/jobs/worker-process.ts +163 -0
- package/src/jobs/worker-supervisor.ts +258 -0
- package/src/modules.ts +58 -0
- package/src/publishing/catalog-hub.ts +18 -0
- package/src/publishing/credentials.ts +31 -0
- package/src/publishing/upload.ts +37 -0
- package/src/residency/leased-adapters.ts +21 -0
- package/src/residency/local-records.ts +16 -0
- package/src/residency/model-residency.ts +405 -0
- package/src/residency/record-index.ts +28 -0
- package/src/residency/resident-estimate.ts +50 -0
- package/src/{anthropic.ts → server/anthropic.ts} +3 -81
- package/src/server/cache-routes.ts +36 -0
- package/src/{serve → server}/chat-request.ts +12 -10
- package/src/{serve → server}/chat-stage.ts +27 -22
- package/src/{client.ts → server/client.ts} +28 -2
- package/src/{serve → server}/completion-executor.ts +42 -42
- package/src/server/decision-body.ts +55 -0
- package/src/server/decision-routes.ts +134 -0
- package/src/server/decision-wire.ts +228 -0
- package/src/{serve → server}/discovery-routes.ts +42 -83
- package/src/server/generated-token-history.ts +57 -0
- package/src/{serve → server}/http.ts +25 -13
- package/src/{serve → server}/inference-request.ts +5 -5
- package/src/server/job-routes.ts +20 -0
- package/src/server/media-prompt.ts +63 -0
- package/src/server/memory-completion-client.ts +46 -0
- package/src/server/model-routes.ts +192 -0
- package/src/{serve → server}/openai-wire.ts +3 -2
- package/src/{serve → server}/pipeline.ts +1 -1
- package/src/server/prompt-builder.ts +13 -0
- package/src/server/prompt-contracts.ts +17 -0
- package/src/server/proxy-routes.ts +187 -0
- package/src/server/publishing-routes.ts +41 -0
- package/src/{serve → server}/request-plan.ts +8 -8
- package/src/{serve → server}/request-prep.ts +82 -68
- package/src/{serve → server}/responses-client.ts +30 -13
- package/src/{responses.ts → server/responses.ts} +12 -93
- package/src/server/routes.ts +205 -0
- package/src/server/start.ts +72 -0
- package/src/server/status-routes.ts +108 -0
- package/src/{serve → server}/text-completion-stage.ts +25 -4
- package/src/server/text-prompt.ts +12 -0
- package/src/{serve → server}/token-streams.ts +36 -38
- package/src/server/worker-routes.ts +311 -0
- package/src/storage/paths.ts +61 -0
- package/src/web/assets.ts +37 -0
- package/src/web/{src → browser}/api.ts +5 -23
- package/src/web/browser/chat-host.ts +17 -0
- package/src/web/browser/installed-panels.ts +6 -0
- package/src/web/browser/main.ts +55 -0
- package/src/web/browser/markdown.ts +279 -0
- package/src/web/browser/memory-panel.ts +9 -0
- package/src/web/browser/palette.ts +64 -0
- package/src/web/browser/protocol.ts +13 -0
- package/src/web/browser/shell.ts +467 -0
- package/src/web/{src → browser}/status.ts +32 -36
- package/src/web/build.ts +51 -0
- package/src/web/public/app.html +1054 -0
- package/src/web/{sw.js → public/sw.js} +4 -3
- package/src/web/{vendor → public/vendor}/hljs-theme.css +2 -2
- package/src/web/{vendor → public/vendor}/hljs.js +2 -2
- package/THIRD_PARTY_LICENSES.md +0 -51
- package/docs/README.md +0 -94
- package/docs/archive/hlg-sampling.md +0 -350
- package/docs/archive/investigations/audio-input-plan.md +0 -356
- package/docs/archive/investigations/chunk-finetune-experiment.md +0 -84
- package/docs/archive/investigations/colibri-glm52-port.md +0 -1486
- package/docs/archive/investigations/colibri-oracle-pin.md +0 -1284
- package/docs/archive/investigations/curve-bisector-routing.md +0 -123
- package/docs/archive/investigations/decode-roofline-lookagain.md +0 -348
- package/docs/archive/investigations/diffusion-gemma-port.md +0 -602
- package/docs/archive/investigations/dspark-handoff.md +0 -224
- package/docs/archive/investigations/expert-offload-single-user-moe.md +0 -292
- package/docs/archive/investigations/faithful-l1-consolidation.md +0 -310
- package/docs/archive/investigations/hlg-sampling-investigation.md +0 -970
- package/docs/archive/investigations/indexshare-performance-spike.md +0 -357
- package/docs/archive/investigations/kernel-perf-review-2026-07.md +0 -171
- package/docs/archive/investigations/lab-build-journal.md +0 -157
- package/docs/archive/investigations/minicpm5-decode-megakernel.md +0 -489
- package/docs/archive/investigations/openjarvis-fit.md +0 -198
- package/docs/archive/investigations/openwiki-evaluation.md +0 -249
- package/docs/archive/investigations/optimization_plan.md +0 -346
- package/docs/archive/investigations/orpo-base-uf-experiment-and-directions.md +0 -120
- package/docs/archive/investigations/orpo-flash-cce-pin-leak.md +0 -167
- package/docs/archive/investigations/orpo-uf-testing-handoff.md +0 -126
- package/docs/archive/investigations/pareto-specialized-runtime-findings.md +0 -92
- package/docs/archive/investigations/pi-builtin-investigation.md +0 -308
- package/docs/archive/investigations/pre-colibri-stabilization.md +0 -329
- package/docs/archive/investigations/qwen38-27b-campaign-log.md +0 -1092
- package/docs/archive/investigations/segmented-backward-handoff.md +0 -148
- package/docs/archive/investigations/starter-model-port-handoff.md +0 -342
- package/docs/archive/investigations/steel-flash-cce-handoff.md +0 -489
- package/docs/archive/investigations/the-dreaming-handoff.md +0 -141
- package/docs/archive/investigations/trainer-validation-experiment.md +0 -173
- package/docs/archive/planning/chat-ui-vision-memory-first.md +0 -199
- package/docs/archive/planning/chat-ui-vision.md +0 -50
- package/docs/archive/planning/chunk-task-roadmap.md +0 -77
- package/docs/archive/planning/curve-sampler-research-plan.md +0 -331
- package/docs/archive/planning/journal.md +0 -316
- package/docs/archive/planning/memory-docs-and-dag-plan.md +0 -194
- package/docs/archive/planning/pi-integration-review.md +0 -178
- package/docs/archive/planning/release-notes-v0.0.10.md +0 -57
- package/docs/archive/planning/release-notes-v0.0.11.md +0 -239
- package/docs/archive/planning/release-notes-v0.0.12.md +0 -80
- package/docs/archive/planning/release-notes-v0.0.13.md +0 -77
- package/docs/archive/planning/release-notes-v0.0.9.md +0 -122
- package/docs/archive/planning/release-notes-v0.2.0.md +0 -49
- package/docs/archive/planning/release-notes-v0.3.0.md +0 -35
- package/docs/archive/planning/web-ui-pass-plan.md +0 -213
- package/docs/archive/planning/website-readme-pass-plan.md +0 -224
- package/docs/design/batching.md +0 -1841
- package/docs/design/decode-speed-program.md +0 -2349
- package/docs/design/dreaming-nightly-pipeline.md +0 -684
- package/docs/design/generic-model-support.md +0 -705
- package/docs/design/kv-cache.md +0 -824
- package/docs/design/orpo-training.md +0 -680
- package/docs/design/speculative-decoding.md +0 -1990
- package/docs/design/turboquant.md +0 -2212
- package/docs/design/unified-engine-frontier-plan.md +0 -1880
- package/docs/design/web-chat-redesign.md +0 -695
- package/docs/planning/PRODUCT_ROADMAP.md +0 -594
- package/docs/planning/ResearchTopics.md +0 -119
- package/docs/planning/release-notes-v0.4.0.md +0 -45
- package/docs/planning/video-series.md +0 -421
- package/docs/reference/benchmarks.md +0 -3613
- package/docs/reference/cli.md +0 -585
- package/docs/reference/distribution.md +0 -355
- package/docs/reference/environment.md +0 -398
- package/docs/reference/glossary.md +0 -39
- package/docs/reference/library-api.md +0 -674
- package/docs/reference/memory.md +0 -260
- package/docs/reference/models.md +0 -334
- package/docs/reference/server-api.md +0 -1775
- package/docs/reference/server-config.md +0 -846
- package/docs/reference/training.md +0 -462
- package/docs/reference/troubleshooting.md +0 -158
- package/src/admission-plan.ts +0 -18
- package/src/audio/conformer.ts +0 -647
- package/src/audio/decode.ts +0 -133
- package/src/audio/features.ts +0 -202
- package/src/audio/transcode.ts +0 -67
- package/src/backends/mlx/assistant-target.ts +0 -46
- package/src/backends/mlx/autoregressive-draft-rows.ts +0 -120
- package/src/backends/mlx/autoregressive.ts +0 -108
- package/src/backends/mlx/batch-group.ts +0 -1790
- package/src/backends/mlx/cache-attachment.ts +0 -67
- package/src/backends/mlx/cache-layout.ts +0 -31
- package/src/backends/mlx/checkpoint-state.ts +0 -53
- package/src/backends/mlx/continuation-persistence.ts +0 -106
- package/src/backends/mlx/continuation-request.ts +0 -43
- package/src/backends/mlx/continuation.ts +0 -77
- package/src/backends/mlx/diffusion.ts +0 -39
- package/src/backends/mlx/draft-target.ts +0 -32
- package/src/backends/mlx/evaluation.ts +0 -105
- package/src/backends/mlx/fill-append.ts +0 -45
- package/src/backends/mlx/full-kv-row-append.ts +0 -32
- package/src/backends/mlx/full-kv-row-donor.ts +0 -54
- package/src/backends/mlx/full-transitioning-kv-rows.ts +0 -65
- package/src/backends/mlx/gateway-binding.ts +0 -130
- package/src/backends/mlx/graph.ts +0 -38
- package/src/backends/mlx/kv-codec.ts +0 -9
- package/src/backends/mlx/kv-maintenance.ts +0 -133
- package/src/backends/mlx/kv-tensor-rows.ts +0 -212
- package/src/backends/mlx/memory-client.ts +0 -189
- package/src/backends/mlx/methods.ts +0 -90
- package/src/backends/mlx/model-prompt.ts +0 -164
- package/src/backends/mlx/model-serving.ts +0 -79
- package/src/backends/mlx/prefill-cohort.ts +0 -89
- package/src/backends/mlx/prefill-rows.ts +0 -200
- package/src/backends/mlx/prefill.ts +0 -33
- package/src/backends/mlx/projected-context-rows.ts +0 -72
- package/src/backends/mlx/request-policy.ts +0 -20
- package/src/backends/mlx/request-state-policy.ts +0 -41
- package/src/backends/mlx/rollback.ts +0 -89
- package/src/backends/mlx/row-sampling.ts +0 -49
- package/src/backends/mlx/serial-executor.ts +0 -242
- package/src/backends/mlx/speculative-group.ts +0 -359
- package/src/backends/mlx/speculative-round.ts +0 -161
- package/src/backends/mlx/speculative.ts +0 -114
- package/src/backends/mlx/state-rows.ts +0 -104
- package/src/backends/mlx/state-views.ts +0 -46
- package/src/backends/mlx/target-layout-capability.ts +0 -35
- package/src/backends/mlx/text-engine.ts +0 -73
- package/src/backends/mlx/transitioning-kv-rows.ts +0 -165
- package/src/chat-template.ts +0 -382
- package/src/cli.ts +0 -3168
- package/src/config.ts +0 -399
- package/src/contracts/completion.ts +0 -16
- package/src/contracts/execution.ts +0 -60
- package/src/contracts/generation.ts +0 -103
- package/src/contracts/host.ts +0 -7
- package/src/contracts/pi-provider.ts +0 -30
- package/src/contracts/pi.ts +0 -259
- package/src/contracts/prefix-cache.ts +0 -23
- package/src/contracts/resources.ts +0 -25
- package/src/contracts/sampling.ts +0 -23
- package/src/contracts/scheduling.ts +0 -45
- package/src/contracts/tsconfig.json +0 -14
- package/src/dataset/generators.ts +0 -1087
- package/src/dataset/index.ts +0 -12
- package/src/dataset/job.ts +0 -81
- package/src/dataset/llm.ts +0 -71
- package/src/dataset/registry.ts +0 -528
- package/src/diffusion/diffusion-generate.ts +0 -507
- package/src/download.ts +0 -505
- package/src/embed.ts +0 -77
- package/src/engine/admission.ts +0 -59
- package/src/engine/cancellation.ts +0 -38
- package/src/engine/engine.ts +0 -56
- package/src/engine/execution-coordinator.ts +0 -70
- package/src/engine/execution-plan.ts +0 -53
- package/src/engine/resources.ts +0 -53
- package/src/engine/scheduler.ts +0 -72
- package/src/engine/session.ts +0 -247
- package/src/eval/README.md +0 -130
- package/src/eval/ifeval.ts +0 -81
- package/src/eval/kl-prompts.ts +0 -29
- package/src/eval/kl.ts +0 -590
- package/src/eval/perplexity.ts +0 -164
- package/src/eval/quality-db.ts +0 -92
- package/src/eval/reference-logits.ts +0 -207
- package/src/eval/runner.ts +0 -124
- package/src/eval/score.ts +0 -49
- package/src/eval/tasks/bfcl.ts +0 -535
- package/src/eval/tasks/gsm8k.ts +0 -98
- package/src/eval/tasks/hashhop.ts +0 -139
- package/src/eval/tasks/humaneval.ts +0 -275
- package/src/eval/tasks/ifeval.ts +0 -525
- package/src/eval/tasks/mmlu.ts +0 -157
- package/src/eval/ultrafeedback.ts +0 -60
- package/src/evaldb.ts +0 -199
- package/src/expert-io.ts +0 -393
- package/src/expert-offload-build.ts +0 -77
- package/src/expert-offload.ts +0 -76
- package/src/expert-residency.ts +0 -1134
- package/src/expert-usage.ts +0 -415
- package/src/fill/echo-index.ts +0 -188
- package/src/fill/fill-session.ts +0 -452
- package/src/fill/proposal.ts +0 -68
- package/src/fill/schema-rows.ts +0 -506
- package/src/fill/tool-boundary.ts +0 -103
- package/src/fit.ts +0 -249
- package/src/generate.ts +0 -1346
- package/src/grammar.ts +0 -591
- package/src/harness-pi.ts +0 -184
- package/src/hf-push.ts +0 -628
- package/src/hub-rest.ts +0 -242
- package/src/index.ts +0 -63
- package/src/inference/denoising.ts +0 -17
- package/src/inference/draft-acceptance.ts +0 -41
- package/src/inference/draft-output.ts +0 -49
- package/src/inference/graph.ts +0 -33
- package/src/inference/prefill.ts +0 -39
- package/src/inference/rollback.ts +0 -7
- package/src/jobs/index.ts +0 -31
- package/src/jobs/job-entry.ts +0 -107
- package/src/jobs/task-client.ts +0 -19
- package/src/jobs/types.ts +0 -2
- package/src/kv-scheme.ts +0 -257
- package/src/kv-store.ts +0 -1099
- package/src/lab/README.md +0 -18
- package/src/lab/curve/curve-designer.html +0 -303
- package/src/lab/curve/curve-sampler.ts +0 -122
- package/src/lab/expert-trace/expert-trace.ts +0 -69
- package/src/lab/paged-kv/paged-kv-rows.ts +0 -83
- package/src/lab/paged-kv/paged-kv.ts +0 -270
- package/src/library.ts +0 -40
- package/src/lora.ts +0 -408
- package/src/md-modules.d.ts +0 -49
- package/src/media-fetch.ts +0 -442
- package/src/memory/article.ts +0 -558
- package/src/memory/chunk-validation.ts +0 -150
- package/src/memory/chunk.ts +0 -318
- package/src/memory/cluster.ts +0 -351
- package/src/memory/coverage.ts +0 -134
- package/src/memory/crosslink.ts +0 -0
- package/src/memory/db.ts +0 -253
- package/src/memory/entity.ts +0 -203
- package/src/memory/gate.ts +0 -187
- package/src/memory/history.ts +0 -152
- package/src/memory/ingest.ts +0 -335
- package/src/memory/model.ts +0 -183
- package/src/memory/normalize.ts +0 -671
- package/src/memory/parse.ts +0 -22
- package/src/memory/pipeline.ts +0 -275
- package/src/memory/prompts.ts +0 -65
- package/src/memory/query.ts +0 -334
- package/src/memory/reconcile.ts +0 -734
- package/src/memory/redirect.ts +0 -108
- package/src/memory/reindex.ts +0 -325
- package/src/memory/resolve.ts +0 -307
- package/src/memory/rest.ts +0 -245
- package/src/memory/route.ts +0 -272
- package/src/memory/schedule.ts +0 -165
- package/src/memory/stages.ts +0 -777
- package/src/memory/synthesize.ts +0 -1638
- package/src/memory/tools.ts +0 -749
- package/src/memory/vault.ts +0 -1194
- package/src/memory/wikify.ts +0 -1064
- package/src/mlx/array.ts +0 -388
- package/src/mlx/autograd.ts +0 -304
- package/src/mlx/checkpoint.ts +0 -149
- package/src/mlx/compile.ts +0 -168
- package/src/mlx/custom-vjp.ts +0 -188
- package/src/mlx/ffi.ts +0 -512
- package/src/mlx/materialize.ts +0 -23
- package/src/mlx/metal-kernel.ts +0 -159
- package/src/mlx/ops.ts +0 -1021
- package/src/mlx/quantized-rows.ts +0 -36
- package/src/mlx/token-bitmask.ts +0 -52
- package/src/mlx/turboquant-kv-decode.ts +0 -82
- package/src/mlx/turboquant-kv-inverse.ts +0 -122
- package/src/mlx/turboquant-ops.ts +0 -861
- package/src/mlx/turboquant-tables.ts +0 -237
- package/src/mmap.ts +0 -90
- package/src/model/artifact-identity.ts +0 -73
- package/src/model/batched-kv.ts +0 -235
- package/src/model/batched-mask.ts +0 -279
- package/src/model/batched-quant.ts +0 -212
- package/src/model/batched-quantized-kv.ts +0 -144
- package/src/model/batched-rotating-quant.ts +0 -353
- package/src/model/batched-rotating-state.ts +0 -149
- package/src/model/batched-rotating.ts +0 -338
- package/src/model/batched-row-storage.ts +0 -179
- package/src/model/batched-ssm.ts +0 -52
- package/src/model/batched-turboquant-kv.ts +0 -93
- package/src/model/compiled-decode.ts +0 -712
- package/src/model/decoded-kv-donor.ts +0 -12
- package/src/model/delayed-quantized-kv.ts +0 -72
- package/src/model/delayed-rotating-quantized-kv.ts +0 -93
- package/src/model/delayed-turboquant-kv.ts +0 -55
- package/src/model/diffusion-gemma.ts +0 -1026
- package/src/model/factory.ts +0 -189
- package/src/model/fingerprint.ts +0 -33
- package/src/model/flash-attention.ts +0 -694
- package/src/model/full-prefill-padding.ts +0 -56
- package/src/model/full-prefill-row.ts +0 -105
- package/src/model/gemma4-base.ts +0 -2351
- package/src/model/gemma4.ts +0 -1315
- package/src/model/generated/gemma4-12b.ts +0 -429
- package/src/model/generated/gemma4-26b.ts +0 -448
- package/src/model/generated/gemma4-e4b.ts +0 -717
- package/src/model/generated/index.ts +0 -18
- package/src/model/glm52-atlas.ts +0 -511
- package/src/model/glm52-cache.ts +0 -792
- package/src/model/glm52-config.ts +0 -262
- package/src/model/glm52-container.ts +0 -305
- package/src/model/glm52-coupling.ts +0 -388
- package/src/model/glm52-dsa.ts +0 -606
- package/src/model/glm52-expert-layout.ts +0 -253
- package/src/model/glm52-layout.ts +0 -152
- package/src/model/glm52-memory.ts +0 -500
- package/src/model/glm52-mla.ts +0 -792
- package/src/model/glm52-moe.ts +0 -223
- package/src/model/glm52-pilot.ts +0 -537
- package/src/model/glm52-reference.ts +0 -445
- package/src/model/glm52-residency.ts +0 -557
- package/src/model/glm52-resident-weights.ts +0 -135
- package/src/model/glm52-streamed-experts.ts +0 -453
- package/src/model/glm52-streamed-metal.ts +0 -610
- package/src/model/glm52-weights.ts +0 -503
- package/src/model/glm52.ts +0 -1132
- package/src/model/implementation.ts +0 -69
- package/src/model/kv-attention-view.ts +0 -67
- package/src/model/minicpm5.ts +0 -376
- package/src/model/profile.ts +0 -443
- package/src/model/qwen-append.ts +0 -12
- package/src/model/qwen-conv.ts +0 -52
- package/src/model/qwen3-delta-state.ts +0 -74
- package/src/model/qwen3-delta.ts +0 -451
- package/src/model/qwen3-moe.ts +0 -404
- package/src/model/qwen3-mrope.ts +0 -252
- package/src/model/qwen3.ts +0 -277
- package/src/model/qwen3_5-checkpoint.ts +0 -164
- package/src/model/qwen3_5.ts +0 -819
- package/src/model/rotating-kv-donor.ts +0 -35
- package/src/model/rotating-kv-layout.ts +0 -262
- package/src/model/rotating-row-transaction.ts +0 -35
- package/src/model/speculative-rotating-kv.ts +0 -101
- package/src/model/ssm-prefill-padding.ts +0 -39
- package/src/model/support.ts +0 -102
- package/src/model/trellis-balanced-scatter.ts +0 -51
- package/src/model/trellis-linear.ts +0 -672
- package/src/model/trellis-shared-m.ts +0 -151
- package/src/model/trellis-shared-scatter.ts +0 -144
- package/src/model/trellis-splitk-prefill.ts +0 -158
- package/src/model/trellis-tiled-prefill.ts +0 -136
- package/src/model/trellis-vector-expand.ts +0 -62
- package/src/model/trellis-wide-prefill.ts +0 -94
- package/src/model/turboquant-codec.ts +0 -83
- package/src/model/universal/archs.ts +0 -332
- package/src/model/universal/dense.ts +0 -612
- package/src/model/universal/modules.ts +0 -170
- package/src/model/universal/rope.ts +0 -381
- package/src/native/expert_io.c +0 -523
- package/src/native/frame_extract.swift +0 -140
- package/src/native-pack.ts +0 -162
- package/src/pi-provider.ts +0 -53
- package/src/pi-session.ts +0 -81
- package/src/pi-terminal.ts +0 -274
- package/src/pi-web.ts +0 -1932
- package/src/preflight.ts +0 -158
- package/src/prompt-cache.ts +0 -460
- package/src/quantize/allocator.ts +0 -627
- package/src/quantize/atomic-output.ts +0 -38
- package/src/quantize/calibration.ts +0 -181
- package/src/quantize/config-writer.ts +0 -145
- package/src/quantize/index.ts +0 -84
- package/src/quantize/job.ts +0 -162
- package/src/quantize/quantizer.ts +0 -666
- package/src/quantize/rotate.ts +0 -603
- package/src/quantize/safetensors-writer.ts +0 -249
- package/src/quantize/sensitivity.ts +0 -216
- package/src/quantize/trellis.ts +0 -586
- package/src/quantize/weight-transform.ts +0 -292
- package/src/registry.ts +0 -602
- package/src/runtime-config.ts +0 -81
- package/src/safetensors.ts +0 -158
- package/src/sampler.ts +0 -1011
- package/src/serve/admin-routes.ts +0 -184
- package/src/serve/assets/curve-terrain.html +0 -105
- package/src/serve/assets/training-inference-map.html +0 -575
- package/src/serve/aux-routes.ts +0 -132
- package/src/serve/batch-scheduler.ts +0 -8
- package/src/serve/checkpoint-identity.ts +0 -64
- package/src/serve/generation-gateway.ts +0 -496
- package/src/serve/isolate.ts +0 -609
- package/src/serve/kv-budget.ts +0 -14
- package/src/serve/lab-routes.ts +0 -341
- package/src/serve/lane-registry.ts +0 -45
- package/src/serve/model-admin-routes.ts +0 -155
- package/src/serve/model-binding.ts +0 -43
- package/src/serve/model-host.ts +0 -511
- package/src/serve/parent-application.ts +0 -56
- package/src/serve/prompt-response-trace.ts +0 -147
- package/src/serve/response-lease.ts +0 -24
- package/src/serve/session-search.ts +0 -326
- package/src/serve/static-routes.ts +0 -101
- package/src/server.ts +0 -1314
- package/src/spec/assistant-rows.ts +0 -92
- package/src/spec/assistant-source.ts +0 -126
- package/src/spec/cache-identity.ts +0 -22
- package/src/spec/deepspec-rows.ts +0 -29
- package/src/spec/deepspec-source.ts +0 -145
- package/src/spec/dflash-source.ts +0 -149
- package/src/spec/drafter.ts +0 -304
- package/src/spec/dspark/ab-stats.ts +0 -136
- package/src/spec/dspark/calibration.ts +0 -112
- package/src/spec/dspark/data-dflash.ts +0 -156
- package/src/spec/dspark/data.ts +0 -201
- package/src/spec/dspark/deepspec-module.ts +0 -776
- package/src/spec/dspark/generate-dflash.ts +0 -245
- package/src/spec/dspark/generate.ts +0 -290
- package/src/spec/dspark/loader.ts +0 -23
- package/src/spec/dspark/loss.ts +0 -213
- package/src/spec/dspark/module-dflash.ts +0 -560
- package/src/spec/dspark/module.ts +0 -540
- package/src/spec/dspark/quantize-drafter.ts +0 -91
- package/src/spec/dspark/sample.ts +0 -112
- package/src/spec/generate.ts +0 -276
- package/src/spec/glm52-mtp-graph.ts +0 -29
- package/src/spec/glm52-mtp-rows.ts +0 -132
- package/src/spec/glm52-mtp-source.ts +0 -317
- package/src/spec/glm52-mtp-state.ts +0 -33
- package/src/spec/ngram-source.ts +0 -204
- package/src/spec/projected-draft-rows.ts +0 -73
- package/src/spec/qwen-mtp-module.ts +0 -171
- package/src/spec/qwen-mtp-rows.ts +0 -219
- package/src/spec/qwen-mtp-source.ts +0 -383
- package/src/spec/qwen-mtp-state.ts +0 -30
- package/src/spec/serve-loop.ts +0 -585
- package/src/spec/source.ts +0 -215
- package/src/spec/two-model-rows.ts +0 -61
- package/src/spec/two-model.ts +0 -113
- package/src/ssd-cache.ts +0 -372
- package/src/ssd-durability.ts +0 -211
- package/src/tokenizer.ts +0 -335
- package/src/tool-approvals.ts +0 -131
- package/src/tool-call.ts +0 -695
- package/src/train/dataset.ts +0 -484
- package/src/train/diffusion-lora.ts +0 -216
- package/src/train/export.ts +0 -32
- package/src/train/flash-cce.ts +0 -1234
- package/src/train/forward.ts +0 -266
- package/src/train/fuse.ts +0 -243
- package/src/train/index.ts +0 -46
- package/src/train/job.ts +0 -205
- package/src/train/lora-params.ts +0 -254
- package/src/train/loss.ts +0 -1591
- package/src/train/merge.ts +0 -195
- package/src/train/optimizer.ts +0 -183
- package/src/train/prefix-shared.ts +0 -490
- package/src/train/rank.ts +0 -137
- package/src/train/segmented.ts +0 -1828
- package/src/train/steel-qmm-header.ts +0 -1461
- package/src/train/trainer.ts +0 -1221
- package/src/train/watch.ts +0 -542
- package/src/vision/diffusion-vision.ts +0 -620
- package/src/vision/embedder.ts +0 -168
- package/src/vision/preprocess.ts +0 -247
- package/src/vision/prompt.ts +0 -373
- package/src/vision/qwen3vl-preprocess.ts +0 -409
- package/src/vision/qwen3vl-prompt.ts +0 -196
- package/src/vision/qwen3vl-tower.ts +0 -555
- package/src/vision/siglip.ts +0 -599
- package/src/vision/video-frames.ts +0 -148
- package/src/web/app.html +0 -2143
- package/src/web/app.js +0 -5703
- package/src/web/skills/memory/SKILL.md +0 -120
- package/src/web/skills/web-research/SKILL.md +0 -46
- package/src/web/skills.ts +0 -97
- package/src/web/src/adapters-panel.ts +0 -288
- package/src/web/src/assistant.ts +0 -451
- package/src/web/src/chat.ts +0 -1057
- package/src/web/src/composer.ts +0 -959
- package/src/web/src/dataset.ts +0 -132
- package/src/web/src/finetune.ts +0 -267
- package/src/web/src/hub.ts +0 -428
- package/src/web/src/main.ts +0 -98
- package/src/web/src/markdown.ts +0 -0
- package/src/web/src/memory-panel.ts +0 -572
- package/src/web/src/model-picker.ts +0 -159
- package/src/web/src/palette.ts +0 -346
- package/src/web/src/protocol.ts +0 -26
- package/src/web/src/quantize.ts +0 -167
- package/src/web/src/rag.ts +0 -274
- package/src/web/src/sessions.ts +0 -469
- package/src/web/src/shell.ts +0 -804
- package/src/web/src/ui-catalog.ts +0 -78
- package/src/web/vendor/README.md +0 -61
- package/src/web-assets.ts +0 -47
- package/src/web-tools.ts +0 -478
- package/src/weights.ts +0 -255
- /package/src/{serve → server}/completion-sink.ts +0 -0
- /package/src/web/{icon.svg → public/icon.svg} +0 -0
- /package/src/web/{manifest.webmanifest → public/manifest.webmanifest} +0 -0
- /package/src/web/{vendor → public/vendor}/hljs-LICENSE +0 -0
package/src/memory/resolve.ts
DELETED
|
@@ -1,307 +0,0 @@
|
|
|
1
|
-
// mlx-bun memory — surface-variant RESOLUTION (the dedup fix for P5-T3/T4).
|
|
2
|
-
//
|
|
3
|
-
// canonicalize() in entity.ts is a *pure string normal form*: casefold, collapse
|
|
4
|
-
// whitespace, strip articles/possessives. That is necessary but NOT sufficient —
|
|
5
|
-
// the self-check turned the spacing, brand-only, and possessive surfaces of ONE
|
|
6
|
-
// product name into several different stems:
|
|
7
|
-
// - "<name>" vs "<na me>" (internal spacing the normal form keeps)
|
|
8
|
-
// - "<brand>" / "<brand> <model>" (brand-only and compositional surfaces that
|
|
9
|
-
// no purely-lexical normalizer can fold onto one canonical)
|
|
10
|
-
//
|
|
11
|
-
// The resolver adds three layers on top of canonicalize():
|
|
12
|
-
// 1. SQUEEZE — strip ALL non-alphanumerics so "<na me>" / "<name>" / "<nam e>"
|
|
13
|
-
// collapse to one key deterministically (fixes the spacing split).
|
|
14
|
-
// 2. ALIAS SEED — a known-entity index (entities + entity_aliases in the store,
|
|
15
|
-
// seeded from goldens/dreaming-entities-gold.json) so brand-only / nickname
|
|
16
|
-
// surfaces (a "<the brand>", a "<the role>") resolve by exact alias hit.
|
|
17
|
-
// 3. TOKEN-SUBSET FUZZY — a conservative compositional match: a surface folds
|
|
18
|
-
// onto an entity iff its content tokens are a SUBSET of that entity's tokens
|
|
19
|
-
// AND the overlap carries at least one DISTINCTIVE token (a token unique to
|
|
20
|
-
// that entity among all known entities). This merges "<my brand model>" → the
|
|
21
|
-
// entity via the distinctive model token while REFUSING to merge
|
|
22
|
-
// "<brand A>" into "<brand B>" (a shared brand token is not distinctive;
|
|
23
|
-
// the model-number tokens differ) — the over-merge guard.
|
|
24
|
-
//
|
|
25
|
-
// Resolution is conservative by construction: when in doubt it MINTS a new
|
|
26
|
-
// canonical rather than fusing two distinct things. ROUTE (route.ts) escalates
|
|
27
|
-
// the genuine misses to the model for binary disambiguation.
|
|
28
|
-
|
|
29
|
-
import { readFileSync } from "node:fs";
|
|
30
|
-
import { join } from "node:path";
|
|
31
|
-
|
|
32
|
-
import { canonicalize } from "./entity";
|
|
33
|
-
import type { MemoryStore } from "./db";
|
|
34
|
-
|
|
35
|
-
/** Tokens that carry no disambiguating signal on their own (articles, glue, and
|
|
36
|
-
* generic gear nouns shared across many entities). Used only for the fuzzy
|
|
37
|
-
* token-subset layer — never for the exact alias index. */
|
|
38
|
-
const STOP = new Set([
|
|
39
|
-
"the", "a", "an", "and", "or", "of", "to", "for", "in", "on", "with", "is",
|
|
40
|
-
"as", "my", "your", "his", "her", "its", "our", "their",
|
|
41
|
-
]);
|
|
42
|
-
|
|
43
|
-
/** Strip ALL non-alphanumerics and casefold: the spacing/punct-insensitive key.
|
|
44
|
-
* A name's spacing and punctuation variants all collapse here. */
|
|
45
|
-
export function squeeze(surface: string): string {
|
|
46
|
-
return surface.toLowerCase().replace(/[^a-z0-9]+/g, "");
|
|
47
|
-
}
|
|
48
|
-
|
|
49
|
-
/** Content tokens of a surface (casefold, split on non-alphanumerics, drop
|
|
50
|
-
* stopwords and single characters). Order-insensitive set. */
|
|
51
|
-
export function contentTokens(surface: string): Set<string> {
|
|
52
|
-
const out = new Set<string>();
|
|
53
|
-
for (const t of surface.toLowerCase().split(/[^a-z0-9]+/)) {
|
|
54
|
-
if (t.length > 1 && !STOP.has(t)) out.add(t);
|
|
55
|
-
}
|
|
56
|
-
return out;
|
|
57
|
-
}
|
|
58
|
-
|
|
59
|
-
/** How a surface resolved (for instrumentation + the deterministic gates). */
|
|
60
|
-
export type MatchKind = "stem" | "squeeze" | "alias" | "fuzzy" | "new";
|
|
61
|
-
|
|
62
|
-
export interface ResolveResult {
|
|
63
|
-
/** The canonical entity name the surface resolved to (or was minted as). */
|
|
64
|
-
name: string;
|
|
65
|
-
/** Which layer produced the resolution. "new" ⇒ a canonical was minted. */
|
|
66
|
-
matched: MatchKind;
|
|
67
|
-
/** True iff this call minted a NEW canonical (no existing entity matched). */
|
|
68
|
-
created: boolean;
|
|
69
|
-
}
|
|
70
|
-
|
|
71
|
-
interface Seed {
|
|
72
|
-
canonical: string;
|
|
73
|
-
aliases: string[];
|
|
74
|
-
}
|
|
75
|
-
|
|
76
|
-
/**
|
|
77
|
-
* In-memory variant resolver over a KNOWN-ENTITY set. Indexes each canonical by
|
|
78
|
-
* its normal-form stem, its squeeze key, and its content tokens, then resolves a
|
|
79
|
-
* surface through stem → squeeze → alias → token-subset-fuzzy, minting a new
|
|
80
|
-
* canonical only when every layer misses.
|
|
81
|
-
*
|
|
82
|
-
* Construct from a seed list, from a {@link MemoryStore} (entities +
|
|
83
|
-
* entity_aliases), or from the curated gold — and compose them (store first,
|
|
84
|
-
* gold to backfill notable brand-only aliases).
|
|
85
|
-
*/
|
|
86
|
-
export class EntityResolver {
|
|
87
|
-
/** normal-form stem → canonical. */
|
|
88
|
-
private byStem = new Map<string, string>();
|
|
89
|
-
/** squeeze key → canonical. */
|
|
90
|
-
private bySqueeze = new Map<string, string>();
|
|
91
|
-
/** canonical → union of content tokens across its name + aliases. */
|
|
92
|
-
private tokensOf = new Map<string, Set<string>>();
|
|
93
|
-
/** content token → set of canonicals carrying it (distinctiveness check). */
|
|
94
|
-
private tokenOwners = new Map<string, Set<string>>();
|
|
95
|
-
|
|
96
|
-
constructor(seeds: Seed[] = []) {
|
|
97
|
-
for (const s of seeds) this.register(s.canonical, s.aliases);
|
|
98
|
-
}
|
|
99
|
-
|
|
100
|
-
/** All known canonical names (registration order is not guaranteed). */
|
|
101
|
-
canonicals(): string[] {
|
|
102
|
-
return [...this.tokensOf.keys()];
|
|
103
|
-
}
|
|
104
|
-
|
|
105
|
-
/** Is `name` an already-known canonical entity? */
|
|
106
|
-
has(name: string): boolean {
|
|
107
|
-
return this.tokensOf.has(name);
|
|
108
|
-
}
|
|
109
|
-
|
|
110
|
-
/** Register a canonical and its aliases into every index. Idempotent; adding
|
|
111
|
-
* aliases to an existing canonical merges them. The canonical's own surface is
|
|
112
|
-
* always indexed as an alias of itself. */
|
|
113
|
-
register(canonical: string, aliases: string[] = []): void {
|
|
114
|
-
if (!this.tokensOf.has(canonical)) this.tokensOf.set(canonical, new Set());
|
|
115
|
-
const tokset = this.tokensOf.get(canonical)!;
|
|
116
|
-
for (const surface of [canonical, ...aliases]) {
|
|
117
|
-
const stem = canonicalize(surface);
|
|
118
|
-
const sq = squeeze(surface);
|
|
119
|
-
if (stem && !this.byStem.has(stem)) this.byStem.set(stem, canonical);
|
|
120
|
-
if (sq && !this.bySqueeze.has(sq)) this.bySqueeze.set(sq, canonical);
|
|
121
|
-
for (const t of contentTokens(surface)) {
|
|
122
|
-
tokset.add(t);
|
|
123
|
-
let owners = this.tokenOwners.get(t);
|
|
124
|
-
if (!owners) this.tokenOwners.set(t, (owners = new Set()));
|
|
125
|
-
owners.add(canonical);
|
|
126
|
-
}
|
|
127
|
-
}
|
|
128
|
-
}
|
|
129
|
-
|
|
130
|
-
/** A token is DISTINCTIVE iff exactly one known canonical carries it — the
|
|
131
|
-
* signal the fuzzy layer requires to merge a compositional surface. */
|
|
132
|
-
private distinctive(token: string): string | null {
|
|
133
|
-
const owners = this.tokenOwners.get(token);
|
|
134
|
-
if (!owners || owners.size !== 1) return null;
|
|
135
|
-
return [...owners][0]!;
|
|
136
|
-
}
|
|
137
|
-
|
|
138
|
-
/**
|
|
139
|
-
* Conservative token-subset fuzzy match: the surface's content tokens must be a
|
|
140
|
-
* non-empty SUBSET of some entity's tokens, and at least one of them must be
|
|
141
|
-
* distinctive to that same entity. Returns the canonical, or null if no
|
|
142
|
-
* unambiguous owner (zero or conflicting distinctive owners ⇒ refuse to merge).
|
|
143
|
-
*/
|
|
144
|
-
private fuzzy(surface: string): string | null {
|
|
145
|
-
const toks = contentTokens(surface);
|
|
146
|
-
if (toks.size === 0) return null;
|
|
147
|
-
// The distinctive tokens present in the surface, and who owns them.
|
|
148
|
-
const owners = new Set<string>();
|
|
149
|
-
for (const t of toks) {
|
|
150
|
-
const owner = this.distinctive(t);
|
|
151
|
-
if (owner) owners.add(owner);
|
|
152
|
-
}
|
|
153
|
-
if (owners.size !== 1) return null; // none, or a tie across entities → no merge
|
|
154
|
-
const cand = [...owners][0]!;
|
|
155
|
-
// Subset guard: every surface token must belong to the candidate, else the
|
|
156
|
-
// surface names something MORE than (a variant of) the candidate.
|
|
157
|
-
const ets = this.tokensOf.get(cand)!;
|
|
158
|
-
for (const t of toks) if (!ets.has(t)) return null;
|
|
159
|
-
return cand;
|
|
160
|
-
}
|
|
161
|
-
|
|
162
|
-
/** Resolve a surface to an EXISTING canonical, or null if every layer misses.
|
|
163
|
-
* Pure: never mutates the index. */
|
|
164
|
-
match(surface: string): { name: string; matched: MatchKind } | null {
|
|
165
|
-
const stem = canonicalize(surface);
|
|
166
|
-
if (stem) {
|
|
167
|
-
const byStem = this.byStem.get(stem);
|
|
168
|
-
if (byStem) return { name: byStem, matched: "stem" };
|
|
169
|
-
}
|
|
170
|
-
const sq = squeeze(surface);
|
|
171
|
-
if (sq) {
|
|
172
|
-
const bySq = this.bySqueeze.get(sq);
|
|
173
|
-
if (bySq) return { name: bySq, matched: "squeeze" };
|
|
174
|
-
}
|
|
175
|
-
const fz = this.fuzzy(surface);
|
|
176
|
-
if (fz) return { name: fz, matched: "fuzzy" };
|
|
177
|
-
return null;
|
|
178
|
-
}
|
|
179
|
-
|
|
180
|
-
/**
|
|
181
|
-
* Resolve a surface, MINTING a new canonical when nothing matches. The minted
|
|
182
|
-
* canonical is the cleaned surface; it is registered so later variants of the
|
|
183
|
-
* same thing fold onto it. Brand-only nicknames still need a gold/store seed —
|
|
184
|
-
* fuzzy cannot invent a "<the brand>" ≡ "<brand model>" link from lexis alone.
|
|
185
|
-
*/
|
|
186
|
-
resolve(surface: string, mintName?: string): ResolveResult {
|
|
187
|
-
const hit = this.match(surface);
|
|
188
|
-
if (hit) return { name: hit.name, matched: hit.matched, created: false };
|
|
189
|
-
const canonical = (mintName ?? surface).trim();
|
|
190
|
-
this.register(canonical, surface === canonical ? [] : [surface]);
|
|
191
|
-
return { name: canonical, matched: "new", created: true };
|
|
192
|
-
}
|
|
193
|
-
|
|
194
|
-
/** Register one canonical per stem with its alias surfaces, grouped from a
|
|
195
|
-
* normalized `alias → stem` map (the read-index's `aliasToStem`). The stem is
|
|
196
|
-
* the canonical; its surfaces become aliases. */
|
|
197
|
-
static fromAliasMap(aliasToStem: Iterable<readonly [string, string]>): EntityResolver {
|
|
198
|
-
const byStem = new Map<string, string[]>();
|
|
199
|
-
for (const [alias, stem] of aliasToStem) {
|
|
200
|
-
let list = byStem.get(stem);
|
|
201
|
-
if (!list) byStem.set(stem, (list = []));
|
|
202
|
-
list.push(alias);
|
|
203
|
-
}
|
|
204
|
-
const r = new EntityResolver();
|
|
205
|
-
for (const [stem, aliases] of byStem) r.register(stem, aliases);
|
|
206
|
-
return r;
|
|
207
|
-
}
|
|
208
|
-
|
|
209
|
-
/** Load the known-entity index from a store's entities + entity_aliases. */
|
|
210
|
-
static fromStore(store: MemoryStore): EntityResolver {
|
|
211
|
-
const r = new EntityResolver();
|
|
212
|
-
const ents = store.db.query("SELECT name FROM entities").all() as { name: string }[];
|
|
213
|
-
for (const e of ents) r.register(e.name, []);
|
|
214
|
-
const aliases = store.db
|
|
215
|
-
.query("SELECT alias, entity_name FROM entity_aliases")
|
|
216
|
-
.all() as { alias: string; entity_name: string }[];
|
|
217
|
-
for (const a of aliases) {
|
|
218
|
-
if (!r.has(a.entity_name)) r.register(a.entity_name, [a.alias]);
|
|
219
|
-
else r.register(a.entity_name, [a.alias]);
|
|
220
|
-
}
|
|
221
|
-
return r;
|
|
222
|
-
}
|
|
223
|
-
}
|
|
224
|
-
|
|
225
|
-
// ---- near-name resolution (sub-concept → its home article) -----------------
|
|
226
|
-
|
|
227
|
-
/** Generic "kind-of-thing" suffix words a discussed sub-concept drops when it is
|
|
228
|
-
* the SAME subject as an existing article (a query that is the article title
|
|
229
|
-
* minus a generic tail — "<subject>" for "<subject> Theory"). They carry no
|
|
230
|
-
* disambiguating signal alone, so a query made ONLY of these never redirects. */
|
|
231
|
-
const GENERIC_SUFFIX = new Set([
|
|
232
|
-
"theory", "framework", "system", "model", "method", "methodology", "approach",
|
|
233
|
-
"principle", "principles", "technique", "concept", "paradigm", "effect", "law",
|
|
234
|
-
"hypothesis", "problem", "process", "pattern", "practice",
|
|
235
|
-
]);
|
|
236
|
-
|
|
237
|
-
/**
|
|
238
|
-
* NEAR-NAME match: resolve a surface to an EXISTING article when it is clearly the
|
|
239
|
-
* SAME subject — the conservative token-subset fuzzy (a query whose content tokens
|
|
240
|
-
* are a distinctive subset of an article's, e.g. the title minus a generic suffix).
|
|
241
|
-
* Reuses {@link EntityResolver.match} (stem → squeeze → token-subset fuzzy) but
|
|
242
|
-
* REFUSES a query made only of generic suffix words (just "Theory"/"Framework"),
|
|
243
|
-
* which would over-merge onto an arbitrary article. Returns the canonical (== the
|
|
244
|
-
* article stem the resolver was seeded with), or null. Pure.
|
|
245
|
-
*/
|
|
246
|
-
export function nearNameMatch(resolver: EntityResolver, surface: string): string | null {
|
|
247
|
-
const toks = contentTokens(surface);
|
|
248
|
-
if (toks.size === 0) return null;
|
|
249
|
-
let hasContentToken = false;
|
|
250
|
-
for (const t of toks) {
|
|
251
|
-
if (!GENERIC_SUFFIX.has(t)) {
|
|
252
|
-
hasContentToken = true;
|
|
253
|
-
break;
|
|
254
|
-
}
|
|
255
|
-
}
|
|
256
|
-
if (!hasContentToken) return null; // a purely-generic query must not redirect
|
|
257
|
-
const hit = resolver.match(surface);
|
|
258
|
-
return hit ? hit.name : null;
|
|
259
|
-
}
|
|
260
|
-
|
|
261
|
-
// ---- gold seed -------------------------------------------------------------
|
|
262
|
-
|
|
263
|
-
export interface VariantGroup {
|
|
264
|
-
canonical: string;
|
|
265
|
-
kind: string;
|
|
266
|
-
domain: string;
|
|
267
|
-
variants: string[];
|
|
268
|
-
}
|
|
269
|
-
export interface NotableEntity {
|
|
270
|
-
name: string;
|
|
271
|
-
kind: string;
|
|
272
|
-
domain: string;
|
|
273
|
-
aliases: string[];
|
|
274
|
-
whyNotable: string;
|
|
275
|
-
}
|
|
276
|
-
export interface DreamingGold {
|
|
277
|
-
variantGroups: VariantGroup[];
|
|
278
|
-
notableEntities: NotableEntity[];
|
|
279
|
-
}
|
|
280
|
-
|
|
281
|
-
const GOLD_PATH = join(import.meta.dir, "..", "..", "goldens", "dreaming-entities-gold.json");
|
|
282
|
-
|
|
283
|
-
/** Read the curated entity gold (variantGroups + notableEntities). */
|
|
284
|
-
export function loadDreamingGold(path: string = GOLD_PATH): DreamingGold {
|
|
285
|
-
return JSON.parse(readFileSync(path, "utf8")) as DreamingGold;
|
|
286
|
-
}
|
|
287
|
-
|
|
288
|
-
/** Seeds derived from the gold: every variant group AND every notable entity
|
|
289
|
-
* becomes a (canonical, aliases) seed. Notable entries that share a canonical
|
|
290
|
-
* with a variant group merge their aliases. */
|
|
291
|
-
export function goldSeeds(gold: DreamingGold = loadDreamingGold()): Seed[] {
|
|
292
|
-
const byName = new Map<string, Set<string>>();
|
|
293
|
-
const add = (canonical: string, aliases: string[]): void => {
|
|
294
|
-
let set = byName.get(canonical);
|
|
295
|
-
if (!set) byName.set(canonical, (set = new Set()));
|
|
296
|
-
for (const a of aliases) set.add(a);
|
|
297
|
-
};
|
|
298
|
-
for (const g of gold.variantGroups) add(g.canonical, g.variants);
|
|
299
|
-
for (const n of gold.notableEntities) add(n.name, n.aliases);
|
|
300
|
-
return [...byName].map(([canonical, aliases]) => ({ canonical, aliases: [...aliases] }));
|
|
301
|
-
}
|
|
302
|
-
|
|
303
|
-
/** A resolver pre-seeded with the curated gold — the production seed for the
|
|
304
|
-
* notable brand-only aliases the model/lexis cannot derive on its own. */
|
|
305
|
-
export function goldResolver(gold?: DreamingGold): EntityResolver {
|
|
306
|
-
return new EntityResolver(goldSeeds(gold));
|
|
307
|
-
}
|
package/src/memory/rest.ts
DELETED
|
@@ -1,245 +0,0 @@
|
|
|
1
|
-
// Memory REST wrappers — thin loopback JSON routes over src/memory/vault.ts
|
|
2
|
-
// for the web chat's Memory panel (docs/design/web-chat-redesign.md §5.5,
|
|
3
|
-
// §9 Phase 2). These are HTTP-only conveniences for the frontend: the
|
|
4
|
-
// agent-tool surface (src/memory/tools.ts) stays read-only and unchanged,
|
|
5
|
-
// and every route here is itself read-only except /init, which delegates to
|
|
6
|
-
// the same `setupVault` the CLI's `mlx-bun memory init` calls (never
|
|
7
|
-
// duplicated logic).
|
|
8
|
-
//
|
|
9
|
-
// Every handler returns a Response and never throws — a missing vault or a
|
|
10
|
-
// bad param degrades to `{ ok: false, error }` (never a 500), matching the
|
|
11
|
-
// rest of server.ts's /api/* conventions. Route dispatch (matching
|
|
12
|
-
// url.pathname + method) lives in server.ts; this file is the pure handler
|
|
13
|
-
// bodies so they're unit-testable without booting a model context (see
|
|
14
|
-
// tests/using/memory-rest.test.ts).
|
|
15
|
-
|
|
16
|
-
import { realpath } from "node:fs/promises";
|
|
17
|
-
import { tmpdir } from "node:os";
|
|
18
|
-
import { dirname, resolve, sep } from "node:path";
|
|
19
|
-
import {
|
|
20
|
-
vaultRoot, vaultStatus, listArticles, listReferenceDocs, readArticle,
|
|
21
|
-
searchArticles, getArticleLinks, resolveArticleRelPath, articleHistory, articleDiff,
|
|
22
|
-
setupVault, type VaultStatus,
|
|
23
|
-
} from "./vault";
|
|
24
|
-
import { parseInfobox, parseLead, parseSeriesBanner, articleStructure } from "./article";
|
|
25
|
-
|
|
26
|
-
function jsonOk<T extends object>(body: T, init?: ResponseInit): Response {
|
|
27
|
-
return Response.json({ ok: true, ...body }, init);
|
|
28
|
-
}
|
|
29
|
-
|
|
30
|
-
function jsonErr(error: string, status = 400): Response {
|
|
31
|
-
return Response.json({ ok: false, error }, { status });
|
|
32
|
-
}
|
|
33
|
-
|
|
34
|
-
/** `{ ok:false, enabled:false }` shape shared by every route when no vault
|
|
35
|
-
* exists yet — never a throw, never a 404: the frontend uses this to
|
|
36
|
-
* render the first-run consent card instead of an error toast. */
|
|
37
|
-
function noVault(root: string): Response {
|
|
38
|
-
return Response.json({ ok: false, enabled: false, error: "no memory vault yet", root });
|
|
39
|
-
}
|
|
40
|
-
|
|
41
|
-
// ---- GET /api/memory/status -------------------------------------------
|
|
42
|
-
|
|
43
|
-
export async function handleMemoryStatus(): Promise<Response> {
|
|
44
|
-
const st = await vaultStatus();
|
|
45
|
-
if (!st.exists) return noVault(st.root);
|
|
46
|
-
return jsonOk({ enabled: true, status: st satisfies VaultStatus });
|
|
47
|
-
}
|
|
48
|
-
|
|
49
|
-
// ---- GET /api/memory/list ----------------------------------------------
|
|
50
|
-
|
|
51
|
-
export async function handleMemoryList(): Promise<Response> {
|
|
52
|
-
const root = vaultRoot();
|
|
53
|
-
const st = await vaultStatus(root);
|
|
54
|
-
if (!st.exists) return noVault(root);
|
|
55
|
-
const [articles, reference] = await Promise.all([listArticles(root), listReferenceDocs(root)]);
|
|
56
|
-
return jsonOk({ articles, reference });
|
|
57
|
-
}
|
|
58
|
-
|
|
59
|
-
// ---- GET /api/memory/search?q=&scope= -----------------------------------
|
|
60
|
-
|
|
61
|
-
export async function handleMemorySearch(url: URL): Promise<Response> {
|
|
62
|
-
const root = vaultRoot();
|
|
63
|
-
const st = await vaultStatus(root);
|
|
64
|
-
if (!st.exists) return noVault(root);
|
|
65
|
-
const q = (url.searchParams.get("q") ?? "").trim();
|
|
66
|
-
if (!q) return jsonErr("q is required");
|
|
67
|
-
const scopeParam = url.searchParams.get("scope");
|
|
68
|
-
const scope = scopeParam === "articles" || scopeParam === "reference" ? scopeParam : "all";
|
|
69
|
-
const limitParam = Number(url.searchParams.get("limit") ?? "50");
|
|
70
|
-
const limit = Number.isFinite(limitParam) && limitParam > 0 ? Math.min(limitParam, 200) : 50;
|
|
71
|
-
try {
|
|
72
|
-
const { summaries, hits } = await searchArticles(root, q, { limit, scope });
|
|
73
|
-
return jsonOk({ summaries, hits });
|
|
74
|
-
} catch (e) {
|
|
75
|
-
return jsonErr((e as Error).message);
|
|
76
|
-
}
|
|
77
|
-
}
|
|
78
|
-
|
|
79
|
-
// ---- GET /api/memory/article?name= --------------------------------------
|
|
80
|
-
//
|
|
81
|
-
// Rendered source + metadata: the article's raw Markdown plus the
|
|
82
|
-
// deterministic structure the panel needs to render it without re-parsing
|
|
83
|
-
// client-side (infobox, lead, series banner, section skeleton).
|
|
84
|
-
|
|
85
|
-
export async function handleMemoryArticle(url: URL): Promise<Response> {
|
|
86
|
-
const root = vaultRoot();
|
|
87
|
-
const st = await vaultStatus(root);
|
|
88
|
-
if (!st.exists) return noVault(root);
|
|
89
|
-
const name = (url.searchParams.get("name") ?? "").trim();
|
|
90
|
-
if (!name) return jsonErr("name is required");
|
|
91
|
-
try {
|
|
92
|
-
const { path, content } = await readArticle(root, name);
|
|
93
|
-
return jsonOk({
|
|
94
|
-
name,
|
|
95
|
-
path,
|
|
96
|
-
content,
|
|
97
|
-
infobox: parseInfobox(content),
|
|
98
|
-
lead: parseLead(content),
|
|
99
|
-
series: parseSeriesBanner(content),
|
|
100
|
-
structure: articleStructure(content),
|
|
101
|
-
});
|
|
102
|
-
} catch (e) {
|
|
103
|
-
return jsonErr((e as Error).message, 404);
|
|
104
|
-
}
|
|
105
|
-
}
|
|
106
|
-
|
|
107
|
-
// ---- GET /api/memory/links?name= -----------------------------------------
|
|
108
|
-
|
|
109
|
-
export async function handleMemoryLinks(url: URL): Promise<Response> {
|
|
110
|
-
const root = vaultRoot();
|
|
111
|
-
const st = await vaultStatus(root);
|
|
112
|
-
if (!st.exists) return noVault(root);
|
|
113
|
-
const name = (url.searchParams.get("name") ?? "").trim();
|
|
114
|
-
if (!name) return jsonErr("name is required");
|
|
115
|
-
try {
|
|
116
|
-
const links = await getArticleLinks(root, name);
|
|
117
|
-
return jsonOk({ name, ...links });
|
|
118
|
-
} catch (e) {
|
|
119
|
-
return jsonErr((e as Error).message, 404);
|
|
120
|
-
}
|
|
121
|
-
}
|
|
122
|
-
|
|
123
|
-
// ---- GET /api/memory/history?name= ---------------------------------------
|
|
124
|
-
|
|
125
|
-
export async function handleMemoryHistory(url: URL): Promise<Response> {
|
|
126
|
-
const root = vaultRoot();
|
|
127
|
-
const st = await vaultStatus(root);
|
|
128
|
-
if (!st.exists) return noVault(root);
|
|
129
|
-
const name = (url.searchParams.get("name") ?? "").trim();
|
|
130
|
-
if (!name) return jsonErr("name is required");
|
|
131
|
-
const limitParam = Number(url.searchParams.get("limit") ?? "50");
|
|
132
|
-
const limit = Number.isFinite(limitParam) && limitParam > 0 ? Math.min(limitParam, 200) : 50;
|
|
133
|
-
try {
|
|
134
|
-
const relPath = await resolveArticleRelPath(root, name); // throws if not found — validates `name`
|
|
135
|
-
const entries = await articleHistory(root, relPath, limit);
|
|
136
|
-
return jsonOk({ name, isGitRepo: st.isGitRepo, entries });
|
|
137
|
-
} catch (e) {
|
|
138
|
-
return jsonErr((e as Error).message, 404);
|
|
139
|
-
}
|
|
140
|
-
}
|
|
141
|
-
|
|
142
|
-
// ---- GET /api/memory/diff?name=&rev= -------------------------------------
|
|
143
|
-
|
|
144
|
-
/** Full 40-hex or an abbreviated (≥4 hex) commit hash — never a ref
|
|
145
|
-
* expression (`HEAD~1`, `main^`, `refs/...`), which could otherwise be
|
|
146
|
-
* abused to walk outside the intended commit or (in principle) smuggle
|
|
147
|
-
* option-like text into the argv. `rev` is passed to Bun.spawn as its own
|
|
148
|
-
* argv element (never through a shell), but this check is the defense
|
|
149
|
-
* against a caller trying to pass anything other than a bare commit id. */
|
|
150
|
-
const REV_RE = /^[0-9a-f]{4,40}$/;
|
|
151
|
-
|
|
152
|
-
export async function handleMemoryDiff(url: URL): Promise<Response> {
|
|
153
|
-
const root = vaultRoot();
|
|
154
|
-
const st = await vaultStatus(root);
|
|
155
|
-
if (!st.exists) return noVault(root);
|
|
156
|
-
const name = (url.searchParams.get("name") ?? "").trim();
|
|
157
|
-
const rev = (url.searchParams.get("rev") ?? "").trim();
|
|
158
|
-
if (!name) return jsonErr("name is required");
|
|
159
|
-
if (!rev) return jsonErr("rev is required");
|
|
160
|
-
if (!REV_RE.test(rev)) return jsonErr("rev must be a commit hash (4-40 hex chars)");
|
|
161
|
-
try {
|
|
162
|
-
const relPath = await resolveArticleRelPath(root, name); // throws if not found — validates `name`
|
|
163
|
-
const diff = await articleDiff(root, rev, relPath);
|
|
164
|
-
if (diff === null) return jsonErr("commit not found or no diff for this article at that revision", 404);
|
|
165
|
-
return jsonOk({ name, rev, diff });
|
|
166
|
-
} catch (e) {
|
|
167
|
-
return jsonErr((e as Error).message, 404);
|
|
168
|
-
}
|
|
169
|
-
}
|
|
170
|
-
|
|
171
|
-
// ---- POST /api/memory/init { path?: string } -----------------------------
|
|
172
|
-
//
|
|
173
|
-
// The consent-card backend: wraps the exact same `setupVault` the CLI's
|
|
174
|
-
// `mlx-bun memory init` calls (src/cli.ts, `case "memory":` → `sub === "init"`)
|
|
175
|
-
// — same idempotent create-dirs/README/Meta-pages/git-init behavior, minus
|
|
176
|
-
// the CLI's interactive extras (seed-from-existing-vault prompt, nightly
|
|
177
|
-
// schedule prompt), which stay CLI-only/TTY-gated and are out of scope here.
|
|
178
|
-
//
|
|
179
|
-
// Unlike the CLI (which only ever calls setupVault(vaultRoot()) — a
|
|
180
|
-
// user-supplied path there is a read-only *seed source*, never the
|
|
181
|
-
// setupVault target), this route takes `path` straight from an untrusted
|
|
182
|
-
// loopback caller. setupVault() mkdir -p's subdirectories into it and, if
|
|
183
|
-
// it isn't already a git repo, runs `git init && git add . && git commit`
|
|
184
|
-
// there — pointed at an arbitrary directory that's a destructive foot-gun
|
|
185
|
-
// (force-commits whatever's already there into a fresh history). There is
|
|
186
|
-
// no product reason for the web panel to redirect the vault root outside
|
|
187
|
-
// the machine's own tmp/vault trees, so `path` is confined to those before
|
|
188
|
-
// it ever reaches setupVault; anything else 400s instead of writing files.
|
|
189
|
-
|
|
190
|
-
/** realpath() of the deepest existing ancestor of `path`, with the
|
|
191
|
-
* non-existent tail (e.g. a not-yet-created vault dir) re-appended
|
|
192
|
-
* literally. Needed because macOS resolves `/var/folders/...` (os.tmpdir())
|
|
193
|
-
* through a `/private` symlink: a brand-new mkdtemp path under it won't
|
|
194
|
-
* literal-prefix-match the realpath'd tmpdir unless we walk up to a real
|
|
195
|
-
* segment the same way. */
|
|
196
|
-
async function realpathNearest(path: string): Promise<string> {
|
|
197
|
-
let cur = resolve(path);
|
|
198
|
-
const tail: string[] = [];
|
|
199
|
-
while (true) {
|
|
200
|
-
try {
|
|
201
|
-
return [await realpath(cur), ...tail.reverse()].join(sep);
|
|
202
|
-
} catch {
|
|
203
|
-
const parent = dirname(cur);
|
|
204
|
-
if (parent === cur) return resolve(path); // hit the fs root without finding anything real
|
|
205
|
-
tail.push(cur.slice(parent.length + 1));
|
|
206
|
-
cur = parent;
|
|
207
|
-
}
|
|
208
|
-
}
|
|
209
|
-
}
|
|
210
|
-
|
|
211
|
-
/** Is `candidate` equal to or a descendant of `ancestor`, after resolving
|
|
212
|
-
* symlinks on both sides? */
|
|
213
|
-
async function isUnder(candidate: string, ancestor: string): Promise<boolean> {
|
|
214
|
-
const realAncestor = await realpathNearest(ancestor);
|
|
215
|
-
const realCandidate = await realpathNearest(candidate);
|
|
216
|
-
return realCandidate === realAncestor || realCandidate.startsWith(realAncestor + sep);
|
|
217
|
-
}
|
|
218
|
-
|
|
219
|
-
/** Confine an init target to the machine's own vault/tmp trees — the only
|
|
220
|
-
* legitimate destinations for this route (default vault root, or a tmp dir
|
|
221
|
-
* for tests). Rejects anything else rather than handing setupVault() an
|
|
222
|
-
* arbitrary, possibly pre-existing directory to mkdir/git-init into. */
|
|
223
|
-
async function isAllowedVaultTarget(path: string): Promise<boolean> {
|
|
224
|
-
return (await isUnder(path, vaultRoot())) || (await isUnder(path, tmpdir()));
|
|
225
|
-
}
|
|
226
|
-
|
|
227
|
-
export async function handleMemoryInit(request: Request): Promise<Response> {
|
|
228
|
-
let body: { path?: string } = {};
|
|
229
|
-
try {
|
|
230
|
-
body = (await request.json()) as { path?: string };
|
|
231
|
-
} catch {
|
|
232
|
-
// empty body is fine — defaults to vaultRoot()
|
|
233
|
-
}
|
|
234
|
-
const root = body.path && body.path.trim() ? body.path.trim() : vaultRoot();
|
|
235
|
-
if (!(await isAllowedVaultTarget(root))) {
|
|
236
|
-
return jsonErr("path must be under the memory vault root or a temp directory", 400);
|
|
237
|
-
}
|
|
238
|
-
try {
|
|
239
|
-
const res = await setupVault(root);
|
|
240
|
-
const st = await vaultStatus(res.root);
|
|
241
|
-
return jsonOk({ result: res, status: st });
|
|
242
|
-
} catch (e) {
|
|
243
|
-
return jsonErr((e as Error).message, 500);
|
|
244
|
-
}
|
|
245
|
-
}
|