mlx-bun 0.4.0 → 0.5.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (158) hide show
  1. package/README.md +57 -32
  2. package/docs/README.md +55 -42
  3. package/docs/archive/investigations/benchmark-ledger-through-v0.4.0.md +3617 -0
  4. package/docs/archive/planning/release-notes-v0.5.0.md +37 -0
  5. package/docs/design/batching.md +220 -7
  6. package/docs/design/decode-speed-program.md +121 -66
  7. package/docs/design/generic-model-support.md +116 -0
  8. package/docs/design/kv-cache.md +380 -211
  9. package/docs/design/speculative-decoding.md +165 -26
  10. package/docs/design/turboquant.md +20 -7
  11. package/docs/design/unified-engine-frontier-plan.md +120 -13
  12. package/docs/planning/PRODUCT_ROADMAP.md +5 -6
  13. package/docs/planning/public-readiness-handoff.md +244 -0
  14. package/docs/planning/public-readiness.md +149 -0
  15. package/docs/reference/benchmarks.md +3463 -3489
  16. package/docs/reference/cli.md +62 -1
  17. package/docs/reference/distribution.md +29 -4
  18. package/docs/reference/environment.md +104 -8
  19. package/docs/reference/library-api.md +18 -4
  20. package/docs/reference/memory.md +1 -1
  21. package/docs/reference/models.md +17 -5
  22. package/docs/reference/server-api.md +181 -11
  23. package/docs/reference/server-config.md +232 -105
  24. package/package.json +2 -2
  25. package/src/audio/audiotoolbox.ts +112 -0
  26. package/src/audio/conformer.ts +8 -0
  27. package/src/audio/mic-capture.ts +98 -0
  28. package/src/audio/silero-vad.ts +379 -0
  29. package/src/audio/whisper-decode.ts +774 -0
  30. package/src/audio/whisper-fast.ts +732 -0
  31. package/src/audio/whisper-format.ts +73 -0
  32. package/src/audio/whisper-mel.ts +185 -0
  33. package/src/audio/whisper-timing.ts +274 -0
  34. package/src/audio/whisper-tokenizer.ts +239 -0
  35. package/src/audio/whisper-transcribe.ts +410 -0
  36. package/src/backends/mlx/autoregressive.ts +22 -1
  37. package/src/backends/mlx/batch-group.ts +111 -27
  38. package/src/backends/mlx/cache-layout.ts +3 -3
  39. package/src/backends/mlx/continuation-persistence.ts +5 -4
  40. package/src/backends/mlx/encoder-cache.ts +25 -0
  41. package/src/backends/mlx/fill-append.ts +21 -8
  42. package/src/backends/mlx/fill-group.ts +497 -0
  43. package/src/backends/mlx/gateway-binding.ts +56 -12
  44. package/src/backends/mlx/kv-maintenance.ts +30 -2
  45. package/src/backends/mlx/mixed-iteration.ts +62 -0
  46. package/src/backends/mlx/model-prompt.ts +46 -32
  47. package/src/backends/mlx/model-serving.ts +4 -3
  48. package/src/backends/mlx/paged-cache-codec.ts +27 -0
  49. package/src/backends/mlx/prefill-cohort.ts +15 -4
  50. package/src/backends/mlx/prefill-policy.ts +26 -0
  51. package/src/backends/mlx/prefill-rows.ts +57 -13
  52. package/src/backends/mlx/prompt-input.ts +29 -0
  53. package/src/backends/mlx/qwen-prompt-input.ts +40 -0
  54. package/src/backends/mlx/request-state-policy.ts +18 -5
  55. package/src/backends/mlx/rollback.ts +6 -3
  56. package/src/backends/mlx/serial-executor.ts +16 -7
  57. package/src/backends/mlx/speculative-group.ts +105 -26
  58. package/src/backends/mlx/speculative.ts +24 -7
  59. package/src/backends/mlx/target-layout-capability.ts +1 -1
  60. package/src/backends/mlx/transitioning-kv-rows.ts +8 -0
  61. package/src/cli.ts +322 -0
  62. package/src/contracts/execution.ts +10 -0
  63. package/src/contracts/object-cache.ts +8 -0
  64. package/src/contracts/pi-provider.ts +1 -1
  65. package/src/contracts/pi.ts +3 -1
  66. package/src/contracts/prefix-cache.ts +13 -3
  67. package/src/contracts/scheduling.ts +9 -0
  68. package/src/engine/execution-plan.ts +23 -12
  69. package/src/engine/execution-tasks.ts +44 -0
  70. package/src/engine/namespaced-cache.ts +13 -0
  71. package/src/engine/scheduler.ts +22 -2
  72. package/src/fill/fill-session.ts +14 -3
  73. package/src/generate.ts +28 -22
  74. package/src/grammar.ts +13 -0
  75. package/src/inference/prefill.ts +6 -0
  76. package/src/kv-scheme.ts +22 -0
  77. package/src/kv-store.ts +115 -79
  78. package/src/lab/README.md +5 -1
  79. package/src/lab/paged-kv/paged-attention.ts +117 -0
  80. package/src/lab/paged-kv/paged-kv-rows.ts +33 -5
  81. package/src/lab/paged-kv/paged-kv.ts +108 -42
  82. package/src/md-modules.d.ts +5 -0
  83. package/src/mlx/array.ts +37 -0
  84. package/src/mlx/ffi.ts +19 -0
  85. package/src/mlx/normalized-argmax.ts +116 -0
  86. package/src/mlx/ops.ts +97 -10
  87. package/src/model/batched-turboquant-kv.ts +6 -5
  88. package/src/model/delayed-quantized-kv.ts +3 -3
  89. package/src/model/delayed-rotating-quantized-kv.ts +1 -1
  90. package/src/model/delayed-turboquant-kv.ts +6 -4
  91. package/src/model/factory.ts +26 -0
  92. package/src/model/fingerprint.ts +12 -0
  93. package/src/model/full-prefill-row.ts +1 -1
  94. package/src/model/gemma4-base.ts +25 -6
  95. package/src/model/gemma4.ts +82 -2
  96. package/src/model/kv-attention-view.ts +4 -1
  97. package/src/model/profile.ts +29 -4
  98. package/src/model/quantized-append-attention.ts +31 -0
  99. package/src/model/qwen3-mrope.ts +7 -9
  100. package/src/model/qwen3_5.ts +99 -37
  101. package/src/model/support.ts +8 -0
  102. package/src/model/token-groups.ts +66 -0
  103. package/src/model/trellis-linear.ts +2 -2
  104. package/src/model/turboquant-codec.ts +6 -0
  105. package/src/model/whisper.ts +435 -0
  106. package/src/native/mic_capture.swift +92 -0
  107. package/src/native-pack.ts +6 -5
  108. package/src/pi-web.ts +9 -1
  109. package/src/prompt-cache.ts +265 -43
  110. package/src/runtime-config.ts +1 -5
  111. package/src/safetensors.ts +9 -2
  112. package/src/sampler.ts +5 -4
  113. package/src/serve/audio-routes.ts +323 -0
  114. package/src/serve/chat-request.ts +2 -0
  115. package/src/serve/chat-stage.ts +17 -12
  116. package/src/serve/completion-executor.ts +9 -1
  117. package/src/serve/discovery-routes.ts +17 -1
  118. package/src/serve/generated-token-history.ts +57 -0
  119. package/src/serve/generation-gateway.ts +52 -11
  120. package/src/serve/model-binding.ts +7 -1
  121. package/src/serve/preparation.ts +3 -3
  122. package/src/serve/prompt-response-trace.ts +14 -0
  123. package/src/serve/request-plan.ts +2 -0
  124. package/src/serve/request-prep.ts +38 -21
  125. package/src/serve/transcription-server.ts +64 -0
  126. package/src/serve/transcription-service.ts +502 -0
  127. package/src/server.ts +132 -164
  128. package/src/spec/assistant-rows.ts +1 -0
  129. package/src/spec/deepspec-rows.ts +1 -1
  130. package/src/spec/dflash-source.ts +1 -1
  131. package/src/spec/glm52-mtp-source.ts +1 -0
  132. package/src/spec/ngram-source.ts +27 -5
  133. package/src/spec/projected-draft-rows.ts +2 -2
  134. package/src/spec/qwen-mtp-rows.ts +11 -0
  135. package/src/spec/qwen-mtp-source.ts +3 -0
  136. package/src/spec/serve-loop.ts +5 -3
  137. package/src/spec/source.ts +14 -1
  138. package/src/spec/two-model-rows.ts +1 -0
  139. package/src/spec/two-model.ts +2 -2
  140. package/src/ssd-cache.ts +76 -17
  141. package/src/ssd-durability.ts +20 -22
  142. package/src/storage/host-buffer.ts +24 -0
  143. package/src/storage/kv-writer.ts +83 -0
  144. package/src/storage/kv-writer.worker.js +264 -0
  145. package/src/storage/retention-policy.ts +37 -0
  146. package/src/tiered-prompt-cache.ts +45 -0
  147. package/src/vision/embedder.ts +8 -0
  148. package/src/vision/prompt.ts +101 -65
  149. package/src/vision/qwen3vl-prompt.ts +101 -64
  150. package/src/vision/qwen3vl-tower.ts +11 -0
  151. package/src/vision/siglip.ts +8 -0
  152. package/src/web/app.html +12 -0
  153. package/src/web/app.js +160 -0
  154. package/src/web/src/chat.ts +4 -1
  155. package/src/web/src/composer.ts +3 -0
  156. package/src/web/src/voice.ts +141 -0
  157. package/src/web/sw.js +1 -1
  158. /package/docs/{planning → archive/planning}/release-notes-v0.4.0.md +0 -0

There are too many changes on this page to be displayed.


The amount of changes on this page would crash your brower.

You can still verify the content by downloading the package file manually.