Repository navigation

release: sync fork and add signed notarized macOS DMG pipeline - #2

Merged

jonathan308 merged 345 commits into

mainfrom

release/signed-dmg

Aug 25, 2026

Merged

release: sync fork and add signed notarized macOS DMG pipeline#2
jonathan308 merged 345 commits into
mainfrom
release/signed-dmg

This pull request is big! We're only showing the most recent 250 commits

Commits on Aug 13, 2026
Commits on Aug 14, 2026
Commits on Aug 15, 2026
- fix(oq): strip HF org prefix from quantized model names (#2652)
  monroewilliams
  authored
  d92ed3c
- fix(qwen35): skip native qmm routing at group_size=128 on NAX hardware (#2657)
  True2456
  authored
  80aa600
- fix(qwen35): gate gs128 fallback on stock NAX
  jundot
  committed
  cc9944e
- fix(vlm): thread tools through chat path (#2645)
  mkmsyk
  authored
  7eea370
- feat(cache): add storage codecs and policy for persisted GDN sidecars (#2644)
  DT-HJ
  authored
  3b2e02e
- chore(cache): drop duplicate gdn sidecar properties and repeated test key
  jundot
  committed
  5926ce7
- test: cover VLM tool schema forwarding (#2646)
  jundot
  committed
  ae1883c
- feat(vlm): sync Muse Glimmer vendor with mlx-vlm #1848 numerics (#2643)
  nazerim
  authored
  1ee0aa9
- feat(qwen): support Qwen3.8 FP8 and reasoning levels (#2653)
  scaryrawr
  authored
  2bb2cc5
- fix(vlm): align Muse Glimmer FP32 scaling paths
  jundot
  committed
  a719ba9
- fix(memory): reclaim MLX pool before hot-cache shrink (#2635)
  PhilipJohnBasile
  authored
  ee2d68d
- fix(benchmark): preserve VLM engine for Lightning MTP (#2651)
  jundot
  committed
  db4331e
- fix(mac): scope main window URL to current app (#2658)
  jundot
  committed
  71b9d52
- fix: make mac updater swap crash-safe (#2522)
  jundot
  committed
  101b211
- feat(ui): add Enhanced Readability accessibility toggle (#2542)
  LXD-8
  authored
  5ee881b
- fix(dspark): stop thinking-budget counter drift from MTP speculative calls (#2617)
  mvdbos
  authored
  cdb6117
- fix(ui): refine enhanced readability controls
  jundot
  committed
  3fcacbd
- fix(responses): start thinking parser when prompt opens <think> (#2623)
  leepokai
  authored
  e4bab83
- test: remove updater relaunch smoke (#2522)
  jundot
  committed
  f04f377
- fix(tool-calling): stop deep nesting escaping the parse chain (#2593)
  
  richgoodson
  and
  discoStew
  authored
  1693022
- fix(muse): keep tool calls when the turn opens with one (#2599)
  dawgflymd
  authored
  9174102
- fix: support unaligned DeepSeek V4 indexer tails (#2674)
  jundot
  authored
  997c9b6
- fix: make Claude cross-session messaging opt-in (#2601)
  fparrav
  authored
  22cab14
- fix: honor Claude cross-session opt-out semantics (#2601)
  jundot
  committed
  80f4031
- feat(cache): surface hot cache promotion failures in stats
  jundot
  committed
  9dbc3f3
- chore: bump version to 0.6.0rc1
  jundot
  committed
  e81e825
- formula: bump to 0.6.0rc1
  github-actions[bot]
  committed
  b2fdc71
Commits on Aug 16, 2026
- feat(mtp): support Gemma4 unified Lightning MTP (#2683)
  djr747
  authored
  1177b88
- chore(deps): bump ddgs from 9.14.1 to 9.14.4 (#2685)
  dependabot[bot]
  authored
  4a1c635
- fix(admin): support free-form reasoning_effort values (#2675)
  Pietroski
  authored
  5890d0d
- fix: improve reasoning effort controls
  jundot
  committed
  03bbdec
- fix(cluster): Keep display names out of node IDs (#2679)
  jundot
  committed
  c4095af
- fix(eval): handle prompt-opened thinking output (#2661)
  jundot
  committed
  42813b4
- fix(mtp): preserve rollback for depth-8 Lightning verify (#2690)
  adamw3455
  authored
  1a0f691
- feat(model-loading): support Qwen3.8 mixed ModelOpt NVFP4 (#2659)
  LKyaha
  authored
  c82cabc
- feat(ui): use PingFang for Chinese and improve accessibility (#2671)
  
  authored
  8bdeacb
- fix(embeddings): honour checkpoint pooling metadata (#2594)
  cedricamram
  authored
  a32b6b1
- fix(ui): make focus ring visible in dark mode
  jundot
  committed
  fd14f7f
- feat: upload intelligence benchmark results to omlx.ai (#2693)
  jundot
  authored
  3fd83ff
- fix(embeddings): keep last-token pooling compile-safe
  jundot
  committed
  734219e
- fix(discovery): skip incompletely downloaded sharded models (#2432)
  JimStenstrom
  authored
  150279d
- fix: update Homebrew install requirements (#2694)
  jundot
  committed
  5326687
- fix(oq): clamp the sensitivity perturbation to a valid affine group size (#2433)
  JimStenstrom
  authored
  6b3d7e5
- fix(admin): localize model settings labels (#2480)
  
  aosama
  authored and
  jundot
  committed
  9be9219
- fix(i18n): sync model settings labels across locales (#2476)
  jundot
  committed
  a857d52
- fix(deepseek_v4): always emit the generation anchor for system-final prompts (#2510)
  jonathan308
  authored
  81c4394
- fix(deepseek_v4): preserve task generation transitions
  jundot
  committed
  19a7b1e
- fix(engine_core): don't re-abort requests already finished or pending abort (#2512)
  jonathan308
  authored
  0fbc5a4
- test(engine_core): cover repeated bulk aborts
  jundot
  committed
  ac58d7c
- fix(deepseek_v4): make tool-call turn re-rendering byte-stable with model output (#2513)
  JimStenstrom
  authored
  b11ec0d
- Chore: Bump version to 0.6.0
  jundot
  committed
  fb60222
- fix(admin): clarify and reorganize global settings
  jundot
  committed
  ecc79a8
- fix(server): surface unenforced response_format on /v1/responses (#2560)
  
  fxd0h
  and
  claude
  authored
  5d41139
- fix(memory): align DeepSeek-V4 prefill estimate with WSDPA route (#2639)
  mvdbos
  authored
  5175562
- feat(mcp): add dashboard toggle to expose backend MCP tools to clients (#2699)
  cgs4github
  authored
  1a7e567
- feat: update model preset profiles
  jundot
  committed
  daf74d1
- fix(cluster): discover peer runtimes and verify parity (#2697)
  hellodk
  authored
  f983355
- fix(mcp): complete expose-tools toggle follow-up (#2699)
  jundot
  committed
  4f5ba16
- fix(cluster): allow mixed Python minor versions
  jundot
  committed
  ddb1ee0
- fix: honor web search test result limit (#2696)
  jundot
  committed
  9f115c7
- fix: make chat tool round limits configurable (#2702)
  jundot
  committed
  818f1a5
- fix: quiesce active engines before admin unload
  jundot
  committed
  5ba3d6a
- fix(mac): reset restart budget after stable health
  jundot
  committed
  c9952e3
- fix(cluster): align runtime version checks and status (#2704, #2705)
  jundot
  committed
  78a1464
- chore: finalize 0.6.0 release
  jundot
  committed
  b16a1d1
- formula: bump to 0.6.0
  github-actions[bot]
  committed
  854060f
Commits on Aug 17, 2026
- fix: publish org-qualified model names to omlx.ai
  jundot
  committed
  fceb5af
- fix(deepseek_v4): map client reasoning_effort aliases instead of 400 (#2724)
  jonathan308
  authored
  3355679
- fix: normalize reasoning effort across chat templates (#2740)
  jundot
  committed
  b876be8
- fix: load root MTP weights for Qwen VLMs (#2732)
  jundot
  committed
  a593a55
- fix: guard model type during batched preflight (#2740)
  jundot
  committed
  c77db0d
- fix: keep filtered HF cache models discoverable (#2742)
  jundot
  committed
  cfdb079
- perf(vlm-mtp): prefer batching under scheduler contention (#2752)
  
  DiscoStew6082
  and
  discoStew
  authored
  fc640a8
- Added Qwen3.5 architecture ANE Kernel, benchmark options, and model settings.
  fabianfeary
  committed
  edb7290
- fix(vlm): normalize Qwen3.5 patch embedding layout (#2754)
  frank-beans
  authored
  c8940ca
- perf: speed up qwen3.5 MTP verify and prefill (#2751)
  jundot
  authored
  293d697
- fix: preserve mid-system cache after tool results (#2753)
  q-p
  authored
  d5637c5
- perf: fuse qwen MoE decode router top-k (#2755)
  jundot
  authored
  a9de32c
- feat: upload model repo id separately from model name
  jundot
  committed
  f610c6a
- fix: validate tool-adjacent mid-system templates (#2483)
  jundot
  committed
  bddc9f8
- fix(cluster): compare probe versions against the source the peer reports from (#2758)
  hellodk
  authored
  eed0844
- feat(qwen3.5): add experimental dual-ANE/GPU prompt processing (#2756)
  jundot
  committed
  fbb98dc
- feat(qwen3.5): extend ANE prefill to oQ4e and harden runtime (#2760)
  jundot
  authored
  b2873ed
- fix(admin): preserve measured ANE select defaults
  jundot
  committed
  9b494be
- chore: bump version to 0.6.1
  jundot
  committed
  b587575
- formula: bump to 0.6.1
  github-actions[bot]
  committed
  6307e39
Commits on Aug 18, 2026
Commits on Aug 19, 2026
- chore: bump dflash-mlx pin for DFlash 2 runtime support
  jundot
  committed
  8b0aa99
- feat(dflash): support DFlash 2 and update draft defaults (#2840)
  
  liang2kl
  and
  jianc99
  authored
  cfa6653
- feat(dflash): finalize DFlash 2 rollout (#2850)
  jundot
  authored
  c1a3d44
- feat(qwen35): enable affine q6 and q8 ANE prefill (#2833)
  GordoAR
  authored
  12841f2
- fix(qwen35): route q8 GDN b/a through the q8 token threshold
  jundot
  committed
  ac34387
- fix(cluster): initialize settings before memory guard fallback (#2835)
  
  xunlinkx
  and
  tronik
  authored
  d6b823f
- fix(cluster): validate remote process identity and escalate reaping on stop (#2722)
  xunlinkx
  authored
  cf91538
- fix(cluster): stop the planner from re-scanning every model on each autoconfigure tick (#2883)
  
  alytaphoenix
  and
  Alyta Phoenix
  authored
  cea0420
- fix(cluster): two-Mac clustering reliability (poll jitter, VLM tensor default, cross-user paths, ssh_user) (#2819)
  
  authored
  677dc45
- fix(cluster): make quantized Nemotron-H run tensor-parallel at TP=2 (#2844)
  
  authored
  d85c0d9
- fix(cluster): retry an unsupported reasoning_effort against rank-zero (#2869)
  
  authored
  254793d
- feat(cluster): add stable machine codes to failure guidance
  jundot
  committed
  ed35acc
- feat(cluster): server-owned incident feed with monotonic-merge polling (B1) (#2866)
  
  authored
  ef53202
- fix(ane): reject tuning candidates whose ANE never executed (#2829)
  beaglemoo
  authored
  dde94f7
- fix(ane): follow ups for the idle-candidate tuner guard
  jundot
  committed
  375d3f5
- fix(ane): measure tuner prompts with a non-ANE tail chunk
  jundot
  committed
  a389f19
- fix(cluster): read operator settings without publishing the global singleton (#2835 follow-up)
  jundot
  committed
  2caa3ee
- fix(cluster): keep rank markers as crash evidence and report reap results (#2722 follow-up)
  jundot
  committed
  5830658
- fix(cluster): dashboard sync guards, peer-home source paths, drop ssh_user (#2819 follow-up)
  jundot
  committed
  1f49541
- fix(cluster): close planner guard gaps for quantized nemotron_h TP (#2844 follow-up)
  jundot
  committed
  15ed484
- fix(cluster): try the normalized reasoning_effort tier first (#2869 follow-up)
  jundot
  committed
  e92f6b5
- fix(cluster): record incidents off the event loop, add store epoch (#2866 follow-up)
  jundot
  committed
  7f6419f
- fix(cluster): include shard stats in the layout cache fingerprint (#2883 follow-up)
  jundot
  committed
  9f13324
- Add Q5 ANE prefill support (#2889)
  
  onthehub97
  and
  fabianfeary
  authored
  c7ca2c5
- Add wide GPU prefill blocks, internal ANE tiling, and benchmark fidelity and instrumentation (#2890)
  
  onthehub97
  and
  fabianfeary
  authored
  1e03b5e
- fix(admin): clear default_model pointer when is_default is unset (#2884)
  
  alytaphoenix
  and
  Alyta Phoenix
  authored
  582fc84
- fix(qwen35): follow ups for wide prefill tiling
  jundot
  committed
  44bf3c2
- Qwen35: ANE Tuner V2 (#2891)
  
  authored
  166c8f4
- fix(patches): heal reinstalled nemotron chain surfaces instead of trusting the class flag
  jundot
  committed
  af9ac31
- fix(cluster): preflight cluster health so a dead rank fails before the 200 commits
  jundot
  committed
  f487604
- fix(ane): tuner v2 follow ups
  jundot
  committed
  4c00bbe
- fix(admin): accept versioned DFlash2 names in draft-model picker (#2879)
  williamxie1989
  authored
  ec9769d
- fix(api): strip Claude Code total_tokens budget markers from system text (#2882)
  
  Pietroski
  and
  claude
  authored
  4ee255b
- fix(cli): strip leaked -- separator from launch extra_args (#2818)
  fparrav
  authored
  99549f6
- feat(admin): add max_tokens override for external accuracy benchmarks (#2863)
  williamxie1989
  authored
  5ff08cd
- fix(responses): surface max_output_tokens truncation as incomplete (#2817)
  mvdbos
  authored
  c8bfcc2
- fix(cli): preserve forwarded argument separators
  jundot
  committed
  280e2ce
- fix(responses): emit incomplete terminal event
  jundot
  committed
  aadfb42
- perf(laguna): port mlxfast-challenge decode optimizations (#2827)
  0xClandestine
  authored
  18b2f9a
- chore: bump version to 0.6.3rc1
  jundot
  committed
  274d17f
- fix: make Laguna compiled-tail test hardware-independent
  jundot
  committed
  0ed0187
- formula: bump to 0.6.3rc1
  github-actions[bot]
  committed
  146d272
Commits on Aug 20, 2026
- feat(ane): warm compiled procedures at load so the first request measures inference (#2898)
  beaglemoo
  authored
  740854c
- feat(ane): surface Qwen ANE prefill no-ops in logs and /api/status (#2904)
  
  finaltv971
  and
  anthony BAUCAL
  authored
  0e72928
- fix: preserve Qwen ANE app profile (#2932)
  popfido
  authored
  be5aaf5
- Qwen35 ANE: CPU Sharing (#2892)
  
  authored
  3d8e661
- docs: add M5 to Apple Silicon requirements (#2925)
  qdaszx
  authored
  99b5398
- fix: keep packed q6 GDN b/a suffix on the banked ANE path
  jundot
  committed
  4ff5858
- fix: ANE follow-ups after the CPU sharing merge
  jundot
  committed
  7576cf7
- fix: warm the CPU sharing path at load
  jundot
  committed
  a79ce4b
- fix: harden ANE tuner staging, GDN floor, and memory release
  jundot
  committed
  2a7bac9
- fix: reserve ANE prefill I/O surfaces in admission
  jundot
  committed
  dcb317f
- feat: M2-tuned MMA indexer score kernel for DSV4F (#2802)
  
  nathanodle
  and
  nathanodle
  authored
  1f6302e
- style: drop an extra blank line after the mma score gate helper
  jundot
  committed
  5a81899
- chore: remove unused BK template param and stale doc ref in the mma score kernel
  jundot
  committed
  5b6f2a7
- test: cover batched inputs in the mma score bit-exact suite
  jundot
  committed
  c6cb4b6
- fix(memory): size hybrid KV blocks by cache layers (#2912)
  
  saichowdary007
  and
  sai
  authored
  4af370f
- Fix reasoning effort editor layout (#2922)
  popfido
  authored
  7b47331
- fix: write settings.json atomically and move corrupt files aside (#2926)
  alexferrao
  authored
  45c7eb9
- fix(settings): use per-process temp names for atomic saves
  jundot
  committed
  a42d837
- perf: stage ANE bank weights incrementally at compile (#2781)
  jundot
  committed
  df05710
- fix: drain committed qmm buffers when a CPU share dispatch throws
  jundot
  committed
  df3a0c2
- fix(cluster): recover from SSH pairing failures (#2930)
  ashhart
  authored
  e0be3c0
- fix(cluster): keep SSH key rotation state consistent (#2930)
  jundot
  committed
  4810f55
- fix(cluster): harden CUDA enrollment and ConnectX verification (#2929)
  ashhart
  authored
  cce2bca
- fix(cluster): enforce CUDA activation and SSH path guards (#2929)
  jundot
  committed
  d4efd93
- fix(ane): isolate ANE warmup failures per module (#2946)
  beaglemoo
  authored
  336e558
- chore: bump version to 0.6.3rc2
  jundot
  committed
  4cb5516
- formula: bump to 0.6.3rc2
  github-actions[bot]
  committed
  fa3e94b
Commits on Aug 21, 2026
- feat(Qwen35) Fused Qwen MLP/down offload and hardware-adaptive tuner (#2935)
  
  onthehub97
  and
  fabianfeary
  authored
  d2110d1
- Added tail padding (#2966)
  
  onthehub97
  and
  fabianfeary
  authored
  f14796f
- fix: stream fused ANE bank staging through an incremental builder
  jundot
  committed
  c267fc6
- fix: guard the fused dispatch site against mid-flight exceptions
  jundot
  committed
  e200e0d
- fix: extend admission pricing and CPU warmup to fused and ANE-down states
  jundot
  committed
  a1ebecb
- fix: reject fused down settings that disable ANE at load
  jundot
  committed
  0436bdf
- fix: re-enter Lightning MTP after performance parking (#3004)
  jundot
  authored
  c0d2eec
- fix(mtp): scope the prime window to the folded span (#3005)
  beaglemoo
  authored
  295eba4
- fix(mtp): keep prime-window overflow latched
  jundot
  committed
  78c799c
- fix(responses): route function_call_output images to vlm instead of prompt text (#2989)
  jundot
  committed
  2df39bf
Commits on Aug 23, 2026
Commits on Aug 24, 2026
Commits on Aug 25, 2026
- Prepare signed notarized fork DMG releases
  jonathan308
  committed
  5d17d34
- Merge remote-tracking branch 'fork/main' into release/signed-dmg
  jonathan308
  committed
  49e91b8

Commits

Commits on Aug 13, 2026

Commits on Aug 14, 2026

Commits on Aug 15, 2026

Commits on Aug 16, 2026

Commits on Aug 17, 2026

Commits on Aug 18, 2026

Commits on Aug 19, 2026