Commits

Commit history

1. Commits on Sep 21, 2026

   • perf(qwen3.5): cut ~550 dispatches/cycle from the decode control flow (#168)

     Show description for b122e8c
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     b122e8cView commit details

     Copy full SHA for b122e8c

     Browse repository at this point

   • fix(inference): remove panic paths from the production inference surface (#167)

     Show description for fcaa4fc
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     fcaa4fcView commit details

     Copy full SHA for fcaa4fc

     Browse repository at this point

   • perf(qwen3.5): fused DFlash2 conv kernel + optional draft quantization (#165)

     Show description for bad5b9e
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     bad5b9eView commit details

     Copy full SHA for bad5b9e

     Browse repository at this point

   • perf(quant): row-merge compatible packed projections (gate|up, kv, qkvz|ba) (#164)

     Show description for 940e300
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     940e300View commit details

     Copy full SHA for 940e300

     Browse repository at this point

   • perf(metal): single-pass qmv_wide tiles for M=6..8 quantized matvecs (#163)

     Show description for cd0cfea
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     cd0cfeaView commit details

     Copy full SHA for cd0cfea

     Browse repository at this point

   • perf(qwen3.5): cut DFlash2 cycle syncs and dead decode work (#161) (#162)

     Show description for 88c0385
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     88c0385View commit details

     Copy full SHA for 88c0385

     Browse repository at this point
2. Commits on Sep 19, 2026

   • feat(lm): support Ternary Bonsai 2 PQ2_0 inference (#160)

     Show description for 784ed80
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     784ed80View commit details

     Copy full SHA for 784ed80

     Browse repository at this point

   • feat(lm): add K2-Horizon and shared inference plumbing (#158)

     Show description for 80d1dfd
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     80d1dfdView commit details

     Copy full SHA for 80d1dfd

     Browse repository at this point

   • fix(lfm2): vLLM-align LFM2/LFM2.5 inference + shared paged-policy abstraction (#157)

     Show description for a95a3f9
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     a95a3f9View commit details

     Copy full SHA for a95a3f9

     Browse repository at this point
3. Commits on Sep 18, 2026

   • feat(catalog): default to Unsloth UD-Q4_K_XL GGUFs and add the native MoE GGUF path (#156)

     Show description for 10697d3
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     10697d3View commit details

     Copy full SHA for 10697d3

     Browse repository at this point
4. Commits on Sep 17, 2026

   • feat(lm): add bounded Qwen3.8 inference and mlx.fast kernel ports (#154)

     Show description for 1787380
     Brooooooklyn

     authored
     1787380View commit details

     Copy full SHA for 1787380

     Browse repository at this point
5. Commits on Sep 16, 2026

   • fix(desktop): downloads behind TLS-intercepting networks; no inference start without models (#155)

     Show description for ae9ea3b
     

     Brooooooklyn

     and

     devin-ai-integration[bot]

     authored
     ae9ea3bView commit details

     Copy full SHA for ae9ea3b

     Browse repository at this point
6. Commits on Sep 15, 2026

   • 0.0.15

     Brooooooklyn

     committed
     e915d77View commit details

     Copy full SHA for e915d77

     Browse repository at this point

   • fix(desktop): rewrite the tokenizers prebuild's install name before signing (#153)

     Show description for 19b796c
     Brooooooklyn

     authored
     19b796cView commit details

     Copy full SHA for 19b796c

     Browse repository at this point

   • test(lfm2): pin the warm-continue comparison to answer-region decisions (#152)

     Show description for b8da5be
     Brooooooklyn

     authored
     b8da5beView commit details

     Copy full SHA for b8da5be

     Browse repository at this point

   • 0.0.14

     Brooooooklyn

     committed
     14953cdView commit details

     Copy full SHA for 14953cd

     Browse repository at this point

   • refactor(workspace): resolve @mlx-node/* through each package's exports map (#151)

     Show description for 30d3372
     Brooooooklyn

     authored
     30d3372View commit details

     Copy full SHA for 30d3372

     Browse repository at this point

   • feat(delegate): label sessions and show estimated token savings (#150)

     Show description for 2caca3a
     Brooooooklyn

     authored
     2caca3aView commit details

     Copy full SHA for 2caca3a

     Browse repository at this point
7. Commits on Sep 14, 2026

   • feat(dashboard): add first-run model onboarding (#149)

     Show description for 766ea92
     Brooooooklyn

     authored
     766ea92View commit details

     Copy full SHA for 766ea92

     Browse repository at this point

   • feat(agent): add local GitHub delegation and thinking controls (#148)

     Show description for 8d50ecf
     Brooooooklyn

     authored
     8d50ecfView commit details

     Copy full SHA for 8d50ecf

     Browse repository at this point
8. Commits on Sep 12, 2026

   • test(qwen3): stabilize mixed prefill/decode parity check (#147)

     Show description for a11526d
     Brooooooklyn

     authored
     a11526dView commit details

     Copy full SHA for a11526d

     Browse repository at this point

   • feat(muse): load native GGUF checkpoints and optimize inference (#146)

     Show description for eaf8373
     Brooooooklyn

     authored
     eaf8373View commit details

     Copy full SHA for eaf8373

     Browse repository at this point
9. Commits on Sep 11, 2026

   • Pin Gemma and Qwen performance fixtures for reproducible benchmarks (#145)

     Show description for 66344b0
     Brooooooklyn

     authored
     66344b0View commit details

     Copy full SHA for 66344b0

     Browse repository at this point

   • feat(gemma4): add native GGUF loading and adaptive decode (#144)

     Show description for 3cdbfe7
     Brooooooklyn

     authored
     3cdbfe7View commit details

     Copy full SHA for 3cdbfe7

     Browse repository at this point
10. Commits on Sep 10, 2026

    • feat: recommend and auto-load Qwen3.8 DFlash2 companions (#143)

      Show description for 185bc1b
      Brooooooklyn

      authored
      185bc1bView commit details

      Copy full SHA for 185bc1b

      Browse repository at this point
11. Commits on Sep 9, 2026

    • fix(dashboard): match persistent scrollbars to app theme

      Brooooooklyn

      committed
      c136d11View commit details

      Copy full SHA for c136d11

      Browse repository at this point

    • perf(agent): accelerate Qwen decode and SSD cache transfers (#141)

      Show description for 770e380
      Brooooooklyn

      authored
      770e380View commit details

      Copy full SHA for 770e380

      Browse repository at this point

    • fix(dashboard): close the settle window on cancel and on the updates half (#140)

      Show description for c14e558
      

      Brooooooklyn

      and

      claude

      authored
      c14e558View commit details

      Copy full SHA for c14e558

      Browse repository at this point

    • feat(desktop): add automatic macOS updates (#139)

      Show description for f2db0a2
      Brooooooklyn

      authored
      f2db0a2View commit details

      Copy full SHA for f2db0a2

      Browse repository at this point
12. Commits on Sep 8, 2026

    • refactor(inference): share model traits and optimize speculative decoding (#138)

      Show description for 78fb189
      Brooooooklyn

      authored
      78fb189View commit details

      Copy full SHA for 78fb189

      Browse repository at this point
13. Commits on Sep 4, 2026

    • 0.0.13

      Brooooooklyn

      committed
      100a03aView commit details

      Copy full SHA for 100a03a

      Browse repository at this point

    • Models page: move the size label to the title row

      Show description for a623e96
      Brooooooklyn

      committed
      a623e96View commit details

      Copy full SHA for a623e96

      Browse repository at this point

    • Models page: keep the size label on one line beside long repo names

      Show description for ef60706
      Brooooooklyn

      committed
      ef60706View commit details

      Copy full SHA for ef60706

      Browse repository at this point
14. Commits on Sep 3, 2026

    • 0.0.12

      Brooooooklyn

      committed
      7a8998dView commit details

      Copy full SHA for 7a8998d

      Browse repository at this point

    • Drain pending session disposals in the stage-1 e2e gate

      Show description for 1e586d1
      Brooooooklyn

      committed
      1e586d1View commit details

      Copy full SHA for 1e586d1

      Browse repository at this point