Commits

Commit history

1. Commits on Nov 3, 2026

   • fix(moe): publish rocm cpu expert inputs before notification (#764)

     Show description for 9cacd08
     

     zihaomu

     and

     LZ-QWQ

     authored
     9cacd08View commit details

     Copy full SHA for 9cacd08

     View code at this pointBrowse repository at this point

   • feat(gguf): read nvfp4 weight scales on qwen4exp ggufs (#845)

     ischencheng

     authored
     d026ffdView commit details

     Copy full SHA for d026ffd

     View code at this pointBrowse repository at this point
2. Commits on Nov 2, 2026

   • perf(cpu-moe): speed up the nvfp4 W4A8 kernels (#851)

     Show description for 3d23551
     jason-fxz

     authored
     3d23551View commit details

     Copy full SHA for 3d23551

     View code at this pointBrowse repository at this point

   • feat(spec): add MTP drafts for the Qwen3.5 family and Qwen3.8-Flash-Next (#844)

     Show description for dacac5f
     jason-fxz

     authored
     dacac5fView commit details

     Copy full SHA for dacac5f

     View code at this pointBrowse repository at this point

   • feat(gguf): serve gpt-oss gguf checkpoints on gpt_oss (#838)

     Show description for ea2916d
     void-mckenzie

     authored
     ea2916dView commit details

     Copy full SHA for ea2916d

     View code at this pointBrowse repository at this point
3. Commits on Oct 30, 2026

   • fix(server): refuse n != 1 on /v1/completions instead of ignoring it (#842)

     A5bhinav

     authored
     ee43b2aView commit details

     Copy full SHA for ee43b2a

     View code at this pointBrowse repository at this point
4. Commits on Oct 29, 2026

   • fix(moe): skip inactive gguf routes in hybrid decode (#823)

     Show description for ecbbce4
     ischencheng

     authored
     ecbbce4View commit details

     Copy full SHA for ecbbce4

     View code at this pointBrowse repository at this point

   • feat(spec): verify draft tokens on the target side (#836)

     Show description for c2c3e82
     jason-fxz

     authored
     c2c3e82View commit details

     Copy full SHA for c2c3e82

     View code at this pointBrowse repository at this point

   • feat(kernel): add an iq1_m mmq kernel for moe steps above 8 tokens (#835)

     ischencheng

     authored
     d2977fbView commit details

     Copy full SHA for d2977fb

     View code at this pointBrowse repository at this point

   • fix(attention): size the decode KV tile to the shared-memory budget (#840)

     Show description for c94535c
     festeraeb

     authored
     c94535cView commit details

     Copy full SHA for c94535c

     View code at this pointBrowse repository at this point

   • refactor(layers): share one GatedDeltaNet op in layers/linear_attn (#841)

     Show description for d043fb6
     jason-fxz

     authored
     d043fb6View commit details

     Copy full SHA for d043fb6

     View code at this pointBrowse repository at this point
5. Commits on Oct 28, 2026

   • refactor(attention): move linear-attention metadata into a LinearAttnBackend (#837)

     Show description for a2ce636
     jason-fxz

     authored
     a2ce636View commit details

     Copy full SHA for a2ce636

     View code at this pointBrowse repository at this point
6. Commits on Oct 27, 2026

   • refactor: add engine and scheduler seams for multi-row decode steps (#833)

     Show description for e328cf6
     jason-fxz

     authored
     e328cf6View commit details

     Copy full SHA for e328cf6

     View code at this pointBrowse repository at this point

   • feat(scheduler): add radix_lookahead keys to the prefix cache (#834)

     Show description for 9221fd7
     jason-fxz

     authored
     9221fd7View commit details

     Copy full SHA for 9221fd7

     View code at this pointBrowse repository at this point

   • fix(server): recover Qwen tool calls after unclosed think (#825)

     madbrain76

     authored
     075cf57View commit details

     Copy full SHA for 075cf57

     View code at this pointBrowse repository at this point

   • feat(gguf): serve image input on minimax-m3 ggufs from a minimax_m3 mmproj (#829)

     Show description for 408e6ae
     jason-fxz

     authored
     408e6aeView commit details

     Copy full SHA for 408e6ae

     View code at this pointBrowse repository at this point

   • feat(gguf): serve image input on muse-glimmer ggufs from a muse-glimmer mmproj (#830)

     Show description for edab62c
     jason-fxz

     authored
     edab62cView commit details

     Copy full SHA for edab62c

     View code at this pointBrowse repository at this point
7. Commits on Oct 26, 2026

   • feat(gguf): serve image input on qwen3vl ggufs from a deepstack qwen3vl_merger mmproj (#827)

     void-mckenzie

     authored
     380693aView commit details

     Copy full SHA for 380693a

     View code at this pointBrowse repository at this point

   • chore(release): 0.2.1 (#824)

     jason-fxz

     authored
     ba31db2View commit details

     Copy full SHA for ba31db2

     View code at this pointBrowse repository at this point
8. Commits on Oct 25, 2026

   • fix(models): discover safetensors shards from the index (#819)

     Show description for 0bf7dbb
     chrisqianz

     authored
     0bf7dbbView commit details

     Copy full SHA for 0bf7dbb

     View code at this pointBrowse repository at this point
9. Commits on Oct 24, 2026

   • feat(gguf): serve image input on deepseek4 ggufs from a deepseek4v mmproj (#816)

     Show description for 271ff07
     void-mckenzie

     authored
     271ff07View commit details

     Copy full SHA for 271ff07

     View code at this pointBrowse repository at this point

   • feat(flint): add nvfp4 modelopt and gguf support (#797)

     Show description for 2d93085
     jesusjodar

     authored
     2d93085View commit details

     Copy full SHA for 2d93085

     View code at this pointBrowse repository at this point
10. Commits on Oct 23, 2026

    • fix(ple): add rocm disk wait-sync support (#776)

      Show description for 32056cf
      LZ-QWQ

      authored
      32056cfView commit details

      Copy full SHA for 32056cf

      View code at this pointBrowse repository at this point

    • feat: add --embed-weights host for a pinned host embedding table (#806)

      Show description for fd863cc
      jason-fxz

      authored
      fd863ccView commit details

      Copy full SHA for fd863cc

      View code at this pointBrowse repository at this point

    • fix(kernel): convert fp32 to e4m3 in one rounding on sm_89 (#805)

      Show description for 6505b6a
      

      ischencheng

      and

      zihaomu

      authored
      6505b6aView commit details

      Copy full SHA for 6505b6a

      View code at this pointBrowse repository at this point

    • perf(attention): drop redundant activation copies in glm5 dsa prefill (#795)

      Show description for 0c50eb9
      void-mckenzie

      authored
      0c50eb9View commit details

      Copy full SHA for 0c50eb9

      View code at this pointBrowse repository at this point

    • feat(gguf): serve image input on glm5-next ggufs from a glm5v mmproj (#810)

      void-mckenzie

      authored
      0b98e9dView commit details

      Copy full SHA for 0b98e9d

      View code at this pointBrowse repository at this point

    • fix(rotary): preserve fp32 inputs in mrope torch fallbacks (#641)

      taking-lying-flat

      authored
      507a8bfView commit details

      Copy full SHA for 507a8bf

      View code at this pointBrowse repository at this point

    • feat(server): add /v1/cache/flush to drop the prefix cache when idle (#811)

      Show description for e0a7e23
      jason-fxz

      authored
      e0a7e23View commit details

      Copy full SHA for e0a7e23

      View code at this pointBrowse repository at this point

    • feat(gguf): read NVFP4 token_embd rows in the vendored get_rows (#801)

      Show description for 20d5d93
      iamanishx

      authored
      20d5d93View commit details

      Copy full SHA for 20d5d93

      View code at this pointBrowse repository at this point

    • feat(gguf): serve qwen3vl and qwen3vlmoe gguf checkpoints text only (#807)

      void-mckenzie

      authored
      ed359eeView commit details

      Copy full SHA for ed359ee

      View code at this pointBrowse repository at this point

    • perf(moe): fuse sigmoid + selection-bias routing into a triton kernel (#804)

      Show description for 19a60a5
      jason-fxz

      authored
      19a60a5View commit details

      Copy full SHA for 19a60a5

      View code at this pointBrowse repository at this point
11. Commits on Oct 22, 2026

    • perf(moe): skip inactive block-fp8 decode routes (#665)

      taking-lying-flat

      authored
      7974c9cView commit details

      Copy full SHA for 7974c9c

      View code at this pointBrowse repository at this point

    • feat(gguf): take image input on Qwen and Gemma-4 GGUF checkpoints from an mmproj (#803)

      Show description for 609191d
      jason-fxz

      authored
      609191dView commit details

      Copy full SHA for 609191d

      View code at this pointBrowse repository at this point

    • feat(gguf): serve DeepSeek-V4 Flash and Pro GGUF checkpoints on deepseek4 (#774)

      Show description for 546c5fe
      void-mckenzie

      authored
      546c5feView commit details

      Copy full SHA for 546c5fe

      View code at this pointBrowse repository at this point