All pull requests

Pull requests

• [Perf][PersonaPlex] Continuously batch streaming Mimi decode

  #2596·nightdecode opened Oct 7, 2026·3 of 5 tasks·

  ·

• [Perf][MiniCPM-o] Avoid SHM for same-process image encoder inputs

  #2595·shaoxiawjc opened Oct 7, 2026·

  ·
  1

• [Perf][PersonaPlex] Fuse Depformer pointwise ops with CuTe DSL

  #2594·LiuYuHui opened Oct 7, 2026·5 tasks·

  ·

• feat(voxt): native Cohere Transcribe 03-2026 runtime

  #2593·khazic opened Oct 7, 2026Collaborator·2 of 5 tasks·

  ·

• feat(voxt): native Whisper small fp16 support

  #2592·khazic opened Oct 7, 2026Collaborator·1 of 4 tasks·

  ·

• feat(voxt): native Whisper large-v3 fp16 support

  #2591·khazic opened Oct 7, 2026Collaborator·1 of 4 tasks·

  ·

• feat(voxt): native Whisper large-v3-turbo runtime

  #2590·khazic opened Oct 7, 2026Collaborator·1 of 4 tasks·

  ·
  1

• [Perf][MiniCPM-o][Video] Parallelize video checkpoint preprocessing

  #2589·charliechenye opened Oct 7, 2026Collaborator·4 of 5 tasks·

  ·
  1

• [Voxt] Run Sortformer v2.1 speaker diarization on the native runtime

  run-ci
  #2588·JiaxinD opened Oct 7, 2026Collaborator·4 tasks done·

  ·

• [Voxt] Run Silero VAD v6 on the native runtime

  run-ci
  #2587·JiaxinD opened Oct 7, 2026Collaborator·4 tasks done·

  ·

• [Voxt] Run Qwen3-ASR 1.7B 6bit and 8bit on the native runtime

  run-ci
  #2586·JiaxinD opened Oct 7, 2026Collaborator·4 tasks done·

  ·

• [Fix][Nemotron-VoiceChat] Stop the Talker input rows leaking GPU memory

  #2585·XinhaoTheo opened Oct 7, 2026Contributor·3 of 4 tasks·

  ·

• [Fix] Apply the server's media policy to chat media

  run-ci
  #2584·rwang5203 opened Oct 6, 2026Collaborator·

  ·
  1

• [Fix] Require the admin key on the profiler routes

  run-ci
  #2583·rwang5203 opened Oct 6, 2026Collaborator·

  ·
  1

• [Fix][Router] Require the admin key to add, update or remove workers

  run-ci
  #2581·rwang5203 opened Oct 6, 2026Collaborator·

  ·
  1

• [Voxt] Native C++ Qwen3-ASR runtime on MLX (no Python runtime)

  run-ci
  #2577·JiaxinD opened Oct 6, 2026Collaborator·4 tasks done·

  ·

• [Perf] Support seeded sampling under thinker async decode

  #2573·YifanLi3 opened Oct 6, 2026·3 of 5 tasks·

  ·

• [Docs] Launch the MiniCPM-o browser demo from the cookbook

  #2572·mtilyxuegao opened Oct 6, 2026Collaborator·2 of 4 tasks·

  ·

• [Nemotron] Add ASR reproduction tools and restore support contributor credits

  #2571·Lenoplus42 opened Oct 6, 2026Collaborator·

  ·

• [Benchmark] Support configurable SocialOmni judges

  run-ci
  #2569·teeryxie opened Oct 6, 2026Contributor·

  ·
  9

• [MiniCPM-o] Replay the speech flow's chunk loop as CUDA graphs

  #2568·lijrjyan opened Oct 6, 2026Collaborator·10/10·4 of 5 tasks·

  ·

• [Scheduling] Keep session units with embedding spans out of the radix tree

  #2567·lijrjyan opened Oct 6, 2026Collaborator·9/10·2 of 5 tasks·

  ·

• [MiniCPM-o] Size duplex KV pools from max_sessions

  #2566·lijrjyan opened Oct 6, 2026Collaborator·8/10·4 of 5 tasks·

  ·

• [MiniCPM-o] Use decode CUDA graphs in duplex sessions

  #2565·lijrjyan opened Oct 6, 2026Collaborator·7/10·4 of 5 tasks·

  ·

• [MiniCPM-o] Expose speech settings and warm vocoder plans

  #2564·lijrjyan opened Oct 6, 2026Collaborator·6/10·4 of 5 tasks·

  ·
  1