1. 10 Feb, 2026 1 commit
    • zhuwenwen's avatar
      [feat] 适配block和channel fp8, 添加dp attention功能 · 2b7b1a31
      zhuwenwen authored
      [pd] 修改非堆成切分的判断, pd分离使用默认调度,set VLLM_USE_PD_SPLIT=1
      [perf](fused-moe): 预打包 Marlin W16A16 MoE 权重,降低 warmup 显存峰值
      [fix]修复丢弃MTP代码报错,pp+chunkprefill多并发input ids更新bug, 修复不开启融合图的断言错误, PP 场景 decode 阶段 token 被误丢弃导致卡住
      2b7b1a31
  2. 16 Aug, 2025 1 commit
  3. 15 Aug, 2025 1 commit
  4. 11 Aug, 2025 1 commit
  5. 05 Aug, 2025 1 commit
  6. 04 Aug, 2025 1 commit
  7. 26 Jun, 2025 1 commit
  8. 19 Jun, 2025 1 commit
  9. 16 Jun, 2025 1 commit
  10. 10 Jun, 2025 1 commit
  11. 07 Jun, 2025 1 commit
  12. 04 Jun, 2025 1 commit
  13. 03 Jun, 2025 1 commit
  14. 28 May, 2025 1 commit
  15. 23 May, 2025 3 commits
  16. 16 May, 2025 1 commit
  17. 14 May, 2025 1 commit
  18. 07 May, 2025 1 commit
  19. 06 May, 2025 1 commit
  20. 04 May, 2025 1 commit
  21. 25 Apr, 2025 1 commit
  22. 21 Apr, 2025 1 commit
  23. 17 Apr, 2025 1 commit
  24. 03 Apr, 2025 1 commit
  25. 02 Apr, 2025 1 commit
  26. 21 Mar, 2025 2 commits
  27. 14 Mar, 2025 1 commit
  28. 11 Mar, 2025 2 commits
  29. 09 Mar, 2025 1 commit
  30. 08 Mar, 2025 1 commit
  31. 03 Mar, 2025 1 commit
  32. 25 Feb, 2025 2 commits
  33. 24 Feb, 2025 2 commits
  34. 23 Feb, 2025 1 commit