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. 31 Dec, 2025 3 commits
  3. 29 Dec, 2025 1 commit
  4. 24 Dec, 2025 2 commits
  5. 22 Dec, 2025 2 commits
  6. 18 Dec, 2025 2 commits
  7. 02 Dec, 2025 1 commit
  8. 27 Nov, 2025 1 commit
  9. 26 Nov, 2025 1 commit
  10. 27 Oct, 2025 1 commit
  11. 16 Sep, 2025 1 commit
  12. 14 Sep, 2025 1 commit
  13. 04 Sep, 2025 1 commit
  14. 13 Aug, 2025 1 commit
  15. 10 Aug, 2025 1 commit
  16. 07 Aug, 2025 2 commits
  17. 06 Aug, 2025 2 commits
  18. 05 Aug, 2025 2 commits
  19. 31 Jul, 2025 1 commit
  20. 12 Jul, 2025 1 commit
  21. 07 Jul, 2025 1 commit
  22. 06 Jul, 2025 1 commit
  23. 30 Jun, 2025 1 commit
  24. 25 Jun, 2025 1 commit
  25. 13 Jun, 2025 3 commits
  26. 06 Jun, 2025 1 commit
  27. 03 Jun, 2025 1 commit
  28. 28 May, 2025 1 commit
  29. 24 May, 2025 2 commits