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 4 commits
  3. 30 Dec, 2025 2 commits
  4. 29 Dec, 2025 6 commits
  5. 24 Dec, 2025 2 commits
  6. 22 Dec, 2025 2 commits
  7. 18 Dec, 2025 2 commits
  8. 09 Dec, 2025 1 commit
  9. 27 Nov, 2025 1 commit
  10. 26 Nov, 2025 1 commit
  11. 09 Oct, 2025 1 commit
  12. 24 Sep, 2025 1 commit
  13. 16 Sep, 2025 1 commit
  14. 04 Sep, 2025 2 commits
  15. 29 Aug, 2025 1 commit
  16. 27 Aug, 2025 1 commit
  17. 25 Aug, 2025 2 commits
  18. 18 Aug, 2025 1 commit
  19. 16 Aug, 2025 2 commits
  20. 15 Aug, 2025 1 commit
  21. 11 Aug, 2025 1 commit