1. 06 Feb, 2026 1 commit
    • laibao's avatar
      • feat(qwen3):新增 vLLM 内置 RMS+RoPE 融合算子,并支持 LightOp 后端切换 · 588538f5
      laibao authored
        - 在 vLLM _C 扩展中新增 rms_rotary_embedding_fuse(注册 op + CUDA kernel),减少对 LightOp 的硬依赖
        - 新增环境变量 VLLM_FUSED_RMS_ROPE_BACKEND=auto|vllm|lightop,auto 优先走 vLLM,缺失时回退 LightOp
        - 更新 Qwen3 / Qwen3-MoE 的 fused 路径按后端选择执行
        - 补充 tc_opt benchmark 结果解析脚本 benchmarks/tc_opt/test/parse_bench_results.py
      588538f5
  2. 14 Jan, 2026 1 commit
  3. 04 Jan, 2026 1 commit
  4. 16 Dec, 2025 1 commit
  5. 13 Nov, 2025 1 commit
  6. 28 Sep, 2025 2 commits
  7. 26 Sep, 2025 1 commit
  8. 21 Sep, 2025 1 commit
  9. 17 Sep, 2025 3 commits
  10. 13 Sep, 2025 1 commit
  11. 12 Sep, 2025 1 commit
  12. 01 Sep, 2025 1 commit
  13. 29 Aug, 2025 1 commit
  14. 25 Aug, 2025 1 commit
  15. 20 Aug, 2025 1 commit
  16. 19 Aug, 2025 1 commit
  17. 13 Aug, 2025 1 commit
  18. 12 Aug, 2025 1 commit
  19. 11 Aug, 2025 1 commit
  20. 09 Aug, 2025 1 commit
  21. 08 Aug, 2025 1 commit
  22. 07 Aug, 2025 1 commit
  23. 30 Jul, 2025 1 commit
  24. 17 Jul, 2025 1 commit
  25. 14 Jul, 2025 1 commit
  26. 12 Jul, 2025 1 commit
  27. 11 Jul, 2025 2 commits
  28. 10 Jul, 2025 1 commit
  29. 03 Jul, 2025 1 commit
  30. 30 Jun, 2025 1 commit
  31. 20 Jun, 2025 1 commit
  32. 12 Jun, 2025 1 commit
  33. 10 Jun, 2025 1 commit
  34. 04 Jun, 2025 1 commit
  35. 03 Jun, 2025 1 commit
  36. 26 May, 2025 1 commit