1. 21 Jan, 2026 1 commit
    • zhuwenwen's avatar
      perf(qwen3): 融合 q/k RMSNorm + RoPE · 3cd3d1e6
      zhuwenwen authored
      新增 VLLM_USE_FUSED_RMS_ROPE 分支,走 fused 路径
      注册 torch.ops.vllm.rms_rotary_embedding_fuse(direct_register_custom_op)
      cos_sin_cache 自动转 device/dtype 并缓存,避免每次重复拷贝
      3cd3d1e6
  2. 20 Nov, 2025 1 commit
  3. 21 Sep, 2025 1 commit
  4. 11 Sep, 2025 1 commit
  5. 26 Aug, 2025 1 commit
  6. 12 Aug, 2025 1 commit
  7. 07 Aug, 2025 1 commit
  8. 02 Aug, 2025 1 commit
  9. 18 Jul, 2025 1 commit
  10. 12 Jul, 2025 1 commit
  11. 11 Jul, 2025 1 commit
  12. 10 Jul, 2025 1 commit
  13. 04 Jul, 2025 1 commit
  14. 27 Jun, 2025 1 commit
  15. 19 Jun, 2025 1 commit
  16. 13 Jun, 2025 1 commit
  17. 11 Jun, 2025 1 commit
  18. 03 Jun, 2025 1 commit
  19. 15 May, 2025 1 commit
  20. 07 May, 2025 1 commit
  21. 24 Apr, 2025 1 commit
  22. 07 Apr, 2025 1 commit