1. 21 Jan, 2026 1 commit
    • laibao's avatar
      perf(qwen3): 融合 q/k RMSNorm + RoPE · 7cd7bf8a
      laibao 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
      7cd7bf8a
  2. 13 Nov, 2025 1 commit
  3. 12 Jul, 2025 1 commit
  4. 11 Jul, 2025 1 commit
  5. 10 Jul, 2025 1 commit
  6. 04 Jul, 2025 1 commit
  7. 27 Jun, 2025 1 commit
  8. 19 Jun, 2025 1 commit
  9. 13 Jun, 2025 1 commit
  10. 11 Jun, 2025 1 commit
  11. 03 Jun, 2025 1 commit
  12. 15 May, 2025 1 commit
  13. 07 May, 2025 1 commit
  14. 24 Apr, 2025 1 commit
  15. 07 Apr, 2025 1 commit