1. 18 Apr, 2026 1 commit
  2. 15 Apr, 2026 1 commit
  3. 10 Apr, 2026 1 commit
  4. 07 Apr, 2026 1 commit
  5. 03 Apr, 2026 3 commits
  6. 02 Apr, 2026 2 commits
  7. 26 Mar, 2026 1 commit
    • laibao's avatar
      feat(v1 attention): 为 ROCm FlashAttention 接入 unified kv layout,并打通... · ea9b8584
      laibao authored
      feat(v1 attention): 为 ROCm FlashAttention 接入 unified kv layout,并打通 mm_prefix、qq_bias 与 use_alibi_sqrt 透传
      在 ROCm FlashAttention 后端增加 unified KV layout 选择逻辑
      接入 unified varlen kernel 调用路径
      在 FlashAttention metadata 中补充 mm_prefix_range 与 qq_bias 透传
      ea9b8584
  8. 24 Mar, 2026 1 commit
  9. 23 Mar, 2026 1 commit
  10. 16 Mar, 2026 2 commits
  11. 12 Mar, 2026 3 commits
  12. 04 Mar, 2026 2 commits
  13. 03 Mar, 2026 1 commit
  14. 02 Mar, 2026 1 commit
  15. 26 Feb, 2026 1 commit
  16. 24 Feb, 2026 1 commit
    • laibao's avatar
      • perf(v1): 增加可选的快速 token-id 拷贝路径 · d3a95d54
      laibao authored
        - 新增环境变量 `VLLM_V1_FAST_TOKEN_ID_COPY`(默认关闭)
        - 在 `CachedRequestState` 中缓存 int32 的 prompt token ids(numpy 数组)
        - 开启后在 `InputBatch` 中使用 `np.copyto` 拷贝 prompt/output token ids
      d3a95d54
  17. 08 Feb, 2026 1 commit
  18. 06 Feb, 2026 1 commit
  19. 05 Feb, 2026 1 commit
  20. 02 Feb, 2026 1 commit
  21. 27 Jan, 2026 1 commit
  22. 26 Jan, 2026 3 commits
  23. 25 Jan, 2026 1 commit
  24. 24 Jan, 2026 4 commits
  25. 23 Jan, 2026 4 commits