1. 08 Jan, 2026 1 commit
  2. 07 Jan, 2026 1 commit
    • laibao's avatar
      perf: 加速 v1 InputBatch.add_request 的 token_ids 拷贝 · 8da572a9
      laibao authored
      新增环境变量开关 VLLM_V1_FAST_TOKEN_ID_COPY(默认关闭)。开启后在 CachedRequestState 中缓存 prompt_token_ids 的 np.int32,并在 add_request 里用 np.copyto 写入 token_ids_cpu,避免长 prompt 场景反复 list->NumPy 转换开销(尤其是抢占/反复进出 batch 时)
      8da572a9
  3. 24 Dec, 2025 1 commit
  4. 22 Dec, 2025 1 commit
  5. 29 Oct, 2025 1 commit
  6. 28 Oct, 2025 1 commit
  7. 20 Oct, 2025 1 commit
  8. 12 Oct, 2025 1 commit
  9. 11 Oct, 2025 2 commits
  10. 01 Oct, 2025 2 commits
  11. 28 Sep, 2025 2 commits
  12. 26 Sep, 2025 5 commits
  13. 25 Sep, 2025 9 commits
  14. 24 Sep, 2025 5 commits
  15. 23 Sep, 2025 7 commits