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. 05 Jan, 2026 3 commits
  4. 24 Dec, 2025 1 commit
  5. 22 Dec, 2025 2 commits
  6. 20 Dec, 2025 1 commit
  7. 12 Dec, 2025 1 commit
  8. 05 Dec, 2025 1 commit
  9. 04 Dec, 2025 3 commits
  10. 03 Dec, 2025 1 commit
  11. 02 Dec, 2025 1 commit
  12. 24 Nov, 2025 1 commit
  13. 21 Nov, 2025 1 commit
  14. 18 Nov, 2025 1 commit
  15. 17 Nov, 2025 1 commit
  16. 14 Nov, 2025 1 commit
  17. 13 Nov, 2025 2 commits
  18. 29 Oct, 2025 1 commit
  19. 28 Oct, 2025 1 commit
  20. 20 Oct, 2025 1 commit
  21. 12 Oct, 2025 1 commit
  22. 11 Oct, 2025 2 commits
  23. 03 Oct, 2025 1 commit
  24. 02 Oct, 2025 2 commits
  25. 01 Oct, 2025 4 commits
  26. 30 Sep, 2025 1 commit
  27. 29 Sep, 2025 1 commit
  28. 28 Sep, 2025 2 commits