Skip to content

[Feature] Triton INT4 per-token-head KV cache quantization - #40835

Merged
tdoublep merged 72 commits into
vllm-project:mainfrom
JartX:refactor/prefill-fastpath-per-token-head-v2
Jun 24, 2026
Merged

tdoublep merged 72 commits into
vllm-project:mainfrom
JartX:refactor/prefill-fastpath-per-token-head-v2

Commits

  1. Commits on Apr 15, 2026

  2. Commits on Apr 16, 2026

    • added mixed

      committed
    • add per_token_head_attention kernel

      committed
    • pre-commit

      committed
    • missing seq_lens_cpu and query

      committed
    • working on kernel

      committed
    • working on kernel

      committed
    • precommit

      committed
    • precommit

      committed
    • prefill kernel

      committed
    • int8 wmma

      committed
    • int8 wmma

      committed
    • remove dead code

      committed
    • remove comment

      committed
    • remove comment

      committed
  3. Commits on Apr 19, 2026

  4. Commits on Apr 24, 2026

  5. Commits on May 4, 2026

  6. Commits on May 25, 2026

  7. Commits on May 28, 2026

  8. Commits on Jun 2, 2026

  9. Commits on Jun 3, 2026

  10. Commits on Jun 4, 2026

  11. Commits on Jun 8, 2026

  12. Commits on Jun 9, 2026

  13. Commits on Jun 10, 2026

  14. Commits on Jun 11, 2026

  15. Commits on Jun 12, 2026

  16. Commits on Jun 18, 2026

  17. Commits on Jun 22, 2026

  18. Commits on Jun 23, 2026

  19. Commits on Jun 24, 2026