1. 24 Dec, 2023 1 commit
  2. 23 Dec, 2023 1 commit
  3. 22 Dec, 2023 7 commits
  4. 20 Dec, 2023 4 commits
    • Tri Dao's avatar
      Format flash_attn_interface.py · bc28eacc
      Tri Dao authored
      bc28eacc
    • Tri Dao's avatar
      [Gen] Remove minor dead code · 0a146185
      Tri Dao authored
      0a146185
    • Sanghun Cho's avatar
      Support alibi, by Sanghun Cho from Kakao Brain · e4f726fc
      Sanghun Cho authored
      
      
      * hard-code alibi in fwd
      
      * use params.h as hun_heads
      
      * hard-code alibi in bwd
      
      * add alibi on/off option
      
      * compute alibi_start, ratio outside of kernels
      
      * fix minor merge conflict
      
      * add test_alibi.py
      
      * change apply_alibi() location before masking
      
      * add alibi in splitkv kernel
      
      * fix backward func # of returns
      
      * add out-of-bound check in apply_alibi()
      
      * update test_alibi.py
      
      * update test_alibi.py for kvcache
      
      * simplify alibi parameter interface
      
      * fix performance issue
      by computing alibi outside of branch
      
      * update test_flash_attn_varlen_func() for left padding
      
      * implement alibi_slopes (b, nh) loading
      
      * optimize apply_alibi() a bit
      
      * update test cases for alibi_slopes loading
      
      * reflect stylistic comments
      
      * disable "seqlenq_ngroups_swapped" when using alibi
      
      ---------
      Co-authored-by: default avatarmonk.detective <monk.detective@kakaobrain.com>
      e4f726fc
    • Tri Dao's avatar
      cd089597
  5. 17 Dec, 2023 2 commits
  6. 01 Dec, 2023 1 commit
  7. 28 Nov, 2023 2 commits
  8. 27 Nov, 2023 3 commits
  9. 20 Nov, 2023 6 commits
  10. 14 Nov, 2023 1 commit
  11. 13 Nov, 2023 1 commit
  12. 09 Nov, 2023 1 commit
  13. 24 Oct, 2023 2 commits
  14. 12 Oct, 2023 1 commit
  15. 09 Oct, 2023 1 commit
  16. 08 Oct, 2023 1 commit
  17. 05 Oct, 2023 2 commits
  18. 04 Oct, 2023 2 commits
  19. 03 Oct, 2023 1 commit