transformers
fix device-to-host sync in flash attention
#47771
Open

Commits
  • add is_packed_sequence to flash attention kwargs
    DhanushPillay committed 14 days ago
  • thread query_length into _is_packed_sequence for decode short-circuit
    DhanushPillay committed 14 days ago
  • hoist prefill packed sequence check out of per-layer path
    DhanushPillay committed 14 days ago
  • Merge branch 'main' into perf/remove-fa-sync
    DhanushPillay committed 14 days ago
  • Merge branch 'main' into perf/remove-fa-sync
    DhanushPillay committed 13 days ago
  • Merge branch 'main' into perf/remove-fa-sync
    DhanushPillay committed 13 days ago
  • Merge branch 'main' into perf/remove-fa-sync
    DhanushPillay committed 10 days ago
  • fix bigbird_pegasus qa gradient checkpointing test
    DhanushPillay committed 9 days ago
  • style: fix trailing whitespace and humanize comment
    DhanushPillay committed 9 days ago
  • Merge branch 'main' into perf/remove-fa-sync
    DhanushPillay committed 6 days ago
Loading