Commit Verlauf

Autor SHA1 Nachricht Datum
  Tri Dao 5458c78e6d Remove sink token vor 3 Tagen
  Tri Dao 20b84d6363 Don't use IntraWGOverlap for hdim 64,512 vor 3 Tagen
  Tri Dao 6752d62aa4 Add dynamic splits vor 1 Woche
  Tri Dao ecdb528dea Make rotary test optional in FA3 vor 1 Woche
  Anton Vlasjuk a09abcd32d make seqused optional on top level interface (#1497) vor 2 Wochen
  Tri Dao 893a22ab57 Test varlen_q=True by default for kvcache vor 3 Wochen
  Tri Dao 4e8496a781 Support Qv vor 3 Wochen
  Tri Dao 2a204125ae Implement HeadDim_V != HeadDim_QK, support hdimQK=192, hdimV=128 vor 1 Monat
  Tri Dao 2ac6c986be Fix Sm80 tile_count_semaphore, adjust test tolerance vor 1 Monat
  Tri Dao 7a802796e1 Big refactor and update vor 1 Monat
  Kai Londenberg b443207c1f Paged Attention support for FA3 (#1268) vor 3 Monaten
  milesvant c1d146cbd5 Fix copy-paste error in hopper tests (#1279) vor 4 Monaten
  jayhshah a5a75274bc FA3 kvcache + split kv + gqa parallelization (#1236) vor 4 Monaten
  Ying Zhang be6c1b98c4 small fixes vor 5 Monaten
  Ying Zhang dff976a84a fixes vor 6 Monaten
  Ying Zhang 8cbc8a042f small fixes vor 5 Monaten
  Ying Zhang db80387343 Add seqused_q in fwd / bwd and seqused_k in bwd. vor 6 Monaten
  jayhshah c92ca63268 FA3 FP8 qkv descales + restore max offset for h128 causal + added sync for producer WG (#1173) vor 6 Monaten
  Ying Zhang 53537da422 add a unittest vor 6 Monaten
  Tri Dao c33de664a1 Fix import in test vor 7 Monaten
  Tri Dao bafe253042 [FA3] Bwd vor 7 Monaten
  jayhshah 5018ac6ac5 Fp8 kernel with "in-kernel" transpose of V in producer (#1100) vor 7 Monaten
  Tri Dao 3aae9c18c1 Revert "Changes For FP8 (#1075)" vor 7 Monaten
  ganeshcolfax 1899c970c8 Changes For FP8 (#1075) vor 7 Monaten
  Ying Zhang dfe1a59e4b Add var-seq-len to FA3 fp16 / bf16 fwd (#1072) vor 7 Monaten
  Tri Dao 74b0761ff7 [FA3] BF16 forward vor 7 Monaten
  Tri Dao 7f67966cc7 FA3 initial code release vor 7 Monaten