 Tri Dao
|
bedf877467
[CrossEntropy] Fix where labels address not aligned to 16 bytes
|
пре 2 месеци |
|
Tri Dao
|
8c20cfef49
[Rotary] Support qkv block layout from GQA
|
пре 3 месеци |
|
Tri Dao
|
c7f32a8409
[CrossEntropy] Support precomputed LSE
|
пре 3 месеци |
|
Tri Dao
|
d79f9b41a8
[CrossEntropy] Use online softmax to simplify implementation
|
пре 3 месеци |
 lancerts
|
22339db185
remove an unused import (#960)
|
пре 6 месеци |
|
Tri Dao
|
ec6d22143b
[CrossEntropy] Change ignored_index -> ignore_index
|
пре 7 месеци |
 Curtis "Fjord" Hawthorne
|
d8aacc510c
return z_loss (#768)
|
пре 10 месеци |
|
Tri Dao
|
08124c8f9c
[CrossEntropy] Implement logit_scale option
|
пре 1 година |
|
Tri Dao
|
aaa1474129
[CrossEntropy] Simplify the case of large vocab with Tensor Parallel
|
пре 1 година |
 Shijie
|
abf04a56e1
fix flash ce mp large vocab (#673)
|
пре 1 година |
|
Tri Dao
|
c79de85ffa
[CrossEntropy] Fix triton cross_entropy_loss IMA for >=2B elements
|
пре 1 година |
|
Tri Dao
|
5400fdc4ac
[CE] Implement CrossEntropyLoss in Triton
|
пре 1 година |
