[Hotfix] Fix accuracy and align attention method api with Triton kernel (#5229)

* fix accuracy

* alignment in attention

* fix attention

* fix

* fix bugs

* fix bugs

* fix bugs
This commit is contained in:
Jianghai
2024-01-08 15:56:00 +08:00
committed by FrankLeeeee
parent fa4fbdbffb
commit e545a871b8
6 changed files with 168 additions and 107 deletions

View File

@@ -11,7 +11,6 @@ from colossalai.inference.config import InferenceConfig
from colossalai.inference.core.engine import InferenceEngine
from colossalai.testing import rerun_if_address_is_in_use, spawn
def setup_seed(seed):
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)