[CPU] Fix model failures on Xeon (#29497)
This commit is contained in:
@@ -130,6 +130,9 @@ class EagerRunner(BaseRunner):
|
||||
if is_encoder_decoder
|
||||
else 0
|
||||
),
|
||||
encoder_lens_dtype=(
|
||||
torch.int64 if torch.device(mr.device).type == "cpu" else torch.int32
|
||||
),
|
||||
dp_size=sa.dp_size,
|
||||
)
|
||||
# Eager has no capture step, so warm up here (run-once via mr._kernel_warmed_up).
|
||||
|
||||
Reference in New Issue
Block a user