don't need temp arg to distill method
This commit is contained in:
@@ -125,7 +125,6 @@ class LigerFusedLinearKLTopKLogprobFunction(LigerFusedLinearDistillationBase):
|
|||||||
target_token_ids_chunk,
|
target_token_ids_chunk,
|
||||||
target_logprobs_chunk,
|
target_logprobs_chunk,
|
||||||
target_mask_chunk,
|
target_mask_chunk,
|
||||||
temperature=temperature,
|
|
||||||
)
|
)
|
||||||
|
|
||||||
loss = weight_hard_loss * ce_loss + weight_soft_loss * soft_loss
|
loss = weight_hard_loss * ce_loss + weight_soft_loss * soft_loss
|
||||||
|
|||||||
Reference in New Issue
Block a user