Fix DataCollatorForLanguageModeling (#8621)
This commit is contained in:
@@ -261,7 +261,7 @@ class DataCollatorForLanguageModeling:
|
|||||||
batch["input_ids"], special_tokens_mask=special_tokens_mask
|
batch["input_ids"], special_tokens_mask=special_tokens_mask
|
||||||
)
|
)
|
||||||
else:
|
else:
|
||||||
labels = batch["input_ids"]
|
labels = batch["input_ids"].clone()
|
||||||
if self.tokenizer.pad_token_id is not None:
|
if self.tokenizer.pad_token_id is not None:
|
||||||
labels[labels == self.tokenizer.pad_token_id] = -100
|
labels[labels == self.tokenizer.pad_token_id] = -100
|
||||||
batch["labels"] = labels
|
batch["labels"] = labels
|
||||||
|
|||||||
Reference in New Issue
Block a user