don't add position_ids for evals (#591)
This commit is contained in:
@@ -403,10 +403,11 @@ def process_datasets_for_packing(cfg, train_dataset, eval_dataset):
|
|||||||
|
|
||||||
if cfg.sample_packing:
|
if cfg.sample_packing:
|
||||||
train_dataset = train_dataset.map(add_position_ids, num_proc=os.cpu_count())
|
train_dataset = train_dataset.map(add_position_ids, num_proc=os.cpu_count())
|
||||||
if eval_dataset:
|
if cfg.eval_sample_packing is not False:
|
||||||
eval_dataset = eval_dataset.map(
|
if eval_dataset:
|
||||||
add_position_ids, num_proc=os.cpu_count()
|
eval_dataset = eval_dataset.map(
|
||||||
)
|
add_position_ids, num_proc=os.cpu_count()
|
||||||
|
)
|
||||||
return train_dataset, eval_dataset
|
return train_dataset, eval_dataset
|
||||||
|
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user