only process eval dataset for packing if not None
This commit is contained in:
@@ -182,9 +182,10 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
|
||||
train_dataset = train_dataset.filter(drop_long).map(
|
||||
add_position_ids, num_proc=os.cpu_count()
|
||||
)
|
||||
eval_dataset = eval_dataset.filter(drop_long).map(
|
||||
add_position_ids, num_proc=os.cpu_count()
|
||||
)
|
||||
if eval_dataset:
|
||||
eval_dataset = eval_dataset.filter(drop_long).map(
|
||||
add_position_ids, num_proc=os.cpu_count()
|
||||
)
|
||||
if cfg.sample_packing_eff_est:
|
||||
total_num_tokens = (
|
||||
cfg.total_num_tokens
|
||||
|
||||
Reference in New Issue
Block a user