423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 20:00:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

lazy image load

Browse Source
This commit is contained in:
hiyouga
2024-09-04 02:27:08 +08:00
parent 59d2b31e96
commit 47ea97fb1b
19 changed files with 353 additions and 366 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llamafactory/train/kto/workflow.py
View File

@@ -41,13 +41,14 @@ def run_kto(
):
tokenizer_module = load_tokenizer(model_args)
tokenizer = tokenizer_module["tokenizer"]
dataset_module = get_dataset(model_args, data_args, training_args, stage="kto", **tokenizer_module)
dataset_module, template = get_dataset(model_args, data_args, training_args, stage="kto", **tokenizer_module)
model = load_model(tokenizer, model_args, finetuning_args, training_args.do_train)
data_collator = KTODataCollatorWithPadding(
tokenizer=tokenizer,
template=template,
pad_to_multiple_of=8,
label_pad_token_id=IGNORE_INDEX if data_args.ignore_pad_token_for_loss else tokenizer.pad_token_id,
**tokenizer_module,
)
# Create reference model
@@ -57,7 +58,7 @@ def run_kto(
ref_model = create_ref_model(model_args, finetuning_args)
# Update arguments
training_args.remove_unused_columns = False # important for pairwise dataset
training_args.remove_unused_columns = False # important for multimodal and pairwise dataset
# Initialize our Trainer
trainer = CustomKTOTrainer(