423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 03:40:34 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix #2777 #2895

Browse Source
This commit is contained in:
hiyouga
2024-03-20 17:59:45 +08:00
parent 7b8f502901
commit 9bec3c98a2
12 changed files with 104 additions and 48 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
src/llmtuner/train/dpo/workflow.py
View File

@@ -7,7 +7,7 @@ from ...extras.constants import IGNORE_INDEX
from ...extras.ploting import plot_loss
from ...hparams import ModelArguments
from ...model import load_model, load_tokenizer
from ..utils import create_custom_optimzer, create_modelcard_and_push, create_ref_model
from ..utils import create_modelcard_and_push, create_ref_model
from .collator import DPODataCollatorWithPadding
from .trainer import CustomDPOTrainer
@@ -44,18 +44,17 @@ def run_dpo(
training_args.remove_unused_columns = False # important for pairwise dataset
# Initialize our Trainer
optimizer = create_custom_optimzer(model, dataset, training_args, finetuning_args)
trainer = CustomDPOTrainer(
beta=finetuning_args.dpo_beta,
loss_type=finetuning_args.dpo_loss,
ftx_gamma=finetuning_args.dpo_ftx,
finetuning_args=finetuning_args,
model=model,
ref_model=ref_model,
args=training_args,
tokenizer=tokenizer,
data_collator=data_collator,
callbacks=callbacks,
optimizers=(optimizer, None),
**split_dataset(dataset, data_args, training_args),
)