Skip to content

[usability] make dpo eval dataset optional #889

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
research4pan merged 1 commit into from
Aug 10, 2024
Merged

[usability] make dpo eval dataset optional #889

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions src/lmflow/args.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1344,6 +1344,9 @@ class DPOAlignerArguments:
run_name: Optional[str] = field(
default="dpo", metadata={"help": "The name of the run."}
)
eval_dataset_path: Optional[str] = field(
default=None, metadata={"help": "The path of the eval dataset."}
)


@dataclass
Expand Down
19 changes: 11 additions & 8 deletions src/lmflow/pipeline/dpo_aligner.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,6 +71,8 @@ def __init__(self, model_args, data_args, aligner_args):
self.model_args = model_args
self.data_args = data_args
self.aligner_args = aligner_args
self.train_dataset = None
self.eval_dataset = None

def _initialize_trainer(self, model, tokenizer):
peft_config = LoraConfig(
Expand Down Expand Up @@ -118,7 +120,7 @@ def _initialize_trainer(self, model, tokenizer):
args=training_args,
beta=self.aligner_args.beta,
train_dataset=self.train_dataset,
eval_dataset=self.eval_dataset,
eval_dataset=self.eval_dataset if self.eval_dataset else None,
tokenizer=tokenizer,
peft_config=peft_config,
max_prompt_length=self.aligner_args.beta,
Expand All @@ -136,13 +138,14 @@ def _load_dataset(self):
and len(x["prompt"]) + len(x["rejected"]) <= self.aligner_args.max_length
)
# load evaluation set
self.eval_dataset = get_paired_dataset(data_root=self.data_args.dataset_path,
data_dir="test",
sanity_check=True)
self.eval_dataset = self.eval_dataset.filter(
lambda x: len(x["prompt"]) + len(x["chosen"]) <= self.aligner_args.max_length
and len(x["prompt"]) + len(x["rejected"]) <= self.aligner_args.max_length
)
if self.aligner_args.eval_dataset_path:
self.eval_dataset = get_paired_dataset(data_root=self.aligner_args.eval_dataset_path,
data_dir="test",
sanity_check=True)
self.eval_dataset = self.eval_dataset.filter(
lambda x: len(x["prompt"]) + len(x["chosen"]) <= self.aligner_args.max_length
and len(x["prompt"]) + len(x["rejected"]) <= self.aligner_args.max_length
)

def align(self, model, dataset, reward_model):
tokenizer = model.get_tokenizer()
Expand Down