Skip to content

add seed in DPO for reproduce the training result. disable hpu graph … #646

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
regisss merged 2 commits into from
Jan 22, 2024
Merged

add seed in DPO for reproduce the training result. disable hpu graph … #646

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
65d7a68
add seed in DPO for reproduce the training result. disable hpu graph …
sywangyi Jan 18, 2024
fd7b0e1
Refinements
regisss Jan 22, 2024
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
12 changes: 11 additions & 1 deletion examples/trl/dpo.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -9,6 +9,7 @@

from optimum.habana import GaudiConfig, GaudiTrainingArguments
from optimum.habana.trl import GaudiDPOTrainer
from optimum.habana.utils import set_seed


# Define and parse arguments.
Expand Down Expand Up @@ -80,6 +81,9 @@ class ScriptArguments:
"https://github.com/huggingface/transformers/issues/22482#issuecomment-1595790992"
},
)
seed: Optional[int] = field(
default=0, metadata={"help": "Random seed that will be set at the beginning of training."}
)


def get_stack_exchange_paired(
Expand Down Expand Up @@ -129,6 +133,7 @@ def return_prompt_and_responses(samples) -> Dict[str, str]:
if __name__ == "__main__":
parser = HfArgumentParser(ScriptArguments)
script_args = parser.parse_args_into_dataclasses()[0]

# 1. initialize training arguments:
training_args = GaudiTrainingArguments(
per_device_train_batch_size=script_args.per_device_train_batch_size,
Expand All @@ -151,9 +156,14 @@ def return_prompt_and_responses(samples) -> Dict[str, str]:
run_name="dpo_llama2",
use_habana=True,
use_lazy_mode=True,
use_hpu_graphs_for_training=True,
use_hpu_graphs_for_training=not script_args.gradient_checkpointing,
use_hpu_graphs_for_inference=True,
seed=script_args.seed,
)

# Set seed before initializing model.
set_seed(training_args.seed)

# 2. load a pretrained model
model = AutoModelForCausalLM.from_pretrained(
script_args.model_name_or_path,
Expand Down