Skip to content
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

Add Optional ZeRO-3 Weight Gathering for GRPO in Sequence Generation #2667

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Add Optional ZeRO-3 Weight Gathering for GRPO in Sequence Generation #2667

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
13 changes: 13 additions & 0 deletions trl/trainer/grpo_config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,6 +58,11 @@ class GRPOConfig(TrainingArguments):
[`~transformers.TrainingArguments`].
beta (`float`, *optional*, defaults to `0.04`):
KL coefficient.

ds3_gather_for_generation (`bool`, *optional*, defaults to `True`):
This setting applies to DeepSpeed ZeRO-3. If enabled, the policy model weights are gathered for generation,
improving generation speed. However, disabling this option allows training models that exceed the VRAM
capacity of a single GPU, albeit at the cost of slower generation.
"""

# Parameters that control the model and reference model
Expand Down Expand Up @@ -110,3 +115,11 @@ class GRPOConfig(TrainingArguments):
default=0.04,
metadata={"help": "KL coefficient."},
)
ds3_gather_for_generation: bool = field(
default=True,
metadata={
"help": "This setting applies to DeepSpeed ZeRO-3. If enabled, the policy model weights are gathered for "
"generation, improving generation speed. However, disabling this option allows training models that "
"exceed the VRAM capacity of a single GPU, albeit at the cost of slower generation."
},
)
4 changes: 3 additions & 1 deletion trl/trainer/grpo_trainer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -322,7 +322,9 @@ def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=N
prompt_inputs["attention_mask"] = prompt_inputs["attention_mask"][:, -self.max_prompt_length :]

# Generate completions
with unwrap_model_for_generation(model, self.accelerator) as unwrapped_model:
with unwrap_model_for_generation(
self.model, self.accelerator, gather_deepspeed3_params=self.args.ds3_gather_for_generation
) as unwrapped_model:
prompt_completion_ids = unwrapped_model.generate(**prompt_inputs, generation_config=self.generation_config)
prompt_length = prompt_inputs["input_ids"].size(1)
completion_ids = prompt_completion_ids[:, prompt_length:]
Expand Down