Skip to content

Enable RL training from base model #44

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

Closed
garyzhang99 wants to merge 2 commits into from
Closed

Enable RL training from base model #44

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6e8ab38
add use base format
May 21, 2025
5251a79
remove added tokenzier to make the pr only have single change
May 21, 2025
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
2 changes: 2 additions & 0 deletions trinity/common/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,8 @@ class FormatConfig:
# for unpaired preference dataset
label_key: str = ""

use_base_format: bool = False
Copy link
Collaborator

@pan-x-c pan-x-c May 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Using a new workflow type can do the same thing. Don't add a new field here.



@dataclass
class GenerationConfig:
Expand Down
19 changes: 17 additions & 2 deletions trinity/common/workflows/workflow.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -190,12 +190,27 @@ def format_messages(self):
messages.append({"role": "assistant", "content": self.reply_prefix})
return messages

def format_prompt(self):
Copy link
Collaborator

@pan-x-c pan-x-c May 23, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Add a new workflow named BaseModelWorkflow may be better.

prompt_text = ""
if self.system_prompt:
prompt_text += self.system_prompt
prompt_text += "\nTask:\n" + self.task_desc + "\nResponse:\n"
else:
prompt_text += "\nTask:\n" + self.task_desc + "\nResponse:\n"
return prompt_text

def run(self) -> List[Experience]:
# TODO: Optimize the generate function
messages = self.format_messages()
if self.format_args.use_base_format:
prompt_text = self.format_prompt()
else:
messages = self.format_messages()

logger.debug("start chat")
responses = self.model.chat(messages, **self.rollout_args)
if self.format_args.use_base_format:
responses = self.model.generate([prompt_text], **self.rollout_args)
else:
responses = self.model.chat(messages, **self.rollout_args)
for response in responses:
reward = self.reward_fn( # type: ignore [misc]
response=response.response_text, # type: ignore [arg-type]
Expand Down