Skip to content

Fix multi-turn logprobs #170

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
pan-x-c merged 2 commits into from
Aug 7, 2025
Merged

Fix multi-turn logprobs #170

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
22377d3
fix bugs
pan-x-c Aug 7, 2025
95ddf73
fix logprobs
pan-x-c Aug 7, 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
1 change: 1 addition & 0 deletions tests/common/vllm_test.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -199,6 +199,7 @@ async def test_generate(
self.assertTrue(
torch.equal(result_dict["assistant_masks"][0][prompt_length:], exp.action_mask)
)
self.assertTrue(exp.logprobs.shape[0] == exp.tokens.shape[0] - prompt_length)
self.assertTrue(torch.equal(result_dict["input_ids"][0], exp.tokens))
self.assertRaises(ValueError, self.model_wrapper.get_openai_client)
if self.config.explorer.rollout_model.enable_history:
Expand Down
9 changes: 5 additions & 4 deletions trinity/common/models/vllm_model.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,7 +172,8 @@ async def generate(self, prompt: str, **kwargs) -> Sequence[Experience]:
return experiences

async def logprobs(self, token_ids: List[int]) -> torch.Tensor:
"""Calculate the logprobs of the given tokens in async.
"""Calculate the logprobs of the given tokens in async. Please slice the result carefully
to align with the actual response length.

Args:
token_ids (List[int]): The input token ids (seq_length).
Expand Down Expand Up @@ -217,11 +218,11 @@ async def convert_messages_to_experience(self, messages: List[dict]) -> Experien
self.chat_template = self.tokenizer.get_chat_template()
token_ids, action_mask, prompt_length = self.action_mask_method(
self.tokenizer, messages, self.chat_template
)
logprobs = await self.logprobs(token_ids=token_ids.tolist())
) # (seq_length, ), (seq_length, )
logprobs = await self.logprobs(token_ids=token_ids.tolist()) # (seq_length - 1,)
return Experience(
tokens=token_ids,
logprobs=logprobs,
logprobs=logprobs[prompt_length - 1 :],
prompt_length=prompt_length,
action_mask=action_mask[prompt_length:], # Exclude the prompt tokens
)
Expand Down