lm-sys / lm-sys/FastChat

Llama 3.1 chat template has <|begin_of_text|> encoded twice

Open
#3,535 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
39.5k
Forks
4.8k
PR merge metrics
No merged PRs in 30d

Description

In conversation.py (my comment):
```
elif self.sep_style == SeparatorStyle.LLAMA3:
# No! It's already added in encode_dialog_prompt chat_format.py
#ret = "<|begin_of_text|>"
if self.system_message:
ret += system_prompt

```
And in encode_dialog_prompt:

```
tokens = []
tokens.append(self.tokenizer.special_tokens["<|begin_of_text|>"])
for message in messages:

```

Not sure which one to keep, but only one of them for sure.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Compare the LLAMA3 branch in conversation.py with encode_dialog_prompt in chat_format.py. Trace how the Llama 3.1 conversation prompt is assembled, then ensure the <|begin_of_text|> token is added only once. Done means the generated prompt no longer contains the token twice; no test file is named in the issue.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
ai
Issue type
Bug
Difficulty
2/5
Estimated time
1-3 hours
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.