lm-sys / lm-sys/FastChat

Inconsistency in Output Format of `get_prompt` Function in Llama-2

Open
#2,956 1 comment 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 my experience with the `get_prompt` function in the llama-2 conversation format, I've noticed a discrepancy in the output format. The expected format is `[INST] <>\n{your_system_message}\n<>\n\n{user_message_1} [/INST] {model_reply_1}`.

However, what the function actually returns is `[INST] <>\n{your_system_message}\n<>\n\n{user_message_1} [/INST] {model_reply_1}`.

This additional `` at the end diverges from the standard llama-2 format.

Reference:

https://github.com/lm-sys/FastChat/blob/5f46ff4b0c0e9b5f4d3378a7a19bf3974a19763b/fastchat/conversation.py#L130-L145

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

Open fastchat/conversation.py and inspect the get_prompt implementation around the referenced lines 130-145. Compare its Llama-2 output with the standard format in the issue, then verify that the returned prompt no longer contains the extra token at the end.

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
Clearly specified
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.