lm-sys / lm-sys/FastChat

Difference between the prompts of v0 and v1

Open
#819 0 comments 2 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

Thanks for your great job! I have checked the code of v0 and v1, and found some differences between the prompts:
```python
# Vicuna-v0
sep = "###"
train_prompt = "system.### Human: xxx.\n### Assistant: yyy.\n### Human: xxx.\n### Assistant: yyy.\n"
target = " ### ### Assistant: yyy.\n### ### Assistant: yyy.\n"
eval_prompt = "system.###Human: xxx.###Assistant:"

# Vicuna-v1
sep = " "
sep2 = ""
train_prompt = "system. Human: xxx. Assistant: yyy.Human: xxx. Assistant: yyy."
target = " yyy. yyy."
eval_prompt = "system. Human: xxx. Assistant:"
```

Is there a bug in the eval_prompt of v0? It seems the " " after "###" and "\n" are ignored.
Besides, is it essential to predict the "###" before the " Human" when training?

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

Start with the v0 and v1 prompt code shown in the issue, focusing on sep, sep2, train_prompt, target, and eval_prompt. Confirm the prompt and token differences, then determine whether the v0 spacing is intentional and whether predicting ### before Human is required; no file or test is named in the report.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.