Unstable results on MT-bench
Open
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 39.5k
- Forks
- 4.8k
- PR merge metrics
- No merged PRs in 30d
Description
In the MT-bench evaluation code, the temperature of gpt4 is set to 0, but everytime I run gen_judgment.py, I get a different score for the same model answer file. The difference can be as large as 0.2. I worder if this is normal.
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the MT-bench evaluation code in gen_judgment.py and run it repeatedly on the same model answer file to reproduce the score variation. Trace how the GPT-4 temperature setting is applied and determine whether repeated runs should be stable; done means explaining the variation or making the results consistent and verifying the behavior across repeated runs.
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