open-compass / open-compass/VLMEvalKit

Unable to reproduce any of the Gemma 3 results

Open
#1,145 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
4.4k
Forks
768
Avg merge
1d 10h
Merged PRs (30d)
17

Description

Hi,

I'm working on re creating some gemma 3 benchmark results. But the scores I get are always very far from the papers or the VLMEvalKit website. It's almost as if the chat template is wrong or something. Example : Running the ChartQA_TEST dataset on Gemma3-4B gives an overall score of 7.2 instead of 30-ish (Also, the paper announced around 70 for this test so this is another mystery). Same for the other datasets I've tried.

Can anyone re-create any gemma 3 results ?

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 Gemma 3 evaluation path and reproduce the ChartQA_TEST run described in the issue, comparing its 7.2 score with the paper and VLMEvalKit website results. Done means identifying the source of the discrepancy or documenting a confirmed, reproducible result across the reported datasets.

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
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.