open-compass / open-compass/VLMEvalKit

How to evaluate my fine-tuned Qwen-VL model on a locally downloaded dataset?

Open
#1,500 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 all,

Thanks very much for your great repo. Now I have the following scenario but I can't figure out what to do by only reading the documentations. I have a fine-tuned Qwen-VL model (locally) which can be loaded by transformers.AutoModelForVision2Seq.from_pretrained, and a HallusionBench dataset (to which I've made some modifications to the images inside) which can be loaded by datasets.load_from_disk. How to evaluate the model and test on my own local modified HallusionBench dataset? Also, is setting CUDA_VISIBLE_DEVICES enough for setting the GPUs to use, and how is batch size determined in the evaluation process?

Appreciations for your help!

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

The issue names no repository file, test, or evaluation entry point. Start with the documentation around HallusionBench and compare the expected inputs with a model loaded by AutoModelForVision2Seq.from_pretrained and data loaded by datasets.load_from_disk; document the local-dataset procedure, CUDA_VISIBLE_DEVICES behavior, and batch-size handling.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
computer-vision, documentation, machine-learning
Issue type
Documentation
Difficulty
3/5
Estimated time
1-2 days
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
45/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.