benchmark_eval.md

Welcome to the benchmark evaluation page!

The evaluation pipeline is designed to be one-clickable and easy to use. However, you may encounter some problems when running the models (e.g. LLaVA, LLaMA-Adapter) that require you to clone their repo to local path. Please feel free to contact us if you have any questions.

We support the following benchmarks:

• MagnifierBench
• MMBench
• MM-VET
• MathVista
• POPE
• MME
• SicenceQA
• SeedBench

And following models:

• LLaVA
• Fuyu
• OtterHD
• Otter-Image
• Otter-Video
• Idefics
• LLaMA-Adapter
• Qwen-VL

many more, see /pipeline/benchmarks/models

https://github.com/Luodian/Otter/tree/main/pipeline/benchmarks/models

Create a yaml file benchmark.yaml with below content:

datasets:
  - name: magnifierbench
    split: test
    data_path: Otter-AI/MagnifierBench
    prompt: Answer with the option letter from the given choices directly.
    api_key: [You GPT-4 API]
  - name: mme
    split: test
  - name: pope
    split: test
    default_output_path: ./logs
  - name: mmvet
    split: test
    api_key: [You GPT-4 API]
    gpt_model: gpt-4-0613
  - name: mathvista
    split: test
    api_key: [You GPT-4 API]
    gpt_model: gpt-4-0613
  - name: mmbench
    split: test
models:
  - name: fuyu
    model_path: adept/fuyu-8b

Then run

python -m pipeline.benchmarks.evaluate --confg benchmark.yaml