---
license: apache-2.0
pipeline_tag: image-to-text
---

# MMAlaya2

MMAlaya2 fine-tunes 20 LoRA modules based on the InternVL-Chat-V1-5 model. These fine-tuned LoRA modules are then merged with the InternVL-Chat-V1-5 model using the PEFT model merging method, TIES.

You can find the inference code [here](https://github.com/open-compass/VLMEvalKit/blob/main/vlmeval/vlm/mmalaya.py#L8).

The [MMBench](https://mmbench.opencompass.org.cn/) benchmark contains 20 categories in the `mmbench_dev_cn_20231003.tsv` dataset. For each category, we first use CoT (Chain of Thought) consistency with the InternVL-Chat-V1-5 model to prepare the training dataset. For specific categories like nature_relation, image_emotion, image_scene, action_recognition, and image_style, we analyze the bad cases made by the InternVL-Chat-V1-5 model. We then prepare images and QA text from online sources to address these issues.

After fine-tuning the 20 LoRAs, they are merged with the InternVL-Chat-V1-5 model using the TIES method. The average score on the `mmbench_test_cn_20231003.tsv` benchmark reached 82.2, which we found noteworthy. As a result, we are sharing this model publicly.

# License

This project is released under the MIT license, in alignment with the InternVL-Chat-V1-5 model's license. InternLM2, however, is licensed under the Apache-2.0 license.

# Citation

If you find this project useful in your research, please consider citing:

```bibtex
@misc{datacanvas2024mmalaya2,
    author = {DataCanvas Ltd.},
    title = {MMAlaya2},
    year = {2024},
    howpublished = {\url{https://huggingface.co/DataCanvas/MMAlaya2}},
}
```