AdaptLLM/Adapt-MLLM-to-Domains

Adapting Multimodal Large Language Models to Domains via Post-Training

This repository provides an implementation preview of our paper: On Domain-Specific Post-Training for Multimodal Large Language Models.

We investigate domain adaptation of MLLMs through post-training, focusing on data synthesis, training pipelines, and task evaluation. (1) Data Synthesis: Using open-source models, we develop a visual instruction synthesizer that effectively generates diverse visual instruction tasks from domain-specific image-caption pairs. Our synthetic tasks surpass those generated by manual rules, GPT-4, and GPT-4V in enhancing the domain-specific performance of MLLMs. (2) Training Pipeline: While the two-stage training--initially on image-caption pairs followed by visual instruction tasks--is commonly adopted for developing general MLLMs, we apply a single-stage training pipeline to enhance task diversity for domain-specific post-training. (3) Task Evaluation: We conduct experiments in two domains, biomedicine and food, by post-training MLLMs of different sources and scales (e.g., Qwen2-VL-2B, LLaVA-v1.6-8B, Llama-3.2-11B), and then evaluating MLLM performance on various domain-specific tasks.

Updates

• [2024/11/29] Released our paper.

About

AdaMLLM represents our latest advancement in building domain-specific foundation models through post-training on synthetic supervised tasks derived from unsupervised contexts.

• AdaptLLM: Adapt LLM to domains
  We employ rule-based methods to extract tasks from domain-specific corpora, reformatting them into reading comprehension tasks for continued pre-training. Our 7B finance model outperforms domain-specific models of much larger scales, such as BloombergGPT-50B.

• AdaMLLM: Adapt Multimodal LLM to domains
  We extend supervised task synthesis to multimodality, introducing a unified visual instruction synthesizer to extract instruction-response pairs from domain-specific image-caption pairs. Our synthetic tasks outperform those generated by manual rules, GPT-4, and GPT-4V in improving domain-specific performance for MLLMs.

Citation

If you find our work helpful, please cite us.

AdaptLLM (ICLR 2024)

@inproceedings{
adaptllm,
title={Adapting Large Language Models via Reading Comprehension},
author={Daixuan Cheng and Shaohan Huang and Furu Wei},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=y886UXPEZ0}
}