Repository navigation
Towards Efficient Multimodal Large Language Models: A Survey on Token Compression (v1.0)
Latestπ Survey Paper
Towards Efficient Multimodal Large Language Models: A Survey on Token Compression
This release contains the first public version (v1.0) of our survey paper.
- Covers token compression methods for multimodal large language models (MLLMs)
- Provides a unified taxonomy across vision encoder, projector, LLM backbone, and hybrid approaches
- Includes analysis, comparisons, applications, and future directions
π Links
- π TechRxiv: https://www.techrxiv.org/your-paper-link
- π» Repository: https://github.com/yaolinli/MLLM-Token-Compression