Preference Tuning with Mixup

MergeMix: A Unified Augmentation Paradigm for Visual and Multi-Modal Understanding

Results of MLLM Benchmark

LLaVA
Qwen

Results of Calibration on MLLMs

Calibration LLaVA

More Results of MLLM Benchmark

LLaVA 1

Different token merge ratios of inference comparison results with augmentations. AVG: The average of the nine benchmarks for comprehensive comparison, except for MME.

LLaVA 2

Different token merge ratios of inference comparison results with augmentations. AVG: The average of the nine benchmarks for comprehensive comparison, except for MME.

LLaVA 3

Different token merge ratios of inference comparison results with augmentations. AVG: The average of the nine benchmarks for comprehensive comparison, except for MME.

Home Page