kirikir13's picture
Duplicate from whyisverysmart/Fourier-Qwen2.5-VL-3B-0.67
d36cd76
|
Raw
History Blame Contribute Delete
1.66 kB
metadata
license: apache-2.0
base_model:
  - Qwen/Qwen2.5-VL-3B-Instruct
tags:
  - vision-language-model
  - efficient-vlm
  - llava
  - qwen2-vl
  - token-compression
  - frequency-domain

Fourier-Qwen2.5-VL-3B-0.67

Official checkpoints for Fourier Compressor: Frequency-Domain Visual Token Compression for Vision-Language Models.

Model Details

Model Base Model Visual Tokens Compression Weights
Fourier-LLaVA-v1.5-7B-256 LLaVA-v1.5-7B 256 55.6% 🤗 HF
Fourier-LLaVA-v1.5-7B-144 LLaVA-v1.5-7B 144 75.0% 🤗 HF
Fourier-LLaVA-v1.5-7B-64 LLaVA-v1.5-7B 64 88.9% 🤗 HF
Fourier-LLaVA-v1.5-7B-36 LLaVA-v1.5-7B 36 93.8% 🤗 HF
Fourier-LLaVA-v1.5-13B-144 LLaVA-v1.5-13B 144 75.0% 🤗 HF
Fourier-Qwen2-VL-2B-0.67 Qwen2-VL-2B-Instruct Dynamic 55.6% 🤗 HF
Fourier-Qwen2.5-VL-3B-0.67 Qwen2.5-VL-3B-Instruct Dynamic 55.6% 🤗 HF

Links