5).
Training-free vision token compression method tailored for linear attention architectures
V-CoLA preserves 99.5% of baseline performance with 50% of vision tokens and achieves up to 6.15x prefill speedup on hybrid vision-language models.
Chinese Tech
Hao Jiang · Yiru Mao · Tianpeng Bu · Hao Zhou · Hongtao Duan · Wang Jing · +6 more
Alibaba Cloud Computing, Alibaba Group
Research Digest··3 min read
Jiang et al.
Why this paper
From Alibaba Cloud Computing, Alibaba Group
In one line
V-CoLA compresses vision tokens in linear-attention VLMs, retaining 99.5% performance with 50% tokens and achieving 1.86-6.15x speedup.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
Concepts
§