Compressing visual tokens in vision-language models: 3x more requests per GPU on Qwen2-VL
• 1
AI & ML interests
None defined yet.
Articles
pinned
Running
Datasent
🟪
Running on T4
Agents
2
Polygen RAG -- Encode Once, Query Many
🗂
A visual corpus encoded once, queried many times
Sleeping
2
Datasent's Polygen Tokenization Demo
📈
Tokenize any file into a compact Polygen token
Running on T4
Agents
2
Polygen for Qwen2-VL
🎯
Compress visual tokens. Same answer, proven per image.