📝 Network Memory Footprint Compression Through Jointly Learnable Codebooks and Mappings 🔭

"JLCM allows a very efficient approximation of any DNN: as such, a Llama 7B can be compressed down to 2Go and loaded on 5-year-old smartphones." [gal30b+] 🤖 #CV

🔗 https://arxiv.org/abs/2309.17361v1 #arxiv

https://creative.ai/system/media_attachments/files/111/172/994/825/239/480/original/d36454cc24b7440e.jpg

https://creative.ai/system/media_attachments/files/111/172/994/880/648/031/original/1edb10a11e23e58d.jpg

https://creative.ai/system/media_attachments/files/111/172/994/937/209/552/original/4e1615bad11a6c87.jpg

https://creative.ai/system/media_attachments/files/111/172/994/990/796/073/original/3ae9fd35bf987914.jpg