Gemma 3 270M: Compact model for hyper-efficient AI 11 months ago
Thank you a lot for working on these models! If you think it would make sense, I think a bigger sized Gemma model would be really cool. Models in the 70B parameter range can be run at q4 on two 3090 or similar hardware and should offer considerable performance improvement over 27B. There’s also the DGX Spark as a possible target.