HuggingFaceFW/fineweb
Viewer • Updated • 52.5B • 384k • 3.37k
Incredibly small 3 expert Mixture of Experts trained to be decent in coding, math, and generic conversation. I'm still training it :'D
I suggest using BeeLlama.cpp for squeezing as much performance as possible out of this thing :]
This model will include an mmproj to try to have basic image support using the frozen embeddings of wkcn/TinyCLIP-ViT-8M-16-Text-3M-YFCC15M.