talk to booper
A 34-million-parameter transformer that was pretrained on web text and then fine-tuned on Discord threads. It is very small and very confident. Say hi.
8 layers512 wide16k vocab
1024 token windowCPU inference
A 34-million-parameter transformer that was pretrained on web text and then fine-tuned on Discord threads. It is very small and very confident. Say hi.
Defaults are the ones booper actually ships with.