Model details

#3
by deathcrush - opened

Hi @philschmid ,

I experimented with this https://huggingface.co/spaces/osanseviero/i-like-flan/tree/main demo and found that I could use FLAN-T5-XXL for very interesting research. I am looking to understand a bit more what model I was running and how it relates to the google checkpoint from the hub. I can see you used Artifact-AI/flan-t5-xxl-sharded-fp16. What exactly is this model? Did the authors just cast the model weights to fp16 and then saved the model to reduce inference latency and cost?

Sign up or log in to comment