16 or 32 bit?

#19
by ChrisGoringe - opened

The model card says "The native weights of this model were exported in bfloat16 precision"

But the config.json file says "torch_dtype": "float32" - and the fact that a 9B parameter model is 36GB to download suggests the same.

If it's really only bfloat16, could we have an 18GB download?

They updated the model

Sign up or log in to comment