Wrong automatic detection of nb of parameters & precision (dataset card bug)
#4
by
Jeronymous
- opened
This is wrong : that model has 6.7 B parameter, with a precision of 4 bits.
I guess there is a wrong assumption that the precision is float 16, and then the estimation of the number of parameters goes wrong (looking at the file size ?).
I wonder whether it can be fixed by adding something in the yaml header :thinking:
Good point! We will ask Hugging Face. It is atm not clear from the model card doc.