Minor issues

#1
by lunahr - opened

There are some issues with this repo and the data.

  • There is no metadata. Try to add some basic metadata to enable Hugging Face metrics.
  • The model is a hybrid FP32 / BF16 model, this may cause problems. If you didn't intend this to happen, try asking Claude and ChatGPT o1 to quantize all the tensors to BF16 to avoid issues.
  • You don't have a tokenizer. You need to add a tokenizer of a compatible type, Gemma 2's in this case.

There are some issues with this repo and the data.

  • There is no metadata. Try to add some basic metadata to enable Hugging Face metrics.
  • The model is a hybrid FP32 / BF16 model, this may cause problems. If you didn't intend this to happen, try asking Claude and ChatGPT o1 to quantize all the tensors to BF16 to avoid issues.

Gotcha, I’ll get it done asap!

Downloads aren't tracked, likely because you don't have a tokenizer so the model cannot be recognized as a valid Transformer.

Try adding a tokenizer of the correct model architecture you intended.

I think I’ve heard they’re tracked via the config. I’ll make one once I get a chance.

I think I’ve heard they’re tracked via the config. I’ll make one once I get a chance.

Still, no tokenizer means the model can't be used.

Let me go do that then. Edit: I have uploaded the config and tokenizer from the previous repo.

It's looking correct to me now. I'm eager to see the known issue get fixed.

Maybe you would also want to set the

library_name: transformers

in the metadata as well

Maybe you would also want to set the

library_name: transformers

in the metadata as well

Forgot to mention but I fixed this!

I’m gonna mark this as closed as there doesn’t seem like there’s anything else to be done for now. Feel free to make another conversation if there’s something else!

QuietImpostor changed discussion status to closed

Sign up or log in to comment