Spaces:
Running
on
CPU Upgrade
"Unknown model size" when submitting model
When trying to submit the google/mt5-large model for LLM leaderboard, it throws a "Unknown model size" error.
This error did not happen when submitting google/mt5-base or google/mt5-xl model.
Any way to resolve this?
It also happened for umt5-small
Hi! We extract model size from the safetensors weights (or the name in last resort), so you would need to provide models stored as safetensors - it's also a safer format for model storing than pickles for example.
For these older models that are not in safetensors, can there be (i) exceptions made or (ii) some way to rewrite the models artefacts to add the safetensors to the model files?
There are spaces which can port the model weights to safetensors iirc - the whole documentation is here if you want to do it for these models
Thank you @clefourrier for the pointers! I'll see how to work it out for these older models =)
I'm closing this discussion, please feel free to reopen it in case of any questions or open a new one!