Newer HF transformers version saves model weights as model.safetensors instead of pytorch_model.bin

[guo0O0o]

Following sequence tagging Chemprot example gives errors on newer versions of HF transformers (currently on 4.41.2). Apparently model.save_pretrained('folder/') now saves model as .safetensors rather than .bin

[tmills]

[tmills]

hmmm... I went to replicate this now and I can't get it to fail.

[tmills]

From @guo0O0o : i think you can also just add the flag --save_safetensors False at the end of your script