I was trying to run this line from translation-tf.ipynb
from transformers import TFAutoModelForSeq2SeqLM, DataCollatorForSeq2Seq
model = TFAutoModelForSeq2SeqLM.from_pretrained('Helsinki-NLP/opus-mt-en-ar')
404 Client Error: Not Found for url: https://huggingface.co/Helsinki-NLP/opus-mt-en-ar/resolve/main/tf_model.h5
But it was runinng in the previous version like this:
Any help? how could I use the new instructions?
While you initialize
TFAutoModelForSeq2SeqLM you need to initialize it with
TFAutoModelForSeq2SeqLM.from_pretrained('Helsinki-NLP/opus-mt-en-ar', from_pt = True) because the TF model itself doesn’t exist but
TFAutoModelForSeq2SeqLM is implemented, meaning, you can convert PyTorch weights to TensorFlow and use it as a TF model.
that solves my problem
Note that in the upcoming version of Transformers, the error message will tell you this explicitly:
OSError: Helsinki-NLP/opus-mt-en-fr does not appear to have a file named tf_model.h5 but there is a file for PyTorch weights. Use `from_pt=True` to load this model from those weights.