Using XLA fast text generation with Pegasus models

Intermediate
2

Hi,

I see you’re using AutoModelForSeq2SeqLM but you need to use TFAutoModelForSeq2SeqLM. I see that the repo you’re trying to load only has a PyTorch checkpoint, so the proper way to load it would be:

from transformers import TFAutoModelForSeq2SeqLM

model = TFAutoModelForSeq2SeqLM.from_pretrained("tuner007/pegasus_paraphrase", from_pt=True)