Helsinki-NLP

/

opus-mt-sv-en

like 14

Follow

Helsinki-NLP Research Group 1.08k

The model will output:
["I'm sorry. I'm sorry. I'm sorry."]

when prompted on an empty string:

Code to reproduce the error:

'''from transformers import MarianTokenizer, AutoModelForSeq2SeqLM
name = "Helsinki-NLP/opus-mt-sv-en"
model = AutoModelForSeq2SeqLM.from_pretrained(name)
tokenizer = MarianTokenizer.from_pretrained(name)
inputs = tokenizer(" ", padding=True, truncation=True, return_tensors="pt")
translated_outputs = model.generate(**inputs)
translations = [tokenizer.decode(output, skip_special_tokens=True) for output in translated_outputs]'''

Anyone having the same issue and possibly a solution to fix it?

Thanks!