Text2Text Generation
Transformers
PyTorch
Safetensors
English
t5
Inference Endpoints
text-generation-inference

Infinite loop of processing

#3
by Ejentos - opened

Hi there,
I am trying to correct human speech and faced with the issue when XL and XL-composite models go to infinite processing. Here is the text example:
"fix grammar: hello! It's me. da da da da da da da da da da."
For sure 'da da da da' doesn't make any sense, but should it be an infinite loop? Any solutions?

Grammarly org

Hello. Can you please try setting a repetition_penalty or doing beam search when you call generate.

@jbochi thanks for the reply.
Indeed it solves the issue with the infinite loop of processing, but at the same time it might rephrase sentences dramatically (or skip some).

machineteacher changed discussion status to closed

Sign up or log in to comment