Open lshamis opened 5 years ago
Hey there, this is an off-the-shelf model that usually spits out gibberish. If you want custom text, I've realised that fine-tuning it on your own corpora may help tremendously. Have you given that a try?
I'm seeing the same issue it just outputs garbage whereas many other examples online are quite good like https://transformer.huggingface.co/doc/gpt2-large why is this implementation outputting garbage?
I am also getting this issue. It seems like it could be an issue where the data isn't actually being passed to the model so it just starts generating unrelated text.
In interactive mode, it seems to completely ignore the prompt and just create random text with a lot of repetition. Tried changing the temperature to 0.5, no improvement. I agree that the prompt appears to not be taken into account, at all.
Describe the bug Ran the README example with the following prompts:
['forklifts', 'tell me about forklifts', 'what is the current state of the art in forklifts']
None of generated entries contained the work 'forklift', or were in anyway related.To Reproduce Steps to reproduce the behavior:
Expected behavior
Screenshots Not applicable
Desktop (please complete the following information): OS: Ubuntu (doesn't matter)
Additional context