Kurt Shuster
Kurt Shuster
from a cursory examination, it looks like this might be failing because we can't fit the model activations in the first gpu spot. i'll need to investigate a bit more
The majority of our recent models include the [Empathetic Dialogues](https://github.com/facebookresearch/EmpatheticDialogues) dataset in their training data, so in theory should be able to display empathy when conversing with a chat partner.
on my personal env, I have `urllib3==1.26.9`, so maybe just try installing with that?
what version of parlai are you using? could you try upgrading?
could you possibly try reinstalling? Additionally, not sure how it is relevant to this issue, but note that ParlAI currently supports python 3.8+
could you try installing from source?
can you show me the full install procedure + stack trace? this _should_ work if installed from source
at the moment, I don't believe there are any pure language-modeling tasks within parlai. we'd be open to accepting additional datasets
These are built-in safety measures for which the bot identifies that its response may be unsafe; please see the [Safety section](https://parl.ai/projects/blenderbot2/) of the project page for more information
The model was only trained to generate `POTENTIALLY_UNSAFE`