Welcome to detoxify Discussions!

[laurahanu]

👋 Welcome!

We’re using Discussions as a place to connect with other members of our community. We hope that you:

• Ask questions you’re wondering about.
• Share ideas.
• Engage with other community members.
• Welcome others and are open-minded. Remember that this is a community we
  build together 💪.

To get started, comment below with an introduction of yourself and tell us about what you do with this community.

[SallyBean]

Great repo!

I have a question, I hope someone can help?

Are the toxicity scores provided by the Unitary models probability scores, in the same way that perspective API returns these values?

"The only score type currently offered is a probability score. It indicates how likely it is that a reader would perceive the comment provided in the request as containing the given attribute. For each attribute, the scores provided represent a probability, with a value between 0 and 1. A higher score indicates a greater likelihood that a reader would perceive the comment as containing the given attribute. For example, a comment like “You are an idiot” may receive a probability score of 0.8 for attribute TOXICITY, indicating that 8 out of 10 people would perceive that comment as toxic. "

Or do they represent the extent of the toxicity?

Thanks so much!

[Rohan101314]

I could not get detoxify to work on lamda and had a question.

{"errorMessage": "/var/task/nvidia/cufft/lib/libcufft.so.10: failed to map segment from shared object", "errorType": "OSError", "requestId":

The error message I get is posted above and I am not sure how to fix this. I am trying to import detoxify and deploy it via a container on aws. The library has gpu bound assets and causes this issue according to me. How do I workaround this on lambda?

I have tried changing models but, the imports stick, causing the issue. I was thinking about either using sagemaker or trying to get anaconda to work around by installing nvidea based packages.

I have been stuck for a while on this and am unsure about how I could get this to work on lambda without expanding to other technologies?