Possible bug in model?
#1
by
dk-crazydiv
- opened
Hi, I tried the inference API with a lot of basic inputs which were abusive/angry but the model output is always "joy". Is there perhaps a bug somewhere in inference code? With this data set "I hate you" shouldn't be getting classified as "joy" in my opinion.
Yes, if you enter the example, we can see that it's mapped wrong.
optimism means joy
anger -> optimism
joy -> anger
sadness -> sadness
is no one going to fix the id2label?
it does seem like the assessment above is correct
e.g.
"I am furious": joy
"feeling good about the future": anger
"I am elated": optimism
"feeling so down right now": sadness