Spaces:
Running
on
T4
question about applying road sign pictures
I really love this online demo you offer so that I can test the model conveniently. But when I tried to apply a road sign here, it gave me a wrong reply.
It's a "turn right" road sign, but the model gives me "left arrow":
However, when I just inverse the image simply, I gives me a correct "left arrow".
So, why does the model give the same result for two converse images?
But it performs right in a multiple road sign image:
Is this normal for the llm? Thank u, I'm new to try this advanced model.
Hi
@ymlzovo
, I am not able to say why the model gives the wrong answer 😅, but this happens (otherwise we get 100% correct LLM which is not be the case - not at this moment!).
You can definitely try to get the logits from the model at the corresponding/relevant tokens and do some analysis.