Open G200005 opened 1 week ago
Since the GQA dataset does not have "attributes," all related parts need to be removed. Therefore, I modified target.copy_with_fields(["labels"]) #, "attributes"]) to control attributes, and I have updated the relevant sections accordingly.
It seems that the changes you made are correct when compared with the SHA_GCL code.
Feel free to ask about even minor errors, and I will check and address them. Thank you.
Hello, when I tried to train the GQA dataset I found the following error, suggesting that Field ‘attributes’ not found, according to the error report, I modified the code in sampling.py and inference.py, I wonder if this is correct?