Add pyre fixme for downstream type errors from D65753120

Summary:
X-link: meta-pytorch/captum#1439

X-link: facebook/Ax#3055

X-link: ctrl-labs/src2#38515

X-link: ctrl-labs/src2#38514

bypass-github-export-checks "The check is bugged, I exported all the required exports"

Reviewed By: jermenkoo

Differential Revision: D65826205

fbshipit-source-id: b08f00caaac8f3cc235bc280c4fe3f99089a0753

Author: ezyang

torcheval/metrics/functional/classification/confusion_matrix.py

@@ -270,13 +270,17 @@ def _confusion_matrix_update_input_check(
                 f"got {input.shape}."
             )
     else:
+        # pyre-fixme[58]: `>=` is not supported for operand types `Tensor` and
+        #  `Optional[int]`.
         if torch.max(input) >= num_classes:
             raise ValueError(
                 "Got `input` prediction class which is too large for the number of classes, "
                 f"num_classes: {num_classes} must be strictly greater than max class predicted: {torch.max(input)}."
             )
 
     # check if num classes is high enough to cover targets.
+    # pyre-fixme[58]: `>=` is not supported for operand types `Tensor` and
+    #  `Optional[int]`.
     if torch.max(target) >= num_classes:
         raise ValueError(
             "Got `target` class which is larger than the number of classes, "