Merge remote-tracking branch 'origin/fix_ci' into add_h100_ci

Author: jainapurva

test/integration/test_integration.py

@@ -1821,7 +1821,7 @@ def test_autoquant_int4wo(self, device, dtype):
             self.assertGreater(compute_error(ref, out), 20)
 
     @parameterized.expand(COMMON_DEVICE_DTYPE)
-    @unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")
+    @unittest.skipIf(not is_sm_at_least_90(), "Need cuda arch greater than SM90")
     @unittest.skipIf(
         not TORCH_VERSION_AT_LEAST_2_5, "autoquant int4 option requires 2.5+."
     )

torchao/quantization/qat/utils.py

@@ -16,14 +16,6 @@
     _get_per_token_block_size,
 )
 
-# Attribute name representing the forward prehook wrapping the
-# linear input in an `AffineFakeQuantizedTensor` on a linear module.
-#
-# The value of this attribute is a 2-tuple of (prehook, handle).
-# The prehook can be disabled by calling `handle.remove()`, and
-# re-enabled by calling `module.register_forward_pre_hook(prehook)`.
-_QAT_LINEAR_SUBCLASS_INPUT_PREHOOK = "_qat_linear_subclass_input_prehook"
-
 
 class _GenericFakeQuantize(torch.autograd.Function):
     """