[inductor][mkl][inline-inbuilt-nn-modules] Change assertion (pytorch#130219)

Fixes the test in the next PR - `python test/inductor/test_mkldnn_pattern_matcher.py -k TestDynamicPatternMatcher.test_conv3d_unary_dynamic_shapes`

Pull Request resolved: pytorch#130219
Approved by: https://github.com/leslie-fang-intel

Author: anijain2305

torch/_inductor/mkldnn_ir.py

@@ -113,7 +113,7 @@ def _original_deconv_weight_size(
         else:
             assert 0 < len(output_padding) <= dims
             output_padding = pad_listlike(output_padding, dims)
-        assert isinstance(groups, int)
+        assert isinstance(groups, (int, sympy.core.numbers.Integer))
         if transposed:
             # When transposed, the size of the prepacked oneDNN weight is different
             # from the PyTorch weight. We're not able to run aten conv with such