[PyTorch] Fix bug when deducing dtype in linear functional API (#2017)

Fix bug when deducing dtype in linear functional API
Signed-off-by: Tim Moon <tmoon@nvidia.com>

transformer_engine/pytorch/ops/basic/basic_linear.py

@@ -426,7 +426,7 @@ class BasicLinear(BasicOperation):
         if dtype is None:
             if out is not None and isinstance(out, torch.Tensor):
                 dtype = out.dtype
-            elif weight is not None and isinstance(out, torch.Tensor):
+            elif weight is not None and isinstance(weight, torch.Tensor):
                 dtype = weight.dtype
             else:
                 raise ValueError(