[BE]: Enable ruff rule TRY302 and apply fixes (#101874)
Removes useless try statements and unreachable code.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/101874
Approved by: https://github.com/malfet
diff --git a/test/test_mps.py b/test/test_mps.py
index d6e0564..1629da3 100644
--- a/test/test_mps.py
+++ b/test/test_mps.py
@@ -10487,52 +10487,47 @@
# Forward check
#
forward_failed = False
- try:
- mps_sample = cpu_sample.transform(
- lambda x: x.detach().to("mps").requires_grad_(x.requires_grad) if isinstance(x, torch.Tensor) else x)
+ mps_sample = cpu_sample.transform(
+ lambda x: x.detach().to("mps").requires_grad_(x.requires_grad) if isinstance(x, torch.Tensor) else x)
- cpu_args = [cpu_sample.input] + list(cpu_sample.args)
- cpu_kwargs = cpu_sample.kwargs
- mps_args = [mps_sample.input] + list(mps_sample.args)
- mps_kwargs = mps_sample.kwargs
+ cpu_args = [cpu_sample.input] + list(cpu_sample.args)
+ cpu_kwargs = cpu_sample.kwargs
+ mps_args = [mps_sample.input] + list(mps_sample.args)
+ mps_kwargs = mps_sample.kwargs
- # for tensor_split(), the second tensor arg ("tensor_indices_or_sections") must be on CPU only
- if (op.name == "tensor_split" and isinstance(mps_args[1], torch.Tensor)):
- mps_args[1] = cpu_args[1]
+ # for tensor_split(), the second tensor arg ("tensor_indices_or_sections") must be on CPU only
+ if (op.name == "tensor_split" and isinstance(mps_args[1], torch.Tensor)):
+ mps_args[1] = cpu_args[1]
- cpu_out = op(*cpu_args, **cpu_kwargs)
- mps_out = op(*mps_args, **mps_kwargs)
+ cpu_out = op(*cpu_args, **cpu_kwargs)
+ mps_out = op(*mps_args, **mps_kwargs)
- if (op.name in self.FP32_LOW_PRECISION_LIST) and dtype == torch.float32:
- atol = 1e-4
- rtol = 3e-5
- elif op.name == "nn.functional.conv2d" or op.name == "linalg.multi_dot" and dtype == torch.float32:
- atol = 1e-4
- rtol = 3e-5
- elif (op.name in self.FP16_LOW_PRECISION_LIST) and dtype == torch.float16:
- atol = 1e-2
- rtol = 1e-2
- elif (op.name == "masked.mean"):
- atol = 7e-4
- rtol = 2e-3
- elif (op.name == "native_layer_norm"):
- atol = 1e-4
- rtol = 1.3e-5
- elif (op.name == "norm" or op.name == "linalg.norm") and dtype == torch.float16:
- atol = 7e-4
- rtol = 1.5e-3
- elif op.name == "unique" and cpu_kwargs["sorted"] is False:
- continue
- else:
- atol = None
- rtol = None
+ if (op.name in self.FP32_LOW_PRECISION_LIST) and dtype == torch.float32:
+ atol = 1e-4
+ rtol = 3e-5
+ elif op.name == "nn.functional.conv2d" or op.name == "linalg.multi_dot" and dtype == torch.float32:
+ atol = 1e-4
+ rtol = 3e-5
+ elif (op.name in self.FP16_LOW_PRECISION_LIST) and dtype == torch.float16:
+ atol = 1e-2
+ rtol = 1e-2
+ elif (op.name == "masked.mean"):
+ atol = 7e-4
+ rtol = 2e-3
+ elif (op.name == "native_layer_norm"):
+ atol = 1e-4
+ rtol = 1.3e-5
+ elif (op.name == "norm" or op.name == "linalg.norm") and dtype == torch.float16:
+ atol = 7e-4
+ rtol = 1.5e-3
+ elif op.name == "unique" and cpu_kwargs["sorted"] is False:
+ continue
+ else:
+ atol = None
+ rtol = None
- self.assertEqual(cpu_out, mps_out, atol=atol, rtol=rtol)
+ self.assertEqual(cpu_out, mps_out, atol=atol, rtol=rtol)
- except Exception as e:
- raise e
- forward_failed = True
- all_forward_pass = False
#
# Backward check
