error when value is a np.array int

[ecilay]

TensorRT/py/torch_tensorrt/dynamo/conversion/converter_utils.py

Line 273 in 5f66ade

(1,) if isinstance(value, (int, float, bool)) else value.shape,

In this line, if the value is a np.array int, eg, a = np.array(50), then a.shape would be empty, and the ctx.net.add_constant function would throw Type error: incompatible function arguments.

[gs-olive]

Hi - thanks for the report. Are you encountering this bug when compiling a specific model?

Fix Outline

Ensure all np.array tensors have nonzero rank before calling add_constant here:

TensorRT/py/torch_tensorrt/dynamo/conversion/converter_utils.py

Lines 272 to 275 in 5f66ade

	constant = ctx.net.add_constant(
	(1,) if isinstance(value, (int, float, bool)) else value.shape,
	numpy_value.copy() if isinstance(numpy_value, np.ndarray) else numpy_value,
	)

A regression test case to elicit this behavior might be to have a constant variable which is a dimensionless tensor, as in:

class TestModule(nn.Module):
  def __init__(...):
    self.const = torch.tensor(1)
...