Skip to content

Model Inputs and Outputs

model_navigator.triton.InputTensorSpec dataclass 

InputTensorSpec(name, shape, dtype=None, reshape=lambda: ()(), is_shape_tensor=False, optional=False, format=None, allow_ragged_batch=False)

Bases: BaseTensorSpec

Stores specification of single input tensor.

This includes name, shape, dtype and more parameters available for input tensor in Triton Inference Server:

Read more in Triton Inference server model configuration

Parameters:

  • optional (bool, default: False ) –

    Flag marking the input is optional for the model execution

  • format (Optional[InputTensorFormat], default: None ) –

    The format of the input.

  • allow_ragged_batch (bool, default: False ) –

    Flag marking the input is allowed to be "ragged" in a dynamically created batch.

__post_init__ 

__post_init__()

Validate the configuration for early error handling.

Source code in model_navigator/triton/specialized_configs/common.py 
47
48
49
50
51
52
53
54
55
56
57
58
def __post_init__(self) -> None:
    """Validate the configuration for early error handling."""
    if self.dtype:
        self.dtype = cast_dtype(dtype=self.dtype)

    expect_type("name", self.name, str)
    expect_type("shape", self.shape, tuple)
    expect_type("reshape", self.shape, tuple, optional=True)
    expect_type("dtype", self.dtype, np.dtype, optional=True)
    expect_type("is_shape_tensor", self.is_shape_tensor, bool, optional=True)
    is_shape_correct("shape", self.shape)
    is_shape_correct("reshape", self.reshape, optional=True)

model_navigator.triton.InputTensorFormat

Bases: Enum

Format for input tensor.

Read more in Triton Inference server model configuration

Parameters:

  • FORMAT_NONE –

    0

  • FORMAT_NHWC –

    1

  • FORMAT_NCHW –

    2

model_navigator.triton.OutputTensorSpec dataclass 

OutputTensorSpec(name, shape, dtype=None, reshape=lambda: ()(), is_shape_tensor=False, label_filename=None)

Bases: BaseTensorSpec

Stores specification of single output tensor.

This includes name, shape, dtype and more parameters available for output tensor in Triton Inference Server:

Read more in Triton Inference server model configuration

Parameters:

  • label_filename (Optional[str], default: None ) –

    The label file associated with this output.

__post_init__ 

__post_init__()

Validate the configuration for early error handling.

Source code in model_navigator/triton/specialized_configs/common.py 
47
48
49
50
51
52
53
54
55
56
57
58
def __post_init__(self) -> None:
    """Validate the configuration for early error handling."""
    if self.dtype:
        self.dtype = cast_dtype(dtype=self.dtype)

    expect_type("name", self.name, str)
    expect_type("shape", self.shape, tuple)
    expect_type("reshape", self.shape, tuple, optional=True)
    expect_type("dtype", self.dtype, np.dtype, optional=True)
    expect_type("is_shape_tensor", self.is_shape_tensor, bool, optional=True)
    is_shape_correct("shape", self.shape)
    is_shape_correct("reshape", self.reshape, optional=True)