Class TensorRTConfig

python/paddle/tensorrt/export.py:231–342

class TensorRTConfig:
    def __init__(
        self,
        inputs: list,
        min_subgraph_size: int | None = 3,
        save_model_dir: str | None = None,
        disable_ops: str | list | None = None,
        precision_mode: PrecisionMode = PrecisionMode.FP32,
        ops_run_float: str | list | None = None,
        optimization_level: int | None = 3,
        disable_passes: list = [],
        workspace_size: int | None = 1 << 30,
        use_cuda_graph: bool | None = False,
        refit_params_path: str | None = None,
        disable_logging: bool | None = True,
    ) -> None:
        """
        A class for configuring TensorRT optimizations.

        Args:
            inputs (list):
                A list of Input configurations
            min_subgraph_size (int, optional):
                The minimum number of operations in a subgraph for TensorRT to optimize (default is 3).
            save_model_dir (str, optional):
                The directory where the optimized model will be saved (default is not to save).
            disable_ops : (str|list, optional):
                A string representing the names of operations that should not be entering by TensorRT (default is None).
            precision_mode (PrecisionMode, optional):
                Specifies the precision mode for TensorRT optimization. The options are:
                - PrecisionMode.FP32: 32-bit floating point precision (default).
                - PrecisionMode.FP16: 16-bit floating point precision.
                - PrecisionMode.INT8: 8-bit integer precision.
                - PrecisionMode.BF16: 16-bit Brain Floating Point precision. Only supported in TensorRT versions greater than 9.0.
            ops_run_float (str|list, optional):
                A set of operation names that should be executed using FP32 precision regardless of the `tensorrt_precision_mode` setting.
            optimization_level (int, optional):
                Set TensorRT optimization level (default is 3). Only supported in TensorRT versions greater than 8.6.
            disable_passes : (str|list, optional):
                A list of string representing the names of pass that should not be used for origin program (default is []).
            workspace_size (int, optional):
                Specifies the maximum GPU memory (in bytes) that TensorRT can use for the optimization process (default is 1 << 30).
            use_cuda_graph (bool, optional):
                Specify whether TensorRT enables cuda_graph during the optimization process (default is false).
            refit_params_path(str, optional):
                The path to the weights that need to be refitted.
            disable_logging (bool, optional):
                Specifies whether to enable GLOG info output during the optimization process (default is true).
        Returns:
            None

        Examples:
            .. code-block:: pycon

                >>> # example 1:
                >>> from paddle.tensorrt.export import (
                >>>    Input,
                >>>    TensorRTConfig,