onnxruntime API reference

66 public APIs from onnxruntime (microsoft/onnxruntime) — 36 classes, 10 functions, 20 methods. Signatures extracted by static analysis of the actual source.

Repository: microsoft/onnxruntime

Kind	Count
Classes	36
Functions	10
Methods	20

API list

funconnxruntime.print_debug_info()

Print information to help debugging.

classonnxruntime.python.onnxruntime_inference_collection.InferenceSession

This is the main class used to run a model.

classonnxruntime.python.onnxruntime_inference_collection.ModelCompiler

This class is used to compile an ONNX model.

methodonnxruntime.python.onnxruntime_inference_collection.ModelCompiler.compile_to_file(output_model_path:str | None=None)

Compiles to an output file.

classonnxruntime.python.onnxruntime_inference_collection.Session

This is the main class used to run a model.

methodonnxruntime.python.onnxruntime_inference_collection.Session.disable_fallback() -> None

Disable session.run() fallback mechanism.

methodonnxruntime.python.onnxruntime_inference_collection.Session.enable_fallback() -> None

Enable session.Run() fallback mechanism.

methodonnxruntime.python.onnxruntime_inference_collection.Session.get_modelmeta() -> C.ModelMetadata

Return the metadata.

methodonnxruntime.python.onnxruntime_inference_collection.Session.get_session_options() -> C.SessionOptions

Return the session options.

methodonnxruntime.python.onnxruntime_inference_collection.Session.io_binding() -> IOBinding

Return an onnxruntime.IOBinding object`.

method

onnxruntime.python.onnxruntime_inference_collection.Session.run(output_names, input_feed, run_options=None) -> Sequence[np.ndarray | SparseTensor | list | dict]

Compute the predictions.

method

onnxruntime.python.onnxruntime_inference_collection.Session.run_with_ort_values(output_names, input_dict_ort_values, run_options=None) -> Sequence[OrtValue]

Compute the predictions.

methodorttraining.orttraining.python.training.api.checkpoint_state.Parameter.data() -> np.ndarray

The data of the parameter

methodorttraining.orttraining.python.training.api.checkpoint_state.Parameter.grad() -> np.ndarray

The gradient of the parameter

methodorttraining.orttraining.python.training.api.checkpoint_state.Parameter.name() -> str

The name of the parameter

methodorttraining.orttraining.python.training.api.module.Module.eval() -> Module

Sets the Module in evaluation mode.

method

orttraining.orttraining.python.training.api.module.Module.export_model_for_inferencing(inference_model_uri:str | os.PathLike, graph_output_names:list[str]) -> None

Exports the model for inferencing.

methodorttraining.orttraining.python.training.api.module.Module.get_parameters_size(trainable_only:bool=True) -> int

Returns the size of the parameters.

methodorttraining.orttraining.python.training.api.module.Module.train(mode:bool=True) -> Module

Sets the Module in training mode.

methodorttraining.orttraining.python.training.api.optimizer.Optimizer.set_learning_rate(learning_rate:float) -> None

Sets the learning rate for the optimizer.

classorttraining.orttraining.python.training.artifacts.LossType

Loss type to be added to the training model.

classorttraining.orttraining.python.training.onnxblock.blocks.Abs

Adds Abs node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Add

Adds Add node to an onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Clip

Adds Clip node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Div

Adds Div node to an onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Log

Adds Log node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Mul

Adds Mul node to an onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Neg

Adds Neg node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Pow

Adds Pow node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.ReduceAllL2

Adds ReduceAllL2 node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.ReduceMean

Adds ReduceMean node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.ReduceSum

Adds ReduceSum node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Sigmoid

Adds Sigmoid node to the onnx model.

classorttraining.orttraining.python.training.onnxblock.blocks.Sub

Adds Sub node to an onnx model.

classorttraining.orttraining.python.training.optim.fused_adam.FusedAdam

Implements Adam algorithm.

classorttraining.orttraining.python.training.ort_triton._codegen.TritonCodegen

Specialized codegen for Triton backend.

classorttraining.orttraining.python.training.ort_triton._common.CodegenContext

record variable name mapping in term of IRnodes.

classorttraining.orttraining.python.training.ort_triton._common.TensorInfo

Represent a input/output tensor of a node.

classorttraining.orttraining.python.training.ort_triton._ir.ComputeNode

Each operator is represented as a ComputeNode.

classorttraining.orttraining.python.training.ort_triton._ir.IRNode

The base class for all IR nodes.

classorttraining.orttraining.python.training.ort_triton._ir.ReduceForLoopEnd

shared-memory reduction

classorttraining.orttraining.python.training.ort_triton._sorted_graph.SortedGraph

This class is used to 1.

funcorttraining.orttraining.python.training.ort_triton._utils.next_power_of_2(n:int) -> int

Return the smallest power of 2 greater than or equal to n

classorttraining.orttraining.python.training.ortmodule._execution_agent.ExecutionAgentOutput

Wraps an OrtValue and adds an ID.

classorttraining.orttraining.python.training.ortmodule._graph_transition_manager.GraphTransitionManager

Manage the graph transition from 1).

func

orttraining.orttraining.python.training.ortmodule._io.calculate_total_parameter_size_in_bytes(module:torch.nn.Module) -> int

Calculate the total parameter size in bytes

classorttraining.orttraining.python.training.ortmodule._onnx_models.ONNXModels

Encapsulates all ORTModule onnx models.

methodorttraining.orttraining.python.training.ortmodule._runtime_inspector.MemoryObserver.is_enabled() -> bool

Check if memory inspector is enabled.

classorttraining.orttraining.python.training.ortmodule._runtime_inspector.RuntimeInspector

Runtime inspector for ORTModule.

funcorttraining.orttraining.python.training.ortmodule._utils.get_rank() -> int

Returns the rank of the current process.

funcorttraining.orttraining.python.training.ortmodule._utils.get_world_size() -> int

Returns the world size of the current process.

classorttraining.orttraining.python.training.ortmodule.experimental.pipe._ort_pipeline_module.ORTPipelineModule

ORTPipelineModule pipeline module.

classorttraining.orttraining.python.training.ortmodule.graph_optimizers.utils.GraphMatcher

Sub-graph matcher with given pattern.

classorttraining.orttraining.python.training.ortmodule.options.DebugOptions

Configurable debugging options for ORTModule.

classorttraining.orttraining.python.training.utils.ptable.PTable

A table that can be printed to the console.

methodorttraining.orttraining.python.training.utils.ptable.PTable.add_row(columns:list[str]) -> Row

Add a row to the table.

method

orttraining.orttraining.python.training.utils.ptable.PTable.get_string(first_column_width=None, second_column_width=None) -> str

Serialize the table to a string.

classorttraining.orttraining.python.training.utils.ptable.Row

A row in a PTable

classorttraining.orttraining.python.training.utils.torch_io_helper.PrimitiveType

Helper class for Python primitive types.

methodorttraining.orttraining.python.training.utils.torch_io_helper.PrimitiveType.get_tensor(value, device) -> torch.Tensor

Convert `value` to a torch.Tensor.

classplugin-ep-cuda.csharp.pack_nuget.PackError

Raised for any user-actionable failure during packaging.

funcplugin-ep-cuda.csharp.pack_nuget.stage_sources(staging_dir:Path) -> None

Copy project sources into staging, excluding bin/obj.

funcplugin-ep-cuda.python.onnxruntime_ep_cuda.get_ep_name() -> str

Return the CUDA plugin Execution Provider name.

funcplugin-ep-webgpu.csharp.pack_nuget.stage_sources(staging_dir:Path) -> None

Copy project sources into staging, excluding bin/obj.

funcplugin-ep-webgpu.python.build_wheel.build_wheel(source_dir:Path, wheel_dir:Path)

Build the wheel using pip.

funcplugin-ep-webgpu.python.onnxruntime_ep_webgpu.get_ep_name() -> str

Return the WebGPU Execution Provider name.

About this data

These signatures were extracted from the public source of microsoft/onnxruntime using Python's ast module. Argument names, default values, type annotations and return types are taken verbatim from the code. Implementation bodies are never stored. See how it works for details.

Back to all 805 libraries