ONNXModelParser Class Reference

Parses the ONNX model into a workload. More...

Public Member Functions
None	__init__ (self, str|ModelProto onnx_model, str mapping_yaml_path)
ONNXWorkload	run (self)
	Iterate through the onnx model and generate the workload consisting of LayerNodes and DummyNodes. More...
def	get_parser_class (self, NodeProto node)
def	parse_workload_from_onnx_model_and_mapping (self)
	Converts an onnx model into a workload object. More...

Public Attributes
	onnx_model
	workload
	mapping_yaml_path
	mapping_data

Detailed Description

Parses the ONNX model into a workload.

Constructor & Destructor Documentation

__init__()

None __init__	(		self,
		str | ModelProto	onnx_model,
		str	mapping_yaml_path
	)

Here is the call graph for this function:

Member Function Documentation

get_parser_class()

def get_parser_class	(		self,
		NodeProto	node
	)

Here is the caller graph for this function:

parse_workload_from_onnx_model_and_mapping()

def parse_workload_from_onnx_model_and_mapping

(

self

)

Converts an onnx model into a workload object.

We scan the model for all convolutional layers, and setup a Layer object for each of those using the mapping. Then we combine the layers into a workload graph.

If the model isn't in the format with external data, it will be slow to manipulate it, so better to work with raw models with external data. The line below accomplishes this. onnx.save_model(model, 'model_external.onnx', save_as_external_data=True, all_tensors_to_one_file=True, location='model_external_raw_data', size_threshold=1024, convert_attribute=False)

In the future, assume we will have a model saved with external data, then we have to execute the code below if the model isn't inferred yet

This approach is faster for large models because the raw model is used (w/o the external data) if model is not inferred: onnx.shape_inference.infer_shapes_path('path/to/the/model.onnx') # This will save the inferred model to the same file model = onnx.load('path/to/the/model.onnx') # reload the inferred model

Saves for each node_id the inputs and outputs tensor names

Here is the call graph for this function:

run()

ONNXWorkload run

(

self

)

Iterate through the onnx model and generate the workload consisting of LayerNodes and DummyNodes.

Here is the call graph for this function:

Here is the caller graph for this function:

Member Data Documentation

mapping_data

mapping_data

mapping_yaml_path

mapping_yaml_path

onnx_model

onnx_model

workload

workload

---

The documentation for this class was generated from the following file:

• /home/runner/work/zigzag/zigzag/zigzag/parser/onnx/onnx_model_parser.py