afe.backends.mpk.node

Functions for making MPK JSON node descriptions.

Functions

`flatten_tuple`(→ List[_A])	Convert un-nested tuples to a list.
`get_node_size`(→ int)	Get node size in bytes.
`get_output_node_names`(→ afe.ir.defines.DataValue[str])	Get the output node names.
`get_plugin_input_nodes`(...)	Convert a list of MPK JSON nodes to a list of inputs for MLA or EV74 plugin.

Module Contents

afe.backends.mpk.node.flatten_tuple(data: afe.ir.defines.DataValue[_A]) → List[_A][source]

Convert un-nested tuples to a list. Given a tuple of 0, 2, or more tensors, the tuple elements are returned. Given a tensor, the tensor is returned as a list of 1 item. An exception is raised for other inputs.

Parameters:: data – Data value
Returns:: Flattened data value

afe.backends.mpk.node.get_node_size(node: afe.ir.node.AwesomeNode) → int[source]

Get node size in bytes.

Parameters:: node – AwesomeNode
Returns:: Number of bytes.

afe.backends.mpk.node.get_output_node_names(node_name: str, output_type: afe.ir.defines.DataValue[afe.ir.tensor_type.TensorType], output_names: Dict[str, str]) → afe.ir.defines.DataValue[str][source]: Get the output node names. Check if output node is an output of the model and if it is original model name is added to the ModelSDK name.

afe.backends.mpk.node.get_plugin_input_nodes(input_nodes: List[afe.backends.mpk.defines.InOutNodesMPKData]) → List[afe.backends.mpk.defines.PluginInputNodeMPKData][source]

Convert a list of MPK JSON nodes to a list of inputs for MLA or EV74 plugin.

Parameters:: input_node – MPK JSON objects representing inputs to the AwesomeNode.
Returns:: PluginInputNodeMPKData class with input node.