Core Python APIs - 3.5 English

Vitis AI User Guide (UG1414)

Document ID
Release Date
3.5 English
Table 1. get_target_info()
Description Parameters Return
Gets target information including batch, fingerprint, target name. You can use info for batching or get target name information. None A DeviceInfo object. See core classes section for more details about this object type.
Table 2. create_wego_graph(input_graph_def, feed_dict={}, accuracy_mode= vitis_vai.enums.AccuracyMode.Default)
Description Parameters Return
Python wrapper for the VAI transformation.
  1. input_graph_def: GraphDef object containing a model to be transformed.
  2. feed_dict: Infer shape configuration when input model without fixed input shape.
  3. accuracy_mode:
    • vitis_vai.enums.AccuracyMode.Default: Running without CPU FixNeuron.
    • vitis_vai.enums.AccuracyMode.ReserveReduantFixNeurons: Running with CPU FixNeruon
New GraphDef with VaiWeGOOps placed in graph replacing subgraphs.
Note: WeGO eliminates CPU FixNeurons operators within quantized model to achieve optimal performance by default. However for those models containing many CPU FixNeurons operators, the models’ accuracy maybe decrease by deploying them with default value(Vitis_vai.enums.AccuracyMode.Default).In such cases, you can switch to Vitis_vai.enums.AccuracyMode.ReserveReduantFixNeurons to achieve better accuracy.