vai_p_tensorflow Usage - 1.4.1 English

Vitis AI Optimizer User Guide (UG1333)

Document ID
UG1333
Release Date
2021-10-29
Version
1.4.1 English

The following arguments are available when running vai_p_tensorflow:

Table 1. vai_p_tensorflow Arguments
Argument Type Action Default Description
action string - "" Which action to run. Valid actions include ‘ana', 'prune', 'transform', and 'flops'.
workspace string [‘ana’, ‘prune’] "" Directory for saving output files.
input_graph string [‘ana’, ‘prune’, ‘flops’] "" Path of a GraphDef protobuf file that defines the network’s architecture.
input_ckpt string [‘ana’, ‘prune’, ‘transform’] "" Path of a checkpoint file. It is the prefix of filenames created for the checkpoint.
eval_fn_path string [‘ana’] "" A Python file path used for model evaluation.
target string [‘ana’] "" The output node name that indicates the performance of the model.
max_num_batches int [‘ana’] None Maximum number of batches to evaluate.By default, use all.
output_graph string [‘prune’] "" Path of a GraphDef protobuf file for saving the pruned network.
output_ckpt string [‘prune’, ‘transform’’] "" Path of a checkpoint file for saving weights.
gpu string [‘ana’] "" GPU device IDs to use separated by ‘,’.
sparsity float [‘prune’] None The desired sparsity of network after pruning.
exclude repeated [‘ana’, ‘prune’] None Convolution nodes excluded from pruning.
input_nodes repeated [‘flops’] None Input nodes of the inference graph.
input_node_shapes repeated [‘flops’] None Shape of input nodes.
output_nodes repeated [‘ana’, ‘prune’, ‘flops’] None Output nodes of the inference graph.
channel_batch int [‘prune’] 2 The number of output channels is a multiple of this value after pruning.