Once an initial inference has been run with a model, sample dataset, and target, you can view performance results on the Configurations Page.
The components specified below provide visual representation of a model performance on a selected dataset and help find potential bottlenecks and areas for improvement:
The Model Analyzer is used for generating estimated performance information on neural networks. The tool analyzes of the following characteristics:
Parameter | Explanation | Unit of Measurement |
---|---|---|
Flop | Total number of floating-point operations required to infer a model. Summed up over known layers only. | Number of operations |
Iop | Total number of integer operations required to infer a model. Summed up over known layers only. | Number of operations |
Total number of weights | Total number of trainable network parameters excluding custom constants. Summed up over known layers only. | Number of weights |
Minimum Memory Consumption | Theoretical minimum of memory used by a network for inference given that the memory is reused as much as possible. Minimum Memory Consumption does not depend on weights. | Number of activations |
Maximum Memory Consumption | Theoretical maximum of memory used by a network for inference given that the memory is not reused, which means all internal feature maps are stored in the memory simultaneously. Maximum Memory Consumption does not depend on weights. | Number of activations |
Sparsity | Percentage of zero weights | Percentage |
Model analysis data is collected when the model is imported. All parameters depend on the size of a batch. Currently, information is gathered on the default model batch.
To view analysis data, click Details next to the name of a model in the table:
The details appear on the right: