Generate an operator configuration file from a list of tflite models

Signed-off-by: Georgios Pinitas <georgios.pinitas@arm.com>
Change-Id: I1b13da6558bd11d49747162d66c81255ccec1498
Reviewed-on: https://review.mlplatform.org/c/ml/ComputeLibrary/+/6166
Tested-by: Arm Jenkins <bsgcomp@arm.com>
Reviewed-by: SiCong Li <sicong.li@arm.com>
Reviewed-by: Sheri Zhang <sheri.zhang@arm.com>
Comments-Addressed: Arm Jenkins <bsgcomp@arm.com>

1 files changed, 37 insertions, 0 deletions

diff --git a/python/scripts/report-model-ops/README.md b/python/scripts/report-model-ops/README.md
new file mode 100644
index 0000000000..aa8efa88d3
--- /dev/null
+++ b/python/scripts/report-model-ops/README.md
@@ -0,0 +1,37 @@
+# Extract list of operators from a list of TfLite models
+
+## Introduction
+
+Purpose of this script is to inspect a list of user-provided TfLite models and report
+the list of operators that are used as well as the data-types that the models operate on.
+The script can subsequently generate a configuration file that can be provided to the
+Compute Library build system and generate a library that contains only the operators required
+by the given model(s) to run.
+
+Utilizing this script, use-case tailored Compute Library dynamic libraries can be created,
+helping reduce the overall binary size requirements.
+
+## Usage example
+
+Assuming that the virtual environment is activated and the requirements are present,
+we can run the following command:
+
+```bash
+./report_model_ops.py -m modelA.tfile modelB.tflite -c build_config.json
+```
+
+## Input arguments
+
+***models (required)*** :
+A list of comma separated model files.
+
+Supported model formats are:
+
+* TfLite
+
+***config (optional)*** :
+The configuration file to be created on JSON format that can be provided to ComputeLibrary's
+build system and generate a library with the given list of operators and data-types
+
+***debug (optional)*** :
+Flag that enables debug information