Skip to content

generate-model-data.md

Latest commit

 

History

History
50 lines (41 loc) · 1.38 KB

generate-model-data.md

File metadata and controls

50 lines (41 loc) · 1.38 KB
Raw

model compilation script Guide

Overview

generate-model-data.py compiles TFLite model with RUHMI AI model compiler.

  1. Copy the model file to the shared directory being used by the Docker container.
  2. Compile the model with RUHMI.
python3 /generate-model-data.py -d model_deployment_dir -m mobilenet_v2_1.0_224_INT8.tflite

[NOTE] Remember to replace mobilenet_v2_1.0_224_INT8.tflite with the correct filepath of the model.
If RUHMI is being used natively, please update the command to use the path to the generate-model-data.py script.
Feel free to use a different output directory name instead of model_deployment_dir.

If the compilation completes successfully, you will see the following.

Output Directory Structure

The script creates one directory per model plus a top-level config.yaml.
The generated output will be ported into the application project.

<output_dir>/
  config.yaml
  <model_name>/
    input-0.bin
    expected-output-0.bin
    project.mdp
    build/
      IP/
        compilation/
        ir_dumps/
        ...

config.yaml is generated in a format similar to:

models:
  - name: mobilenet_v1
    data_directory: mobilenet_v1/
    inputs:
      - name: serving_default_input:0
        data_type: int8
        file_name: input-0.bin
        shape: [1, 224, 224, 3]