10.3. Exporting Models

This guide covers how to export trained models from Model Composer for deployment on TI microcontrollers.

10.3.1. Export Options

Model Composer provides several export formats:

Format

Description

CCS Project

Complete Code Composer Studio project

Artifacts Only

Model library and headers

ONNX Model

Intermediate model format

Configuration

YAML config for CLI tools

10.3.2. Exporting as CCS Project

The CCS Project export creates a ready-to-build project:

Step 1: Access Export

  1. Open your trained project in Model Composer

  2. Navigate to Export tab

  3. Select CCS Project

Step 2: Configure Export

  1. Select target device family

  2. Choose SDK version (if applicable)

  3. Select example type:

    • Minimal: Basic inference loop

    • Full: With ADC, GPIO examples

Step 3: Download

  1. Click Export

  2. Wait for packaging

  3. Download ZIP file

Step 4: Import to CCS

  1. Open Code Composer Studio

  2. File → Import → CCS Projects

  3. Select archive file (the ZIP)

  4. Click Finish

Project Contents:

MyProject_CCS/
├── .project              # CCS project file
├── .cproject             # CCS configuration
├── main.c                # Application entry
├── model/
│   ├── mod.a            # Compiled model
│   ├── mod.h            # Model interface
│   └── model_config.h   # Configuration
├── features/
│   ├── feature_extraction.c
│   └── feature_extraction.h
├── device/
│   └── device_config.c  # Device-specific init
└── linker.cmd           # Memory layout

10.3.3. Exporting Artifacts Only

For integration into existing projects:

Step 1: Select Artifacts Only

  1. Go to Export tab

  2. Select Artifacts Only

Step 2: Choose Components

Select which files to include:

  • Model library (mod.a) - Required

  • Model header (mod.h) - Required

  • Feature extraction code - Recommended

  • Test vectors - Optional

  • Configuration header - Optional

Step 3: Download

  1. Click Export

  2. Download ZIP

Artifact Contents:

artifacts/
├── mod.a                    # Model library
├── mod.h                    # Interface header
├── model_config.h           # Model parameters
├── feature_extraction.c     # Feature code
├── feature_extraction.h     # Feature header
└── test_vectors/            # (if selected)
    ├── input_0.csv
    └── expected_0.csv

Using Artifacts:

  1. Copy to your existing CCS project

  2. Add include paths for headers

  3. Add library to linker settings

  4. Call model functions from your code

10.3.4. Exporting ONNX Model

For custom deployment or analysis:

Step 1: Select ONNX

  1. Go to Export tab

  2. Select ONNX Model

Step 2: Choose Options

  • Float32 model: Original precision

  • Quantized model: INT8 quantized

  • Both: Download both versions

Step 3: Download

Receive ONNX file(s):

onnx_export/
├── model_float32.onnx    # Full precision
└── model_int8.onnx       # Quantized

Using ONNX:

  • Visualize with Netron

  • Run inference with ONNX Runtime

  • Custom compilation pipelines

  • Cross-platform deployment

10.3.5. Exporting Configuration

Export YAML config for CLI tools:

Step 1: Select Configuration

  1. Go to Export tab

  2. Select Configuration (YAML)

Step 2: Download

Receive config file:

# Exported from Model Composer
common:
  task_type: 'generic_timeseries_classification'
  target_device: 'F28P55'
  run_name: 'arc_fault_v1'

dataset:
  dataset_name: 'dc_arc_fault_example_dsk'
  input_data_path: '/path/to/dataset'

data_processing_feature_extraction:
  feature_extraction_name: 'FFT1024Input_256Feature_1Frame_Full_Bandwidth'
  variables: 1

training:
  model_name: 'ArcFault_model_400_t'
  training_epochs: 30
  batch_size: 256
  quantization: 2
  quantization_method: 'QAT'
  quantization_weight_bitwidth: 8
  quantization_activation_bitwidth: 8

testing:
  enable: True

compilation:
  enable: True

Using Configuration:

cd tinyml-modelzoo
./run_tinyml_modelzoo.sh exported_config.yaml

10.3.6. Batch Export

Export multiple projects at once:

  1. Select projects in dashboard (checkbox)

  2. Click Batch Export

  3. Choose export format

  4. Download combined ZIP

10.3.7. Export History

Track all exports:

  1. Go to Project → History

  2. View Exports tab

  3. Re-download previous exports

  4. Compare export configurations

10.3.8. Validating Exports

After exporting, validate the model:

Check Model Files:

# Verify library exists
ls -la mod.a

# Check header content
head mod.h

Test Vectors:

If test vectors were exported:

// In your test code
#include "test_vectors.h"

void validate_model(void) {
    float output[NUM_CLASSES];

    // Run inference on test vector
    mod_inference(test_input_0, output);

    // Compare with expected
    for (int i = 0; i < NUM_CLASSES; i++) {
        float diff = fabs(output[i] - expected_output_0[i]);
        if (diff > 0.01) {
            // Unexpected difference
            printf("Mismatch at %d: %.4f vs %.4f\n",
                   i, output[i], expected_output_0[i]);
        }
    }
}

10.3.9. Export Troubleshooting

Export Button Grayed Out:

  • Training must complete first

  • Check for compilation errors

  • Refresh the page

Download Fails:

  • Check browser popup blocker

  • Try different browser

  • Check internet connection

CCS Import Fails:

  • Verify CCS version compatibility

  • Check SDK installation

  • Try “Artifacts Only” instead

Model Doesn’t Work on Device:

  • Verify target device matches

  • Check memory constraints

  • Validate with test vectors

10.3.10. Best Practices

Before Export:

  1. Verify training completed successfully

  2. Check accuracy meets requirements

  3. Test on holdout data

  4. Document configuration

Export Selection:

  • New to CCS: Use CCS Project export

  • Existing project: Use Artifacts Only

  • Custom pipeline: Use ONNX

  • Reproducibility: Export Configuration

After Export:

  1. Validate model files

  2. Test with known inputs

  3. Profile on target device

  4. Document performance

10.3.11. Next Steps

After exporting: