Converting TinyStories-33M to ONNX

# Install dependencies (if not already installed)
pip install torch transformers onnxruntime onnxruntime-tools huggingface_hub

# Basic conversion of TinyStories-33M from Hugging Face
python3 ../scripts/universal_onnx_converter.py roneneldan/TinyStories-33M -o tinystories-33m.onnx --opset 14 -l 256

# Quantize to INT8 and save the quantized model
python3 ../scripts/universal_onnx_converter.py roneneldan/TinyStories-33M -o tinystories-33m.onnx --opset 14 -l 256 -q

# Search for models on Hugging Face (optional)
python3 ../scripts/universal_onnx_converter.py --search tinystories

pip install torch transformers onnxruntime

python convert_tinystories_optimized.py

mkdir -p ~/Documents/models
cp tinystories-33m-quantized.onnx ~/Documents/models/

# Create a virtual environment (optional)
python -m venv onnx_env
source onnx_env/bin/activate  # Linux/Mac
# or
onnx_env\Scripts\activate  # Windows

# Install dependencies
pip install -r requirements_convert.txt

# Basic conversion
python convert_tinystories.py

# Or optimized with quantization
python convert_tinystories_optimized.py

# Check file sizes
ls -lh *.onnx

# Test the model
python -c "
import onnxruntime as ort
session = ort.InferenceSession('tinystories-33m-quantized.onnx')
print('Model loaded successfully!')
print('Inputs:', [input.name for input in session.get_inputs()])
print('Outputs:', [output.name for output in session.get_outputs()])
"