Optimize models for edge deployment with quantization and pruning
Convert FP32 weights to INT8 for reduced model size and faster inference
Remove redundant connections while maintaining accuracy
Train smaller student model from larger teacher
Optimize for ONNX Runtime execution