Small Language Model inference pipeline optimized for edge device deployment using quantization, knowledge distillation, and model compression techniques. User queries flow through preprocessing and tokenization stages before reaching the SLM for inference, with model weights and language processing components enabling real-time predictions. This architecture demonstrates how to reduce model size and latency while maintaining accuracy on resource-constrained edge hardware. Fork this diagram on Diagrams.so to customize optimization parameters, add device-specific constraints, or integrate with OCI edge services. Knowledge distillation and quantization are applied pre-deployment to ensure the SLM runs efficiently without cloud connectivity.