端侧AI模型部署与优化指南生成器

You are an expert in on-device AI deployment and model optimization. Help me deploy an AI model to run efficiently on edge devices.

My Setup:

• Target device: [smartphone / Raspberry Pi / embedded board - specify]
• Hardware specs: [CPU/GPU/NPU, RAM, storage]
• Model type: [LLM / vision / speech - specify]
• Base model: [model name and size]
• Latency requirement: [max acceptable inference time]
• Memory budget: [max RAM usage]

Please generate a complete deployment guide covering:

1. Model Optimization

• Quantization strategy (INT8/INT4/mixed-precision) with expected quality-speed tradeoffs
• Knowledge distillation options if the model is too large
• Layer pruning and architecture search recommendations
• Specific commands using tools like llama.cpp, ONNX Runtime, TensorRT, Core ML, LiteRT

2. Runtime Configuration

• Optimal inference engine for the target platform
• Thread/batch configuration
• Memory mapping and KV-cache optimization

3. Integration Code

• Minimal working example to load and run the optimized model
• Streaming output handling and error handling

4. Benchmarking

• How to measure tokens/sec, time-to-first-token, memory peak
• Comparison table template: original vs optimized model

5. Production Checklist

• Model versioning and OTA update strategy
• Privacy considerations for on-device inference

Provide concrete commands and code, not just theory.