Liquid Edge - High-Performance Edge Inference Runtime
Liquid Edge is a production-ready inference runtime designed specifically for edge computing environments. It provides high-performance LLM inference with multiple backend support, comprehensive tokenization capabilities, and optimized memory management.
🚀 Features
- 🎯 Multiple Backends: ONNX Runtime support with more backends planned
- ⚡ High Performance: Optimized for edge devices with minimal latency
- 🔄 Async Support: Full async/await support for non-blocking inference
- 💬 Chat Templates: Jinja2 template support for conversational AI
- 📊 Comprehensive Monitoring: Built-in metrics and performance tracking
- 🛡️ Production Ready: Robust error handling and logging
- 🔧 Flexible Configuration: Feature flags for customized builds
Model Directory Structure
models/my-model/
├── model.onnx # ONNX model file
├── tokenizer.json # HuggingFace tokenizer
├── config.json # Model configuration
├── tokenizer_config.json # Tokenizer configuration
├── special_tokens_map.json # Special tokens mapping
└── chat_template.jinja # Chat template (optional)