odunola/lfm-350m
LFM2.5-350M by Liquid AI is a 350 million parameter instruction-tuned hybrid model, building on the LFM2 architecture with extended pre-training and reinforcement learning. Optimized for on-device deployment, it offers fast edge inference and rivals larger models in performance. This model excels at data extraction, structured outputs, and tool use, supporting multiple languages and a 32,768 token context length.
Loading preview...
LFM2.5-350M: A Compact Model for Edge Deployment
LFM2.5-350M, developed by Liquid AI, is a 350 million parameter instruction-tuned model designed for efficient on-device deployment. It leverages a hybrid architecture, extending the LFM2 framework with 28 trillion tokens of pre-training and large-scale multi-stage reinforcement learning. This model is engineered to deliver high-quality AI performance in a compact footprint, rivaling much larger models.
Key Capabilities & Features
- On-Device Optimization: Achieves fast inference speeds (e.g., 313 tok/s on AMD CPU, 188 tok/s on Snapdragon Gen4) and operates under 1GB of memory.
- Broad Compatibility: Supports various inference frameworks including llama.cpp, MLX, and vLLM, with day-one support for optimized formats like GGUF, ONNX, MLX, and OpenVINO.
- Extended Context & Multilingual Support: Features a 32,768 token context length and supports English, Arabic, Chinese, French, German, Japanese, Korean, Portuguese, and Spanish.
- Advanced Tool Use: Designed for robust function calling, allowing for structured outputs and integration with external tools through a defined chat template and Pythonic function calls.
Ideal Use Cases
- Data Extraction: Efficiently extracts specific information from text.
- Structured Outputs: Generates responses in predefined formats.
- Tool Use: Integrates seamlessly with external tools and APIs for complex tasks.
While highly capable for these specific tasks, it is not recommended for knowledge-intensive tasks or programming due to its specialized optimization.