LiquidAI/LFM2-350M
LFM2-350M is a 354 million parameter hybrid model developed by Liquid AI, designed for edge AI and on-device deployment. It features a new architecture with multiplicative gates and short convolutions, offering 3x faster training and 2x faster inference speeds on CPU compared to Qwen3. This model excels in agentic tasks, data extraction, RAG, creative writing, and multi-turn conversations, outperforming similarly-sized models across various benchmarks.
Loading preview...
LFM2-350M: A Hybrid Model for Edge AI
LFM2-350M is part of Liquid AI's new generation of hybrid models, specifically engineered for efficient edge AI and on-device deployment. This 354 million parameter model introduces a novel architecture combining multiplicative gates and short convolutions, enabling significant performance gains.
Key Capabilities & Features
- Optimized Performance: Achieves 3x faster training and 2x faster decode/prefill speeds on CPU compared to Qwen3, while maintaining a 32,768 token context length.
- Superior Benchmarking: Outperforms other models of similar size across multiple categories, including knowledge, mathematics, instruction following, and multilingual tasks.
- Flexible Deployment: Designed to run efficiently on CPU, GPU, and NPU hardware, making it suitable for smartphones, laptops, and vehicles.
- Multilingual Support: Supports English, Arabic, Chinese, French, German, Japanese, Korean, and Spanish.
- Tool Use: Features a structured tool-use mechanism, allowing for function definition, calling, execution, and interpretation within conversations.
- Training: Trained on 10 trillion tokens using knowledge distillation from LFM1-7B, large-scale SFT, custom DPO, and iterative model merging.
Recommended Use Cases
LFM2-350M is particularly well-suited for fine-tuning on narrow use cases to maximize performance. It is recommended for:
- Agentic tasks
- Data extraction
- Retrieval Augmented Generation (RAG)
- Creative writing
- Multi-turn conversations
However, it is not recommended for knowledge-intensive tasks or those requiring advanced programming skills.