350M (LFM2 hybrid architecture; 16 layers: 10 double-gated LIV conv + 6 GQA)
LFM Open License v1.0
Open-weight 350M instruction-tuned LFM2.5 release positioned as a small-model breakthrough for on-device AI. Official release materials claim best-in-class quality at this size, highlighting that LFM2.5-350M rivals much larger models while sustaining fast edge inference (reported 313 tok/s decode on AMD CPU and 188 tok/s on Snapdragon Gen4) with sub-1GB memory deployment targets. The model card presents scaled training from 10T to 28T tokens plus large multi-stage reinforcement learning, 32,768 context length, and multilingual support (EN, AR, ZH, FR, DE, JA, KO, PT, ES). Release guidance emphasizes practical use in data extraction, structured outputs, and tool use, with day-one support across Transformers, vLLM, llama.cpp, MLX, and ONNX formats (via companion checkpoints). Benchmark tables in the release report strong gains versus prior LFM2-350M and several other small-model baselines across IFEval/IFBench/Multi-IF and tool/function-calling-oriented evaluations.