LFM2.5-24B-A2B: Built for Fast On-Device Agents
Summary
Liquid AI announces LFM2.5-24B-A2B, a 24B-parameter model designed for fast, private on-device agents. The post covers architecture, training progression, benchmarks, and open-weight availability on Hugging Face and Playground, plus edge and GPU inference performance and a multi-stage post-training pipeline.