Frontier Agentic Models for the Edge

A new class of small agentic models is emerging — models that call tools and complete tasks while running entirely on phones and laptops. They point to a future where capable AI agents live at the edge, not just in the data center.

The session

In “Frontier Agentic Models for the Edge,” Maxime Labonne breaks down how to post-train small agentic models using the LFM2.5 recipe: on-policy preference alignment, agentic reinforcement learning, and curriculum training with iterative model merging.

He'll cover the training challenges unique to the 1B scale — doom loops, capability interference, and how to fix them — and share a concrete playbook for fine-tuning and deploying small models across real use cases, from structured data extraction to multi-turn tool use.

Why it matters now

As inference costs and privacy concerns push AI toward the edge, small capable agents are becoming strategically important. This session is a practical, reproducible playbook for building them.

About the speaker

Maxime Labonne is Head of Post-Training at Liquid AI. He holds a Ph.D. in Machine Learning from the Polytechnic Institute of Paris, is a Google Developer Expert in AI/ML, and authored the best-selling “LLM Engineer's Handbook” and “Hands-On Graph Neural Networks Using Python.” He created the popular open-source LLM Course and best-in-class LFM2/2.5 models.

Key Takeaways

1. The LFM2.5 recipe — on-policy alignment, agentic RL, and curriculum training with iterative model merging.
2. Solving 1B-scale issues — how to diagnose and fix doom loops and capability interference.
3. A deployable playbook — ship small agentic models for real tasks, from data extraction to tool use.

Session at a Glance