--- schema: 1 type: video title: >- Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI description: >- A new class of small models is emerging with the ability to reliably follow instructions and call tools while running on-device under 1 GB of memory. In this talk, we'll break down how to post-train frontier small models using the LFM2.5 recipe: on-policy preference alignment, agentic reinforcement learning, and curriculum training with iterative model merging. We'll cover training challenges unique to the 1B scale, like doom loops, capability interference, and how to fix them. The goal is to give you a concrete playbook to fine-tune and deploy small models for your own use cases, from structured data extraction to multi-turn tool use. Speaker info: - https://x.com/maximelabonne - https://www.linkedin.com/in/maxime-labonne/ - https://github.com/mlabonne Timestamps: 0:00:00 - Start 0:00:14 - Introduction to frontier small models at Liquid AI 0:01:02 - Characteristics: memory-bound, task-specific, latency-sensitive 0:02:20 - Architecture: why large embedding layers are inefficient 0:04:01 - LFM2 architecture: using gated short convolutions for speed 0:06:09 - LFM 2.5 recipe: 28T tokens and post-training stages 0:08:34 - Post-training: SFT, preference alignment, and RL best practices 0:10:43 - Identifying "doom loops" in reasoning models 0:11:34 - Solutions: mitigating loops via preference alignment and RL 0:15:29 - Future focus: using agentic tools to overcome memory limits 0:17:58 - Q&A: real-world applications for small vs. large models --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/fLUtUkqYHnQ-everything-i-learned-training-frontier-small-models Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- canonical: 'https://liquidgold.cc/watch/fLUtUkqYHnQ' markdown: 'https://liquidgold.cc/watch/fLUtUkqYHnQ.md' llms: 'https://liquidgold.cc/watch/fLUtUkqYHnQ/llms.txt' parent: 'https://liquidgold.cc/channel/aidotengineer.md' channel: aidotengineer updated: '2026-10-24T14:43:18.191Z' language: null children: [] videoId: fLUtUkqYHnQ durationSeconds: 1213 publishedAt: '2026-04-29T12:00:06.000Z' searchChunks: - ordinal: 0 text: null startSeconds: 25 endSeconds: 72 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=25' heldReason: rights - ordinal: 1 text: null startSeconds: 72 endSeconds: 84 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=72' heldReason: rights - ordinal: 2 text: null startSeconds: 84 endSeconds: 142 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=84' heldReason: rights - ordinal: 3 text: null startSeconds: 155 endSeconds: 245 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=155' heldReason: rights - ordinal: 4 text: null startSeconds: 155 endSeconds: 245 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=155' heldReason: rights - ordinal: 5 text: null startSeconds: 256 endSeconds: 379 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=256' heldReason: rights - ordinal: 6 text: null startSeconds: 256 endSeconds: 379 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=256' heldReason: rights - ordinal: 7 text: null startSeconds: 390 endSeconds: 523 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=390' heldReason: rights - ordinal: 8 text: null startSeconds: 390 endSeconds: 523 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=390' heldReason: rights - ordinal: 9 text: null startSeconds: 523 endSeconds: 534 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=523' heldReason: rights - ordinal: 10 text: null startSeconds: 534 endSeconds: 645 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=534' heldReason: rights - ordinal: 11 text: null startSeconds: 534 endSeconds: 645 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=534' heldReason: rights - ordinal: 12 text: null startSeconds: 534 endSeconds: 645 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=534' heldReason: rights - ordinal: 13 text: null startSeconds: 656 endSeconds: 703 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=656' heldReason: rights - ordinal: 14 text: null startSeconds: 715 endSeconds: 930 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=715' heldReason: rights - ordinal: 15 text: null startSeconds: 715 endSeconds: 930 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=715' heldReason: rights - ordinal: 16 text: null startSeconds: 940 endSeconds: 1078 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=940' heldReason: rights - ordinal: 17 text: null startSeconds: 940 endSeconds: 1078 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=940' heldReason: rights - ordinal: 18 text: null startSeconds: 1089 endSeconds: 1213 watch: 'https://liquidgold.cc/watch/fLUtUkqYHnQ?t=1089' heldReason: rights --- # Everything I Learned Training Frontier Small Models — Maxime Labonne, Liquid AI A new class of small models is emerging with the ability to reliably follow instructions and call tools while running on-device under 1 GB of memory. In this talk, we'll break down how to post-train frontier small models using the LFM2.5 recipe: on-policy preference alignment, agentic reinforcement learning, and curriculum training with iterative model merging. We'll cover training challenges unique to the 1B scale, like doom loops, capability interference, and how to fix them. The goal is to give you a concrete playbook to fine-tune and deploy small models for your own use cases, from structured data extraction to multi-turn tool use. Speaker info: - https://x.com/maximelabonne - https://www.linkedin.com/in/maxime-labonne/ - https://github.com/mlabonne Timestamps: 0:00:00 - Start 0:00:14 - Introduction to frontier small models at Liquid AI 0:01:02 - Characteristics: memory-bound, task-specific, latency-sensitive 0:02:20 - Architecture: why large embedding layers are inefficient 0:04:01 - LFM2 architecture: using gated short convolutions for speed 0:06:09 - LFM 2.5 recipe: 28T tokens and post-training stages 0:08:34 - Post-training: SFT, preference alignment, and RL best practices 0:10:43 - Identifying "doom loops" in reasoning models 0:11:34 - Solutions: mitigating loops via preference alignment and RL 0:15:29 - Future focus: using agentic tools to overcome memory limits 0:17:58 - Q\&A: real-world applications for small vs. large models --- Watch along with the corrected transcript, summary, timestamps and resources on this talk's official AIE page: https://ai.engineer/talks/fLUtUkqYHnQ-everything-i-learned-training-frontier-small-models Join our upcoming AI Engineer conferences (see https://ai.engineer landing page): - AIEi Shanghai, November 5–6 — the first for meeting top Chinese labs and engineers - AIE CODE in SF, November 10–12 — the top AI Coding conference returns - AIEi Sydney, December 7–8 — the first AIEi run alongside NeurIPS 2026 in Sydney --- Canonical: https://liquidgold.cc/watch/fLUtUkqYHnQ · Channel: [AI Engineer](https://liquidgold.cc/channel/aidotengineer.md) > Search chunk text is held until public-source provenance is verified. ## Document chunks > Missing: no stored document chunks are available.