AFM: Arcee Foundation Models

Meet AFM-4.5B—the brand-new Small Language Model (SLM) that kicks off Arcee AI’s next-gen Foundation Model family. With 4.5 billion parameters tuned for real-world enterprise demands, this model runs anywhere from smartphones to 512-GPU SageMaker HyperPods. In this session we sit down with Arcee’s CTO, Lucas Atkins, to unpack the full training pipeline: 6.58 trillion curated tokens, mid-training for precision, YaRN context-window scaling, MergeKit consolidation, supervised fine-tuning, RL with verifiable rewards, and KTO alignment. If you’re an engineer, data scientist, or ML nerd chasing state-of-the-art post-training stacks, hit that reminder and join us live! Join us every Wednesday at 1pm EST for our live events. SUBSCRIBE NOW to get notified! Apply for The AI Engineering Bootcamp today! https://bit.ly/AIEbootcamp LLM Foundations - Email-based course https://bit.ly/4izQVKk Full LLM Engineering Cohort - Open-Source! https://bit.ly/42ha7Xy For team leaders, check out! https://bit.ly/41YtlA7 Join our community to start building, shipping, and sharing with us today! https://bit.ly/43HzDEP Speakers: ​Dr. Greg, Co-Founder & CEO AI Makerspace https://www.linkedin.com/in/gregloughane The Wiz, Co-Founder & CTO AI Makerspace https://www.linkedin.com/in/csalexiuk/ 00:00 – Launch Day: AFM 4.5B — The Next-Level Open-Weight LLM 04:28 – First Impressions: Why This Model Changes the Game 06:45 – Inside the Architecture: RLU² Activations and More 11:34 – Breaking Down ReLU for AI Engineers 30:59 – Scaling Without Compromise: Design Choices That Matter 43:10 – Mid-Training Paradigm Shift: Massive Data, Smarter Strategies 51:42 – Hands-On Performance: Real-World Benchmarks 1:01:24 – Post-Training Experts: The Secret to Peak Model Output 1:05:22 – CTO Insights: Building AFM 4.5B from the Ground Up 1:07:24 – Your Move: Join the AI Engineer Challenge