Workshop: Training Agents at the frontier of AI Industry

Training Agents IRL: four talks and a panel on post-training agents with open models Training Agents IRL is the in-person follow-up to the Training Agents livestream series. Speakers from Microsoft, Mercor, RadixArk and BenchFlow cover enterprise RL, environment design, scaling agent RL with Miles, and OpenEnv Arena. A panel and audience Q&A close the session. The four livestream classes teach the methods discussed here, step by step. If a talk makes you want to try something yourself, start there. CHAPTERS 00:00 Training Agents livestream series recap 01:46 Learnings from enterprise RL (Gobind, Microsoft) 19:23 Environments and benchmarks in industry (Victor, Mercor) 40:25 How to train your agents with Miles (Tao Lin, RadixArk) 52:40 Why we built OpenEnv Arena (Xiangyi Li, BenchFlow) 1:07:54 Panel discussion 1:38:28 Audience Q&A SLIDES All slides huggingface.co/buckets/burtenshaw/training-agents-irl Intro (Training Agents IRL) huggingface.co/buckets/burtenshaw/training-agents-irl/resolv Learnings from enterprise RL huggingface.co/buckets/burtenshaw/training-agents-irl/resolv How to train your agents with Miles huggingface.co/buckets/burtenshaw/training-agents-irl/resolv Why we built OpenEnv Arena huggingface.co/buckets/burtenshaw/training-agents-irl/resolv TRAINING AGENTS LIVESTREAM SERIES Full curriculum (blog) huggingface.co/blog/huggingface/train-agents Playlist youtube.com/playlist?list=PLo2EIpI_JMQvQZm-kVlz4wY1vWF0LBcf5 Class 1, fine-tune a coding agent on traces youtube.com/watch?v=rNgUoH7Wbv8 Code: huggingface.co/burtenshaw/gemma4-pi-mono-youtube-livestream- TRL SFT Trainer: huggingface.co/docs/trl/sft_trainer Class 2, distillation youtube.com/watch?v=GZT_oNQnLV4 Collection: huggingface.co/collections/sergiopaniego/training-agents-cla TRL Distillation Trainer: huggingface.co/docs/trl/distillation_trainer Class 3, reinforcement learning youtube.com/watch?v=ztdTed5egrM Collection: huggingface.co/collections/sergiopaniego/training-agents-cla TRL GRPO Trainer: huggingface.co/docs/trl/grpo_trainer Class 4, from reward functions to environments