Why it matters
Ai2 openly publishes models and infra decisions, so this is a chance to hear directly from a lab about training tradeoffs that later shape a model's cost and behavior.
The tokenmaxxing angle
Talks on scaling training runs and post-training/eval describe the training-time choices that ultimately determine a released model's efficiency and cost-per-token at inference.
From the organizers
Event description confirms a behind-the-scenes look covering "scaling large training runs to fine-tuning and reinforcement learning," held at Ai2's Seattle office as part of Seattle Tech Week.