Training Frontier Knowledge Work Agents A 397b Rl Training Guide With Skyrl

Includes a full SkyRL training script, weights and traces for RL on a 397B model, with ablations and de-risking steps. Teams training knowledge-work agents get a reproducible recipe and a 70% relative Pass@1 gain to compare against.

• token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
• eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.

• articleMonty Engineering Deep Dive
• articleHuman Baselines For Benchmarks Ai Now Outperforms Junior Accountants
• articleHow We 20x Llm Eval Throughput
• articleIntroducing The Ai Productivity Index For Accounting

• articlePutting Task Expertise Into RlThinking Machines
• clipOpen Thoughts Agents: RL environment curation and SFT vs RL tradeoffsAI Engineer
• clipBasic RL training loop for single-turn Q&AAI Engineer

Checking sign-in…

Loading comments…