IFM open-sources RL360: RL infra behind 8,192 concurrent sandboxes and 1.3M rollouts
rupspace · x · 2026-10-23
IFM has open-sourced RL360, the RL infrastructure that trained K2 Horizon's six domain experts: 8,192 concurrent sandboxes, 150 steps in 7 days, and 1.3M rollouts of long-horizon agentic RL running thousands of agents in real sandboxes on minute-long tasks.
The core problem: sandbox crashes mid-task return the same low reward as a wrong answer, penalizing the model for system failures rather than its own mistakes and leaving GPUs underutilized. RL360 was built after months of isolating failures and fixing bottlenecks at scale.
Code and run instructions are available on GitHub.
Related event: IFM Open-Sources RL360, Massive Agentic RL Infrastructure Behind K2 Horizon(2 posts)→
More from coding & agent
- Claude cut a 2.6-hour video export to 20 minutes — but only after being told to rebuild the server — byhewar · 2026-10-23
- 3 flight-booking rules my AI agent checks that I'd never do by hand — Efistoffeles · 2026-10-23
- OpenJet claims Qwen3.8-27B coding runs 1.9x faster than llama.cpp on one RTX 3090 — JeremyCMorgan · 2026-10-23
- Building an autonomous SEO agent in Hermes with three skills: programmatic pages, Keywords Everywhere, DataForSEO — EXM7777 · 2026-10-23
- Building a sports betting AI agent with onchain execution and human approval by default: architecture and lessons — Dry-Comfortable-2678 · 2026-10-23
- How do you audit AI agent decisions in regulated workflows? A field survey — Ok_Way_1993 · 2026-10-23