Field noteOct 22, 20261 min read1 source

IFM open-sourced RL360 , the RL infrastructure behind K2 Horizon's six domain experts, under Apache 2.0.

In one minute

The Institute of Foundation Models (IFM) at MBZUAI has open-sourced RL360, a reinforcement-learning post-training codebase used for the K2 Horizon model family.

What changed: IFM has now open-sourced RL360, the reinforcement-learning infrastructure used behind K2 Horizon’s six domain experts, under Apache 2.0.

The GitHub repository is licensed under Apache 2.0.

The release includes source code, recipes, Docker material and components for the training infrastructure. Its repository history also documents Merge360, tooling for checkpoint merging, with methods including linear soup, task arithmetic, TIES, RAM/RAM+, ULP-gate, Gram-soup and ISO. A supplied recipe downloads a K2 Horizon 375B-A23B base model and five RL experts, then runs the documented merge settings.

IFM describes RL360 as the infrastructure behind K2 Horizon's six domain experts. The repository provides code and configuration rather than a hosted service or a newly released model. It does not establish the compute cost of training, the hardware required, independent reproduction of K2 Horizon results, or performance when the stack is applied to other models.

Public source: https://github.com/ifm-ai/RL360

Evidence / sources

  1. 1github.com/ifm-ai/RL360
All Field Notes

Next reads

Read next

Open the graph
  1. 01topicHarness EngineeringExplore the harness-engineering topic hub.
  2. 02topicAgent RuntimeExplore the agent-runtime topic hub.
  3. 03archiveField NotesOpen the latest editorial analysis.
  4. 04source ledgerSource LedgerInspect the public source evidence graph.
Who read this page?Machine requests, hidden until opened

Loading the privacy-safe route aggregate...

Open the JSON contract