inclusionAI/AReno

GitHub - inclusionAI/AReno: An easy-to-use, fast toolkit to scale up RL post-training on a single node.

🔥Ling-3.0-tiny × ASystem AReno: Local Agentic RL post-training toolkit is here 7.9B total, 1.3B active Run the full loop on a single DGX Spark—no cluster needed Supports Agentic RL, RL…

The asset is not tic-tac-toe. It is the loop: Baseline → feedback → post-train → same-environment retest. Use it for repeated, scoreable work: tool calls, structured extraction, domain instructions,…

Results after 400 steps: ✅rollout/rewards_mean: ~-0.5 → ~0.4 ✅response_len: down to ~850 tokens ✅fewer invalid actions ✅more stable tool calls and decisions

The model also became more concise. response_len dropped to ~850 tokens, meaning less wandering and more stable execution. For agentic tasks, that matters: the model should not only do the right…