MarkTechPost· Asif Razzaq·· 2 days agoAI score58
New LoRA Adapter Lifts Qwen3.8 Coding Success From 39% to 65%
New LoRA Adapter Lifts Qwen3.8 Coding Success From 39% to 65%
AI summary
Hugging Face's harness-hub team released Qwen3.8-27B-pi-adapted, an Apache-2.0 rank-32 LoRA adapter trained with AsyncGRPO reinforcement learning to work inside the pi coding agent, raising SWE-bench Pro (hard 51) from 39.2% to 64.7% with thinking off and Terminal-Bench 2.1 from 39.5% to 55.8%. On tasks both models solved, median turns fell from 59 to 23 on SWE-bench Pro, and the adapter matches base-model-with-thinking accuracy at roughly 4.5x fewer tokens. The results are self-reported on small eval sets and apply only to the pi harness.
Source: MarkTechPost · marktechpost.com