Hugging Face releases Qwen3.8-27B coding adapter
Understand this story
Hugging Face's harness-hub team released Qwen3.8-27B-pi-adapted, a LoRA adapter for the pi coding agent, trained with AsyncGRPO reinforcement learning, rank 32, and released under the Apache-2.0 license. The team self-reports that the adapter raises SWE-bench Pro from 39.2% to 64.7%. The adapter applies only to Qwen3.8-27B and can run on vLLM with LoRA enabled, or load the GGUF adapter via llama.cpp on a Q8_0 base. The above performance figures are all self-reported by the team.
AI overview based on source reports · Updated 2 days ago
Coverage timeline
Follow the coverage to explore different perspectives.
- MarkTechPostNew LoRA Adapter Lifts Qwen3.8 Coding Success From 39% to 65%
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.
Activity for this story
- Current comparable activity
- 3
- Peak comparable activity
- 10Nov 19 02:00
- Change in the past 24 hours
- -50%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.