Debate: Will LLMs Absorb the Agent Harness, or Should Models Be Fine-Tuned
Sebastian Raschka pushes back on the idea that LLMs will absorb the agent harness, calling it part of training; meanwhile Hugging Face shows the opposite approach works, boosting Qwen3-27B's SWE-bench Pro from 39% to 65% via RL plus a small LoRA.
2026-11-21 ~ 2026-11-21 · 3 related posts
- Sebastian Raschka: the harness is part of training, LLMs can't absorb it — rasbt · 2026-11-21
- Qwen3-27B jumps from 39% to 65% on SWE-bench Pro with RL and one small LoRA — TheTuringPost · 2026-11-21
- rasbt on harness vs model fine-tuning: Qwen3.8-27B hit 65% on SWE-bench Pro — rasbt · 2026-11-21