Originally published on dev.tohttps://dev.to/namakoo/the-day-my-agent-started-having-hunches-f7o. Reposted here in full. My last post here was about curating training data for LLMs. This one is stranger: it is about giving a coding agent an involunt...

Over 36 hours we ran four DPO training iterations against Qwen2.5-Coder-7B-Instruct, trying to push HumanEval pass@1 above the base model's 87.20%. The first three iterations failed in different ways -9.15pp, -1.22pp, two NO-GO calls. The fourth rec...

In the previous posthttps://dev.to/namakoo/-curating-python-failures-for-dpo-notes-from-the-rejected-side-2ff0, I described the curation philosophy for IDFU's rejected-side dataset — why I avoid synthetic bug generation, why stub detection matters, w...