pyromind/easyhard-24k

EasyHard-24K EasyHard-24K (v0.02) is a 24,061-sample easy/hard chat SFT dataset spanning coding, math, and multi-turn agent trajectories. The 19,076-sample core comes from comparing ground-truth (GT) assistant turns against Qwen3.5-4B rollouts on the same conversation prefix — easy when the small model already matches GT; hard when it does not (assistant turns are rebuilt with LLM-generated chain-of-thought + GT answers / tool_calls). GT corpora include… See the full description on the dataset page: https://huggingface.co/datasets/pyromind/easyhard-24k.

<iframe src="https://silytics.si/embed/item/3277d917-f9af-4964-81bf-b543cafdfdd2" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.