IO-AIR/arenamax

ArenaMax — human-preferred arena remix 89,589 rows · 55,691,246 real gpt2 tokens. A standalone remix dataset built from human-rated arena outputs: every assistant turn is either a human-chosen arena winner (decisive battles, loser dropped) or a rating-filtered best-of-N choice. Two configs: config file rows what sft sft.jsonl.gz 89,589 unified SFT rows (messages/trainable/gpt2_tokens/tags/meta) preference preference.jsonl.gz 89,589 (89,523 strictly distinct) DPO… See the full description on the dataset page: https://huggingface.co/datasets/IO-AIR/arenamax.

<iframe src="https://silytics.si/embed/item/1d0626c8-2139-4b9f-873c-61296cd89d87" 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 Information may be incomplete or out of date.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 22 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.