PTeterwak/openwebrl-arm-matched-pairs

OpenWebRL ARM preference pairs, matched action counts These are preference pairs for fine-tuning a browser agent (OpenWebRL-4B-SFT, Qwen3-VL-4B) with DPO or RPO. Each row is a browser state with a chosen and a rejected candidate response. Only pairs whose chosen and rejected responses make the same number of tool calls are kept. RPO (DPO + NLL on the chosen response, lr 5e-5, β 0.1) on these pairs gets 40.0% on Online-Mind2Web (300 tasks, o4-mini judge, invalid episodes counted… See the full description on the dataset page: https://huggingface.co/datasets/PTeterwak/openwebrl-arm-matched-pairs.

<iframe src="https://silytics.si/embed/item/5c2aeb1e-b468-46d5-99d5-9dc2ae0e1081" 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 Information may be incomplete or out of date.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 23 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.