datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
OSWorld2
OSWorld2 Kimi-K2.6 exact request traces
This dataset contains 98 complete OSWorld2 task traces recorded from a Kimi-K2.6 EPD 1P1D serving run. It is intended for exact virtual request replay in SGLang; replay sends the recorded model requests and token sequences without launching OSWorld VMs or re-executing computer actions.
Coverage
98 successful tasks
16,100 model requests
13,100 unique content-addressed screenshot blobs
6.72 GB of blob payloads before Hub… See the full description on the dataset page: https://huggingface.co/datasets/ergt2025/OSWorld2.osworld_tasks_files
