Evaluation time on BrowseComp-Plus

[BaohaoLiao]

May I ask how much time do you use to sample the full set of BrowseComp-Plus? I use one GPU for retriever and 7 GPUs (B200) for OpenResearcher with the default setting, it takes 12 hours, and only finish 238 tasks.

[Zhuofeng-Li]

Sorry for the delay. On 8×A100 GPUs (with tp=2), it takes about 3 hours.