This dataset is a refreshed version of the DR-Tulu SFT trajectories from
rl-research/dr-tulu-sft-data.
The trajectory structure and all non-browse content are preserved. For real
assistant browse_webpage calls, the following <tool_output> block is
replaced with a raw Jina page fetch from
https://r.jina.ai/{url}, formatted as
the existing
tool-output schema. Page text is capped at
95,000 cl100k_base tokens, matching the long webpage-budget style… See the full description on the dataset page: https://huggingface.co/datasets/lihaoxin2020/dr-tulu-sft-v2.
