Open yuanrr opened 7 months ago
This also puzzles me. Looking forward to your reply.
The code has been released in #53 .
Hi! For the instruction creation, please check the pipeline in our paper. Specifically, we use ChatGPT to summarize the abstract or introduction of the related dataset paper, obtaining the dataset description
. Then we manually write the task description
and instruction example
, thus using ChatGPT to generate similar ones via multi-turn revision.
I noticed that the video dataset such as NextQA has many different instructions for each sample, how can I create this? (in the provided files in
Ask-Anything/video_chat2/DATA.md
)