如果某个段落是文字+图片，如何获得该段落的分数
#8
by
Herbert65
- opened
如题
The interleaved image and text document is not natively supported in the current compute_score API. Of course, you can hack the forward and prompt to support. However, this task is not optimized because the interleaved text and image documents are not included in our training data. Simply, I would recommend to compute scores on individual text and image content independently, and then use max/mean ops to assign the document score.
got it. Thank you!
numb3r3
changed discussion status to
closed