In the prompt2model paper, we examined performance on several tasks, but performance was not as good on multilingual tasks. We're looking to improve performance on these tasks, so this is a meta-issue regarding this.
We will create individual issues for individual improvements and tag this issue there.
In the prompt2model paper, we examined performance on several tasks, but performance was not as good on multilingual tasks. We're looking to improve performance on these tasks, so this is a meta-issue regarding this.
We will create individual issues for individual improvements and tag this issue there.