Open wdr-RA02 opened 3 months ago
Hi, we have uploaded the instruction data:
conversation: https://drive.google.com/file/d/1E8UEH09y0CiAT4Hg7rm975AR3JCjEHeM/view?usp=drive_link reasoning: https://drive.google.com/file/d/1MTNHFzasCb0F921P0itaH-x8vN2OvxEu/view?usp=drive_link
As for the generation method of categorical data, it does not need to rely on GPT for creation; it can be directly produced (you can observe the prompt in all.py)."
Does the reasoning only contains 800 samples? I don't get it. In the paper I remember the statement of being trained on half the EmoSet-118k dataset @aimmemotion
Thanks in advance, keep it up with the great work!
Hello, first of all thank you for your amazing work!
I wonder if you have any plans for open-sourcing the Emotion Instruction Tuning data you created for instruct tuning the model mentioned in the paper. Due to the fact that GPT-4 API is unavailable for me right now, it is impossible to have an insight of more examples in the dataset.:( I would highly appreciate it if an organized dataset would be available.
Moreover, I assume that with a uniform dataset released would help build a uniformed benchmark or more, promoting the further developments in the field of emotional instruction tuning.
Thank you for your attention on this issue.