search

qiqiApink / MotionGPT

The official PyTorch implementation of the paper "MotionGPT: Finetuned LLMs are General-Purpose Motion Generators"
https://qiqiapink.github.io/MotionGPT/
189 stars 11 forks source link

Confusion of the input of training #10

Closed Harryqu123 closed 11 months ago

Harryqu123 commented 1 year ago

Hi there, thanks for your awesome work. I open this issue since I am confused with the following line of code (https://github.com/qiqiApink/MotionGPT/blob/main/finetune_motion.py#L185) Specifically, in this line of code, data[i]["input_ids"] is used as the input for training. However, according to the prepare step (https://github.com/qiqiApink/MotionGPT/blob/main/scripts/prepare_motion.py#L76), data[i]["input_ids"] seems to be the one that involves GT output. Can I ask should data[i]["input_ids_no_response"] be used here instead? I am sorry if I misunderstood anywhere of the code and hope can get your reply. Many thanks in advance.

Harryqu123 commented 11 months ago

I find my answer here and sorry for disturb. https://github.com/Lightning-AI/lit-llama/issues/279