I'm very excited about your project. I have some questions regarding the tasks. I noticed that you used the "t2m" parameter when instantiating the model in the webui and demo, does this mean that MotionGPT can only perform t2m, and not m2t or m2m tasks simultaneously on a single model? During training, are all tasks trained together to obtain a mixed model, or is each task trained separately to obtain a MotionGPT model dedicated to that specific task?
Additionally, I want to train a new task that does not appear in the template_instruction.json. Which parts of the code should I pay attention to and modify? Is there any guidance available?
Thank you in advance for your response. Your project has inspired me a lot！
I'm very excited about your project. I have some questions regarding the tasks. I noticed that you used the "t2m" parameter when instantiating the model in the webui and demo, does this mean that MotionGPT can only perform t2m, and not m2t or m2m tasks simultaneously on a single model? During training, are all tasks trained together to obtain a mixed model, or is each task trained separately to obtain a MotionGPT model dedicated to that specific task?
Additionally, I want to train a new task that does not appear in the template_instruction.json. Which parts of the code should I pay attention to and modify? Is there any guidance available?
Thank you in advance for your response. Your project has inspired me a lot！