Fine-tune Trellis on my own dataset

[wangwhTj]

I want to use my own data to fine-tune the model. How much data would I need?

[XiaokunSun]

I have the same question, if I want to fine-tune this model, what is the minimum graphics card memory I need?

[Huaji173]

I have the same question, if I want to fine-tune this model, what is the minimum graphics card memory I need?

Yesterday I tried to fine-tune the model of stage 1 (ss_flow_model). When BatchSize was set to 1, the model occupied 23.3GB of GPU memory. And once the sampling step (i_sample) was reached, the program would report a "CUDA out of memory" error.

[loyris]

I have the same question, if I want to fine-tune this model, what is the minimum graphics card memory I need?

Yesterday I tried to fine-tune the model of stage 1 (ss_flow_model). When BatchSize was set to 1, the model occupied 23.3GB of GPU memory. And once the sampling step (i_sample) was reached, the program would report a "CUDA out of memory" error.

maybe A100 GPU can work?

[WiserZhou]

When the batch size == 8, the memory is about 78G using A800.

[GatorSethBarber]

@WiserZhou , how did you set the batch size? I can't find the place to do so in the configuration files