Closed zhaosheng-thu closed 2 months ago
Thank you for your interest. Since my GPU (A6000) cannot handle bsz > 1 due to VRAM constraints, I didn’t address that case. You might need to slightly modify the code to make it work well if it doesn’t currently.
Thank you for your interest. Since my GPU (A6000) cannot handle bsz > 1 due to VRAM constraints, I didn’t address that case. You might need to slightly modify the code to make it work well if it doesn’t currently.
Thanks for your prompt reply. I have tried a bs > 1, and I find that the conditioner ConcatTimestepEmbedderND will return a Tensor with a wrong shape. So for this repository, it currently doesn't support bs > 1. Closed.
Thanks for your work! I'm wondering whether I can choose a batch_size > 1 in the data config during the finetuning procedure. Thanks.