OOM error

[cyxu2017]

what is the GPU memory requirement for running prediction? Is it system dependent? If so is there a simple way to estimate the memory required?

I was running an inference for a complex with N_asym 6, N_token 2372, N_atom 18500, N_msa 4940 on a GPU with 24 GB memory. and the job was killed by a OOM:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 5.37 GiB. GPU

[zhangyuxuann]

@cyxu2017 it's need about 40G+ in your settings. you can refer to https://github.com/bytedance/Protenix/blob/main/docs/model_train_inference_cost.md