Hi，thank you for the excellent work. I am reproducing the pre-training phase, but noticed the repo only provides a single-GPU script while the paper uses 8x NVIDIA L40S. Could you please provide the 8-GPU (DDP) training script?
Also, I'd like to check if my current training speed is normal.
My Setup: 1x RTX 3090 (24GB), PyTorch 2.3.1.
Performance: 38,000 genes: 5 hours for only 2 epochs; 200 HVGs: Still slow(~22.5GB).
How long does the pre-training typically take on your 8-GPU setup?
Thanks!
Hi，thank you for the excellent work. I am reproducing the pre-training phase, but noticed the repo only provides a single-GPU script while the paper uses 8x NVIDIA L40S. Could you please provide the 8-GPU (DDP) training script?
Also, I'd like to check if my current training speed is normal.
My Setup: 1x RTX 3090 (24GB), PyTorch 2.3.1.
Performance: 38,000 genes: 5 hours for only 2 epochs; 200 HVGs: Still slow(~22.5GB).
How long does the pre-training typically take on your 8-GPU setup?
Thanks!