search

UbiquitousLearning / FedAdapter

"Efficient Federated Learning for Modern NLP", to appear at MobiCom 2023.
27 stars 2 forks source link

RuntimeError: CUDA error: invalid device ordinal #2

Open Transformer-man opened 3 months ago

Transformer-man commented 3 months ago

Hello, when I used a single card 4090, an error occurred, RuntimeError: cuda error: invalid device ordinal image

caidongqi commented 2 months ago

Do you check the gpu_mapping.yaml file?

Considering you have only one GPU, you should change line 13 to 4090: [6].

Make sure you have enough GPU memory to run 6 clients in one card. For your reference, each client will need at least 2GB GPU memory.