CUDA out of memory using a custom mgf file about casanovo HOT 3 CLOSED

Hi,

You're likely getting this memory error because the model and training batches don't fit into a single GPU (pre-trained models we provided were trained on 2x 2080 GPUs). You can either try using a smaller batch size or decrease the number of model parameters by lowering dim_model, n_layers etc.

from casanovo.

@melihyilmaz
Hi,
I trained the model using two 2080Ti. If I used the default one mgf file from 9 species, the error of CUDA out of memory will not happen. Is there something wrong with my MGF file?

from casanovo.

I trained the model using two 2080Ti.

Does out of memory problem persist when you try training the model on 2 GPUs with your own data using training batch size 32? If so I'm not sure what the problem is.

You can check whether there's a problem with your mgf file by comparing it with one of the 9 mgf files we provide

from casanovo.