search

Noble-Lab / Polarbear

Polarbear translates between different single-cell data modalities
Apache License 2.0
7 stars 2 forks source link

Can not train Polarbear efficiently #3

Open Jiluzhang opened 10 months ago

Jiluzhang commented 10 months ago

Hi Dr.Zhang,

I am trying to train Polarbear using the data you provide (https://noble.gs.washington.edu/~ranz0/Polarbear/data/), but the training speed is too slow. Only when set batch_size to 1, the model can be trained (on one GPU with 80G memory). Even so, it will take much time (maybe more than half an hour) for one iteration. Could you please provide some training tips or pre-trained model?

Thanks so much Luz

RanZhang08 commented 10 months ago

Hi Luz,

Thanks for your question. The code was tested on our local machines with less than 8G GPU memory with a batch size of 16, could you please provide more information on your system (e.g. GPU memory, CPU memory)?

Thanks, Ran