search

snap-stanford / UCE

UCE is a zero-shot foundation model for single-cell gene expression data
MIT License
136 stars 21 forks source link

Questions about average running time #9

Closed HelloWorldLTY closed 9 months ago

HelloWorldLTY commented 9 months ago

Hi, I am using a cluste with one rtx3090 gpu, but the running time for the demo dataset seems very strange. It seems that some people used only macbook with M1 chip and successfully run the demo case, but I need quite long time here:

image

Is it possible for me to know the reason? What is the expected running time for demo dataset using for exmaple, one A100 GPU? Thanks.

Yanay1 commented 9 months ago

That definitely seems slower than it should be-- can you check if the GPU is being used? It could be that it is running on CPU instead of gpu. On an A100 (80GB) it takes less than 5 minutes to embed the demo dataset.

HelloWorldLTY commented 9 months ago

Thanks, but I install all of the pacakges based on the instructions. I will try and get back to you.

HelloWorldLTY commented 9 months ago

Thanks a lot. It seems work well now. image

I think the problem is caused by installing the accelerate package via conda rather than pip. Would you please add this package to the requirement.txt file? Thanks.

Yanay1 commented 9 months ago

Good catch! I updated the requirements to include accelerate (rather than accelerator).