running time too long

[Li-ZhiD]

I am testing one stereo-seq chip about 10K spots (bin50). The running time is too long:

model.train(device="cuda:0", sampler = "R")

0%|                                                                                            | 0[/800](http://192.168.101.101:7799/800) [00:00<?, ?it[/s](http://192.168.101.101:7799/s)][/mnt/data1/aaa/software/miniconda3/envs/sctm/lib/python3.8/site-packages/pyro/primitives.py:478](http://192.168.101.101:7799/mnt/data1/aaa/software/miniconda3/envs/sctm/lib/python3.8/site-packages/pyro/primitives.py#line=477): UserWarning: encoder.norm_topic.0.weight was not registered in the param store because requires_grad=False. You can silence this warning by calling my_module.train(False)
  warnings.warn(
Epoch Loss:6195.225:   0%|▏                                                          | 2[/800](http://192.168.101.101:7799/800) [13:12<86:38:45, 390.88s[/it](http://192.168.101.101:7799/it)]

[Chengwei94]

@Li-ZhiD I have no idea why its so slow. Did you check if you are using the GPU? If not I would think its an i/o issue if many people are using the same server.

[Li-ZhiD]

Thank you! I reduced my gene number by offering HVGs, and it finished with 2 hours.