jnjnnj

My solution: use one GPU to train for one or more epoches to create a checkpoint. Then continue training on multi-GPUs. It seems that RVC can train on multi-GPUs by...

Hi, @h-mayorquin, sorry for my late reply. Here are my thoughts. It is quite hard to "define" an elegent code style because such styles can be defined from many perspectives....

Well, I'm not sure if the function `write_binary_recording()` writes the metadata of scaling factors, but from my observation (comparing the binary data output by self-written codes and `write_binary_recording()`), it seems...

Hi, @h-mayorquin ,there are several reasons of using `write_binary_recording`. 1. we built our workflow based on the binary file as we were using kilosort3 before, while the function `recording.save_to_folder` seems...

Close this issue since the feature is implemented and discussion about I/O tutorial is moved to #3098

This problem is because the `spike_times.npy` and `spike_clusters.npy` are **assumed to be in seconds**. However, the two files generated by kilosort count in samples. And the easiest solution might be...

This might because of the orignal cluster number is larger than 200 and in this case, rastermap suggests to cluster first