Hi
On my RTX 3080 with 10 GB memory, I get the following error while running rnnt.
[I] Starting running actual test.
terminate called after throwing an instance of 'std::runtime_error'
what(): Critical error in pipeline:
Error when executing GPU operator Spectrogram encountered:
Can't allocate 4653874752 bytes on device 0.
Current pipeline object is no longer valid.
I tested that with gpu_batch_size=512 and 1024. Is there a way to run that with lower parameters or it definitely needs more than 10 GB or memory? Any rule of thumb about minimum memory usage of this workload?
Hi On my RTX 3080 with 10 GB memory, I get the following error while running rnnt.
I tested that with gpu_batch_size=512 and 1024. Is there a way to run that with lower parameters or it definitely needs more than 10 GB or memory? Any rule of thumb about minimum memory usage of this workload?