temp switch back to no profile mode

    fine_transformer = fine_transformer

)

with profile(activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA], record_shapes=True, profile_memory=True, use_cuda=True) as prof:

    with record_function("model_inference"):

generated_wav = audiolm(batch_size = 1)

output_path = f"{prefix}/out.wav"

sample_rate = 16000

torchaudio.save(output_path, generated_wav.cpu(), sample_rate)

filename = f"{prefix}/profile-{datetime.datetime.now().strftime('%Y-%m-%d-%H:%M:%S')}.txt"

with open(filename, "w") as f:

    f.write("cpu time sorted:\n")

    f.write(f"{prof.key_averages(group_by_input_shape=True).table(sort_by='cpu_time_total', row_limit=10)}")

    f.write("\n cuda time sorted:\n")

    f.write(f"{prof.key_averages().table(sort_by='cuda_time_total', row_limit=10)}")

    f.write("\ncpu memory self\n") # excludes children memory allocated

    f.write(f"{prof.key_averages().table(sort_by='self_cpu_memory_usage', row_limit=10)}")

    f.write("\ncpu memory\n") # includes children memory allocated

    f.write(f"{prof.key_averages().table(sort_by='cpu_memory_usage', row_limit=10)}\n")

# with profile(activities=[ProfilerActivity.CPU, ProfilerActivity.CUDA], record_shapes=True, profile_memory=True, use_cuda=True) as prof:

#     with record_function("model_inference"):

#         generated_wav = audiolm(batch_size = 1)

#         output_path = f"{prefix}/out.wav"

#         sample_rate = 16000

#         torchaudio.save(output_path, generated_wav.cpu(), sample_rate)

# filename = f"{prefix}/profile-{datetime.datetime.now().strftime('%Y-%m-%d-%H:%M:%S')}.txt"

# with open(filename, "w") as f:

#     f.write("cpu time sorted:\n")

#     f.write(f"{prof.key_averages(group_by_input_shape=True).table(sort_by='cpu_time_total', row_limit=10)}")

#     f.write("\n cuda time sorted:\n")

#     f.write(f"{prof.key_averages().table(sort_by='cuda_time_total', row_limit=10)}")

#     f.write("\ncpu memory self\n") # excludes children memory allocated

#     f.write(f"{prof.key_averages().table(sort_by='self_cpu_memory_usage', row_limit=10)}")

#     f.write("\ncpu memory\n") # includes children memory allocated

#     f.write(f"{prof.key_averages().table(sort_by='cpu_memory_usage', row_limit=10)}\n")