Hi, I’m trying to train my own model with tensorflow-gpu.
I followed the installation guide here https://www.tensorflow.org/install/gpu but when i try to launch Deepspeech.py, i get an error “Failed to get convultion algorithm. This is probably because cuDNN failed to initialize” and above “Could not create cudnn handle : cudnn_status_internal_error”.
Does someone have an idea to how I can fix that? Thanks
Ok i fixed that by adding:
config = tf.compat.v1.ConfigProto()
config.gpu_options.allow_growth = True
config.log_device_placement = True
sess = tf.compat.v1.Session(config=config)
in train.py
Glad you figured it out 
So you know for next time, you’ve left out too many details for someone else to be able to help you (unless they were sitting next to you)
@nmstoker is right, there is mostly nothing we can do to remotely help you, there’s 0 details on your setup, hardware, versions, and saying you followed this doc does not help really: how many times did we had people “follow the doc” and when asked for details, they were, indeed, not.
In particular, CuDNN errors can point to a lot of things, so with only half of the error message, nothing we can do. If allow growth helped, this would mean you might not have enough GPU RAM to fit the model + your batch size.
But again, I can only throw hypothesis.