Running out of GPU memory

[wwxFromTju]

Hey all:

When I try to run the train.py, it take the all GPU memory, and I try to add per_process_gpu_memory_fraction, it can't work.

so how to change the GPU memory use?

[danijar]

PR #18 recently changed training on GPU. It is possible that it puts too many ops on the GPU. What tensors fails? What GPU do you use? Can you please set use_gpu = False in the config and verify that CPU-only training works?

[danijar]

@wwxFromTju Could you provide an update, please?

[wwxFromTju]

@danijar what update?

[danijar]

I asked a few questions in my first comment that could help debug your problem. Did you have a chance to try those?

[danijar]

I disabled GPU usage by default and implemented splitting episodes (using the chunk_length and batch_size config options) to be able to train on episodes with many time steps or large observations.