NaN or Inf found in input tensor.

[nanzhu2003]

When I run the train_offpolicy_with_trained_encoder.py with the environment of hopper, it occurs the problem that NaN or Inf found in input tensor, also, the success rate is always 0.0, I don't know why. My datasets is collected by myself, which separates the current state from the obeservation, and the obs_dim is 11.

Also，when I run the test_ood_context.py，which model should I load? The paramter of network is different from train_offpolicy_with_trained_encoder. Thank you a lot!