Closed axkoenig closed 3 years ago
Make sure that your Python, PyTorch and stable-baselines3 versions match. Unfortunately some of the saving relies on Python's pickling, which does not always play nice between Python versions (same can happen when PyTorch version changes).
Thanks for this lightning fast reply! That was it, thank you! I was on a different stable_baselines3 version ...
Question
I am training models on a research cluster with SB3 and a custom environment. I then transfer the saved
.zip
models viasftp
to my local computer to visualize the learned behavior in my robotics simulator. I run into an annoying problem here, since somehow the.zip
file is corrupted on the way. When I runmodel.load("my_path")
I get the following error:Loading the model into a new session on the research cluster works without problems, so I really think the model transfer is the problem. How do you guys transfer your trained models from A to B?
Additional context
Here is my main script if this is of any interest