[Bug]: Can't create the PPO model on Macbook M1 #1152
Labels
bug
Something isn't working
Comments
|
I'm not sure it is actually duplicated. Looking the code, it is supposed run on the CPU. |
@qgallouedec You're right. I've run the code on the CPU using Jupyter Notebook inside VSCode. |
|
Thank you. Can you make the code minimal? Therefore, the last line is the one that raises the issue. I think it is a problem of gym rendering in the notebook. |
Okay, let me try to walk you through step by step to reproduce the error.
Error info: Actually, when I run |
|
My intuition was bad. For the moment, I fail in reproducing the bug. |
Sure, here you go. |
|
Do you agree that the minimal code would be from stable_baselines3 import PPO
from stable_baselines3.common.env_util import make_vec_env
env = make_vec_env("CartPole-v1", n_envs=4)
PPO("MlpPolicy", env)Does it work without multiple envs?: from stable_baselines3 import PPO
PPO("MlpPolicy", "CartPole-v1") |
Without multiple envs, it still has the same error. For example:
|
|
I have the same configuration as you and I do not encounter any errors. I've noticed several times that the VSCode jupyter extension is buggy and sometimes behaving a bit strangely. I remember solving one of them by simply reinstalling the virtual environment completely and rebooting my laptop. I think it's really worth a try. |
|
Possibly memory-related: microsoft/vscode-jupyter#9741 (comment). It would explain why the reboot worked for me. |
|
@qgallouedec I've tried rebooting the laptop and reinstalling the Anaconda virtual environment but the error remains the same :( |
|
Does the code fail when executed in a standard python file? |
When running with normal python file (it's my |
|
I've found a package to help me find out the bug. Just import it and add Having run the script with each of those models (PPO, A2C, DQN), I got the result: With model A2C, the error is And with model DQN, it works fine. |
|
Could you run this: import torch.nn as nn
linear = nn.Linear(2, 2)
nn.init.orthogonal_(linear.weight) |
Yepp, no problems with that. |
|
Do you still encounter the error? What Python version do you use? |
|
I can still notice this behaviour in sb3 1.6.2 and python3.9 .Exact same bug "Fatal Python error: Segmentation fault" in case of PPO but DQN is working fine. Please let me know for any additional log |
|
Have a MacBook Pro M1, I can't reproduce the error. The following works as expected: from stable_baselines3 import PPO
PPO("MlpPolicy", "CartPole-v1")
If anyone finds a code/config to reproduce this error, please post it here. |
|
@qgallouedec the exact same two lines i tried running in my local vs code . I hope this has nothing do with which i de i am using.
|
|
What are the Jupyter logs? |
|
Sorry for late response. I resolved my issue just now by reinstalling numpy which is causing the problem. |
|
Late = 18 minutes ? 😆 Great to hear! I'm closing the issue. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
🐛 Bug
I'm using Macbook Pro M1 and running code in Jupyter Notebook. Every time I run the scripts, I get an error and the Kernel crashed. I've tried running the code in Google Colab and it went well. It seems to me that
stable-baselines3library doesn't support M1 chips. What do I need to do now to run the code locally?To Reproduce
Relevant log output / Error message
System Info
No response
Checklist
The text was updated successfully, but these errors were encountered: