Hugging Face 🤗 x Stable-baselines3 v3.0

A library to load and upload Stable-baselines3 models from the Hub with Gymnasium and Gymnasium compatible environments.

⚠️ If you use Gym, you need to install huggingface_sb3==2.3.1

Installation

With pip

pip install huggingface-sb3

Examples

We wrote a tutorial on how to use 🤗 Hub and Stable-Baselines3 here

If you use Colab or a Virtual/Screenless Machine, you can check Case 3 and Case 4.

Case 1: I want to download a model from the Hub

import gymnasium as gym

from huggingface_sb3 import load_from_hub
from stable_baselines3 import PPO
from stable_baselines3.common.evaluation import evaluate_policy

# Retrieve the model from the hub
## repo_id = id of the model repository from the Hugging Face Hub (repo_id = {organization}/{repo_name})
## filename = name of the model zip file from the repository
checkpoint = load_from_hub(
    repo_id="sb3/demo-hf-CartPole-v1",
    filename="ppo-CartPole-v1.zip",
)
model = PPO.load(checkpoint)

# Evaluate the agent and watch it
eval_env = gym.make("CartPole-v1")
mean_reward, std_reward = evaluate_policy(
    model, eval_env, render=False, n_eval_episodes=5, deterministic=True, warn=False
)
print(f"mean_reward={mean_reward:.2f} +/- {std_reward}")

Case 2: I trained an agent and want to upload it to the Hub

With package_to_hub() we'll save, evaluate, generate a model card and record a replay video of your agent before pushing the repo to the hub. It currently works for Gym and Atari environments. If you use another environment, you should use push_to_hub() instead.

First you need to be logged in to Hugging Face:

• If you're using Colab/Jupyter Notebooks:

from huggingface_hub import login

login()

• Else:

huggingface-cli login

For more details about authentication, check out this guide.

Then

With package_to_hub():

import gymnasium as gym

from stable_baselines3 import PPO
from stable_baselines3.common.env_util import make_vec_env
from huggingface_sb3 import package_to_hub

# Create the environment
env_id = "LunarLander-v2"
env = make_vec_env(env_id, n_envs=1)

# Create the evaluation env
eval_env = make_vec_env(env_id, n_envs=1)

# Instantiate the agent
model = PPO("MlpPolicy", env, verbose=1)

# Train the agent
model.learn(total_timesteps=int(5000))

# This method save, evaluate, generate a model card and record a replay video of your agent before pushing the repo to the hub
package_to_hub(model=model, 
               model_name="ppo-LunarLander-v2",
               model_architecture="PPO",
               env_id=env_id,
               eval_env=eval_env,
               repo_id="ThomasSimonini/ppo-LunarLander-v2",
               commit_message="Test commit")

With push_to_hub(): Push to hub only push a file to the Hub, if you want to save, evaluate, generate a model card and record a replay video of your agent before pushing the repo to the hub, use package_to_hub()

import gymnasium as gym

from stable_baselines3 import PPO
from stable_baselines3.common.env_util import make_vec_env
from huggingface_sb3 import push_to_hub

# Create the environment
env_id = "LunarLander-v2"
env = make_vec_env(env_id, n_envs=1)

# Instantiate the agent
model = PPO("MlpPolicy", env, verbose=1)

# Train it for 10000 timesteps
model.learn(total_timesteps=10_000)

# Save the model
model.save("ppo-LunarLander-v2")

# Push this saved model .zip file to the hf repo
# If this repo does not exists it will be created
## repo_id = id of the model repository from the Hugging Face Hub (repo_id = {organization}/{repo_name})
## filename: the name of the file == "name" inside model.save("ppo-LunarLander-v2")
push_to_hub(
    repo_id="ThomasSimonini/ppo-LunarLander-v2",
    filename="ppo-LunarLander-v2.zip",
    commit_message="Added LunarLander-v2 model trained with PPO",
)

Case 3: I use Google Colab with Classic Control/Box2D Gym Environments

• You can use xvbf (virtual screen)

!apt-get install -y xvfb python-opengl > /dev/null 2>&1

• Just put your code inside a python file and run

!xvfb-run -s "-screen 0 1400x900x24" <your_python_file>

Case 4: I use a Virtual/Remote Machine

• You can use xvbf (virtual screen)

xvfb-run -s "-screen 0 1400x900x24" <your_python_file>

Case 5: I want to automate upload/download from the Hub

If you want to upload or download models for many environments, you might want to automate this process. It makes sense to adhere to a fixed naming scheme for models and repositories. You will run into trouble when your environment names contain slashes. Therefore, we provide some helper classes:

import gymnasium as gym
from huggingface_sb3.naming_schemes import EnvironmentName, ModelName, ModelRepoId

env_name = EnvironmentName("seals/Walker2d-v0")
model_name = ModelName("ppo", env_name)
repo_id = ModelRepoId("YourOrganization", model_name)

# prints 'seals-Walker2d-v0'. Notice how the slash is removed so you can use it to 
# construct file paths if you like.
print(env_name)

# you can still access the original gym id if needed
env = gym.make(env_name.gym_id)  

# prints `ppo-seals-Walker2d-v0`
print(model_name)  

# prints: `ppo-seals-Walker2d-v0.zip`. 
# This is where `model.save(model_name)` will place the model file
print(model_name.filename)  

# prints: `YourOrganization/ppo-seals-Walker2d-v0`
print(repo_id)