pip install stable-baselines3 huggingface_sb3 panda_gym shimmy

Then you can load and evaluate the model:

```python
from huggingface_sb3 import load_from_hub
from stable_baselines3 import A2C
from stable_baselines3.common.vec_env import DummyVecEnv, VecNormalize

# Load the model and statistics
repo_id = "LuckLin/a2c-PandaReachDense-v3"
filename = "a2c-PandaReachDense-v3.zip"

checkpoint = load_from_hub(repo_id, filename)
model = A2C.load(checkpoint)

# Load the normalization statistics
stats_path = load_from_hub(repo_id, "vec_normalize.pkl")
env = DummyVecEnv([lambda: gym.make("PandaReachDense-v3")])
env = VecNormalize.load(stats_path, env)

# At test time, we don't update the stats
env.training = False
env.norm_reward = False

# Evaluate
obs = env.reset()
for _ in range(1000):
    action, _states = model.predict(obs, deterministic=True)
    obs, rewards, dones, info = env.step(action)
    env.render()