CoolFace
Modelpublic

dhyuti-n/autorl-ppo-halfcheetah-v5

sourceHugging Faceupdated 4mo agoView on Hugging Face
0likes2downloads
Model Card

AutoRL — PPO on HalfCheetah-v5

Trained automatically by [AutoRL](https://github.com/wandb/autorl) — a multi-agent RL training race powered by W&B, Weave, and OpenAI.

Performance

MetricValue
Mean Return1246.39 ± 139.57
Steps Trained450,000
AlgorithmPPO
EnvironmentHalfCheetah-v5

Usage

python
# pip install stable-baselines3 huggingface_hub gymnasium mujoco imageio
from huggingface_hub import hf_hub_download
from stable_baselines3 import PPO
import gymnasium as gym
import imageio

# Load the winning model from HuggingFace
model_path = hf_hub_download(repo_id="dhyuti-n/autorl-ppo-halfcheetah-v5", filename="model.zip")
model = PPO.load(model_path)

# Record a video rollout
env = gym.make("HalfCheetah-v5", render_mode="rgb_array")
obs, _ = env.reset(seed=0)
frames, done = [], False
while not done:
    frames.append(env.render())
    action, _ = model.predict(obs, deterministic=True)
    obs, _, terminated, truncated, _ = env.step(action)
    done = terminated or truncated
env.close()

imageio.mimsave("rollout.mp4", frames, fps=30)
print("✓ Saved rollout.mp4")