Instructions to use devkumarhf/ur7e-sac-line-following with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- stable-baselines3
How to use devkumarhf/ur7e-sac-line-following with stable-baselines3:
from huggingface_sb3 import load_from_hub checkpoint = load_from_hub( repo_id="devkumarhf/ur7e-sac-line-following", filename="{MODEL FILENAME}.zip", ) - Notebooks
- Google Colab
- Kaggle
Download export_actor.py from devkumarhf/ur7e-sac-line-following: direct link, hf CLI and curl.
- Browser
- Download file 1.75 kB
-
https://huggingface.co/devkumarhf/ur7e-sac-line-following/resolve/main/export_actor.py
- Command line
-
hf download hf://devkumarhf/ur7e-sac-line-following/export_actor.py
-
curl -L -o export_actor.py https://huggingface.co/devkumarhf/ur7e-sac-line-following/resolve/main/export_actor.py
1.75 kB
| """Export the deterministic actor from the original trusted checkpoint.""" | |
| import os | |
| os.environ['CUDA_VISIBLE_DEVICES']='' | |
| from pathlib import Path | |
| import json | |
| import numpy as np | |
| import torch | |
| from checkpoint import load_model | |
| from ur7e_traj_env import UR7eTrajEnv | |
| from trajectory_generator import generate_trajectory | |
| from policy import Policy | |
| root=Path(__file__).resolve().parent;torch.set_num_threads(1);m=load_model();sd=m.actor.state_dict() | |
| keys={'w1':'latent_pi.0.weight','b1':'latent_pi.0.bias','w2':'latent_pi.2.weight','b2':'latent_pi.2.bias','wmu':'mu.weight','bmu':'mu.bias'} | |
| np.savez_compressed(root/'actor.npz',**{k:sd[v].cpu().numpy() for k,v in keys.items()});p=Policy() | |
| errors=[];observations=[] | |
| for seed in range(2026111700,2026111705): | |
| traj,_=generate_trajectory(seed=seed);e=UR7eTrajEnv(fixed_trajectory=traj);o,_=e.reset() | |
| for i in range(100): | |
| a=m.predict(o,deterministic=True)[0];na=p.predict(o);errors.append(float(np.max(np.abs(a-na))));observations.append(o) | |
| o,r,d,t,info=e.step(a) | |
| if d or t:break | |
| e.close() | |
| batch=np.array(observations,dtype=np.float32);expected=np.array([m.predict(o,deterministic=True)[0] for o in batch]);batcherr=float(np.max(np.abs(p.predict(batch)-expected))) | |
| assert max(errors)<1e-4 and batcherr<1e-4,(max(errors),batcherr) | |
| result={'samples':len(errors),'single_max_abs_action_difference':max(errors),'batch_max_abs_action_difference':batcherr,'tolerance':1e-4,'reference':'Original Stable-Baselines3 deterministic SAC.predict on CPU','format':'NumPy NPZ; allow_pickle=False','export':'Deterministic mean actor only; original stochastic policy and optimizers retained in best_model.zip.'} | |
| (root/'evaluation/actor-parity.json').write_text(json.dumps(result,indent=2)+'\n');print(json.dumps(result)) | |