Spaces:
Paused
Paused
sergiopaniego HF Staff
Canonical env: reward weights from the environment, pool from the pinned dataset
2ac71c7 verified | # SPDX-License-Identifier: BSD-3-Clause | |
| """Watercolour environment for OpenEnv. | |
| The model writes a p5.brush sketch, the environment paints it in a headless | |
| browser and compares the result against a pool of reference paintings. The | |
| reward is comparative rather than absolute: a vision judge is asked which of | |
| two paintings is the better watercolour, in both presentation orders, because | |
| that is a question it answers stably and an absolute rating is not. | |
| Examples: | |
| ```python | |
| from envs.watercolour_env import WatercolourAction, WatercolourEnv | |
| with WatercolourEnv(base_url="http://localhost:8000") as env: | |
| observation = env.reset().observation | |
| reply = my_model(observation.system_prompt, observation.prompt) | |
| result = env.step(WatercolourAction(response=reply)) | |
| print(result.reward, result.observation.feedback) | |
| ``` | |
| """ | |
| from .client import WatercolourEnv | |
| from .models import WatercolourAction, WatercolourObservation, WatercolourState | |
| __all__ = [ | |
| "WatercolourEnv", | |
| "WatercolourAction", | |
| "WatercolourObservation", | |
| "WatercolourState", | |
| ] | |