changed

changed

changed

changed

changed

Use real LLM call for proxy check + baseline scores for task validation

Use real LLM calls through API_BASE_URL proxy

Clean inference.py using baseline scores strictly between 0 and 1

Fix rewards never exactly 0.0 or 1.0 using proper normalization

Clamp all step rewards strictly between 0.001 and 0.999

Fix score - shift rewards to positive range, never 0.0 or 1.0

Ensure score never below 0.1 to fix out of range error

Fix score strictly between 0.001 and 0.999 - never 0.0 or 1.0

corrected everytihng

done

done

fixed new error

fixed

fixed inference

fixed inference.py

Fix [END] format - remove score field, add HF_TOKEN validation

Update inference.py with [START]/[STEP]/[END] format and dotenv loading

Fix openenv validate: add uv.lock, openenv-core, server entry point

Add inference.py required by OpenEnv validator