Spaces:
Sleeping
Sleeping
Commit History
-
fix: clamp reward scores to strict (0, 1) range
Nitesh321 +1: Claude Opus 4.6Validator rejects scores of exactly 0.0 or 1.0. Clamp all rewards to [0.01, 0.99] before returning observations. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>843e0f8 -
fix: implement abstract methods in SQLAnalyticsEnv and use sync() wrapper
Nitesh321 +1: Claude Opus 4.6- Implement _step_payload, _parse_result, _parse_state in client.py - Use .sync() wrapper since EnvClient is async-first - Access StepResult.observation for env-specific fields (db_schema, error, result) - Access StepResult.reward and StepResult.done for step-level fields Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>016c7fd -
fix: rewrite inference.py to match official sample key=value log format
Nitesh321 +1: Claude Sonnet 4.6- Use HF_TOKEN/API_KEY instead of OPENAI_API_KEY - Default API base URL to HuggingFace router - Default model to Qwen/Qwen2.5-72B-Instruct - Log format: [START] task= env= model= (not JSON blobs) - [STEP] step= action= reward= done= error= - [END] success= steps= score= rewards= - [END] emitted in finally block (always runs) - Retry logic with exponential backoff Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>d8571a4 -
fix: pass api_key to OpenAI client from OPENAI_API_KEY env var
Nitesh321648f93a
-
fix: add uv.lock and server entry point in project.scripts
Nitesh321de5bfd8 -
fix: add HF_TOKEN and LOCAL_IMAGE_NAME env vars to inference.py per checklist
Nitesh3216749b1a -
fix: add HuggingFace Space metadata with app_port 8000
Nitesh3215ce4d9f -
feat: SQL Analytics OpenEnv environment — 10 tasks, dense rewards, RL-ready
Nitesh321798461f