Spaces:
Sleeping
Sleeping
Commit History
-
Clamp task scores to strictly (0, 1) — never exactly 0.0 or 1.0
- Increased epsilon from 0.001 to 0.01 so scores always display > 0 - Changed reward format from :.2f to :.4f for precision - Clamp done-step rewards in inference.py to [0.01, 0.99] - Fixes validator: "task scores are out of range" Submission by SpandanM11015e2e13
-
Clamp task scores to strictly (0, 1) — never exactly 0.0 or 1.0
- Added _clamp() helper: max(0.001, min(0.999, score)) - Applied to score_task1/2/3 final returns - Changed empty finalize penalty from -0.20 to 0.001 - Fixes Phase 2 Task Validation: scores out of range Submission by SpandanM110853ec0d -
Fix inference.py: always exit 0, fix import path resolution
- Always return exit code 0 (validator treats non-zero as crash) - Add script directory to sys.path for reliable env imports - Add traceback logging on task errors Submission by SpandanM11005ed675 -
Fix inference.py: support in-process env, use ENV_BASE_URL convention
- Runs environment in-process when ENV_BASE_URL is not set - Uses ENV_BASE_URL (OpenEnv standard) instead of ENV_URL - Added heuristic fallback agent when HF_TOKEN unavailable - No longer crashes when server is unreachable - Fixes Phase 2 validation: unhandled exception error Submission by SpandanM110fea1433 -
Fix inference.py: correct default port to 7860, add retry logic
- Changed ENV_URL default from localhost:8000 to localhost:7860 - Added wait_for_env() with 90s timeout to handle container startup - Made HF_TOKEN non-fatal (warning instead of crash) - Fixes Phase 2 validation: connection refused error Submission by SpandanM1104e71d7a -
Fix OpenEnv validation: add uv.lock, scripts entry, openenv-core dep, main()
- Added openenv-core>=0.2.0 to dependencies - Added [project.scripts] server entry point - Added main() function to server/app.py with if __name__ guard - Generated uv.lock via uv lock Submission by SpandanM11059d6790 -
Add pyproject.toml for OpenEnv validation
Submission by SpandanM11052a9626 -
Fix /reset endpoint to accept empty POST body for OpenEnv checker
- Made /reset parse body manually instead of requiring Pydantic model - Handles empty body, invalid JSON, and valid JSON with task field - Updated openenv.yaml app path to match actual entry point Submission by SpandanM1108702c9b -
Comprehensive README rewrite with full documentation
- Added HF Space badge and live demo link - Detailed architecture diagram and novel contributions table - Full action/observation space documentation - Reward function and grading engine breakdown - API endpoint reference with curl examples - Gradio UI tab descriptions - Setup instructions (local, Docker, inference) - Baseline performance table and tech stack Submission by SpandanM110c6b277d -
Drop Docling, PyMuPDF-only: fast multi-page PDF extraction, RL handles refinement by SpandanM110
50a7ab7 -
Fix Docker build: add system libs, PyMuPDF fallback for table extraction by SpandanM110
1a970fb -
Add system libs for Docling PDF processing (libxcb, libGL, poppler) by SpandanM110
11785c0 -
Switch to Docker SDK with FastAPI + mounted Gradio UI by SpandanM110
7433519 -
Fix inference.py to strictly match pre-submission checklist by SpandanM110
911b8c6 -
Integrate Docling for accurate table/KPI extraction from uploaded PDFs by SpandanM110
1285ffb -
Add PDF/image upload tab with document preview and grading by SpandanM110
287fea1 -
Switch to Gradio SDK, mount OpenEnv API on Gradio FastAPI app by SpandanM110
d1987c7 -
Add app_port: 7860 to Space metadata for Docker by SpandanM110
255ac95 -
Fix: redirect root / to Gradio UI at /ui by SpandanM110
715d086 -
Add Gradio UI, switch to Qwen2.5-72B-Instruct on HF Inference API by SpandanM110
5d245c2 -
Initial submission: OCR Table RL Environment for OpenEnv RL Challenge by SpandanM110
abe7bd6