Spaces:
Sleeping
Sleeping
Commit History
-
Fix task validation: accept OpenEnv action dict format in /step, add /grade endpoint, run all 3 tasks in inference.py
- /step now accepts both {message: string} and {action: dict} formats so the platform's standard OpenEnv client can call it without 422 errors - Added POST /grade endpoint — platform can explicitly verify each grader returns a float score in [0,1] without running a full episode - StepResponse now includes score at top level (alongside info.graded_score) when done=True, making the graded score discoverable by any client - inference.py now runs all 3 tasks when called with no arguments so the platform sees [START]..[END] logs for each task during validationa7b1c20 -
Fix Phase 2 Task Validation: add missing OpenEnv required API endpoints
The hackathon platform's Task Validation runs openenv validate --url against the live HF Space. All 6 required criteria were failing because: - /health returned {"status":"ok"} instead of {"status":"healthy"} - /metadata endpoint was missing - /schema endpoint was missing (had /action_schema but not /schema) - /mcp JSON-RPC 2.0 endpoint was missing - /tasks response lacked "grader" field per task - server package excluded from setuptools, breaking uv run server Changes: - /health now returns {"status": "healthy"} - Added /metadata with env name, description, and tasks+graders list - Added /schema returning action/observation/state JSON schemas - Added /mcp with tools/list and tools/call (reset + step tools) - /tasks now includes grader path and has_grader:true per task - pyproject.toml includes server* in setuptools packages.find so uv run server works correctly openenv validate --url http://localhost:7860 now passes 6/6 criteria.b7854d3
-
server fixes
9d69b2d -
server fixes
37b014e -
minior fixes
e60ffc0 -
Fix FileNotFoundError for interactive_demo.html
a0552d4 -
Minor fixes
14f8b5d -
Add HTML landing page for HF Space App tab
9c5d93f -
Add root / endpoint to fix HF Space 404
cf64b85 -
Research improvements: physics fidelity + new server endpoints
Physics: - Effective length factors (k=0.7 braced / k=1.5 sway) for column buckling - RC shear wall code checks (EC2 simplified: in-plane shear + bending) Observation: - carbon_kg: CO2 footprint (1.85 kg/kg steel) - is_braced_frame + effective_length_factor: frame sway classification - displacement_hotspots: top-5 most displaced nodes for load-path awareness Server endpoints: - GET /query_forces?session_id&element_id — member forces (N, V, M, δ) - POST /what_if_remove — counterfactual ΔUR if element removed - GET /render?session_id&floor — SVG plan view color-coded by UR Cleanup: - Remove legacy pgsa_env/ package - 92/92 tests passing, openenv validate: OKebcc4f2 -
Rebuild as structural_design_env: replace pgsa_env with full OpenEnv implementation
- Replace pgsa_env/ with structural_design_env/ (Eurocode 3 physics, DSM solver, three-task graders, reward shaping, 5-consecutive-invalid termination) - Add server/ package with FastAPI app and main() entry point for openenv validate - Add inference.py at root (OpenAI-compatible, reasoning model think-tag stripping) - Add tests/ suite (92 tests passing) - Fix pyproject.toml: openenv-core dep, [project.scripts] server entry, updated URLs - Fix Dockerfile, openenv.yaml, .gitignore, README for new structure63dd587