Fetching metadata from the HF Docker repository...
Jun 15 16:59 Sync from GitHub 7c37dac9 Jun 8 03:16 Add AGENTS.md and pr_compliance_checklist.yaml; fix app.py location (#16) Carlos Salgado
* Add AGENTS.md and pr_compliance_checklist.yaml; fix app.py location
* Resolve merge conflict: keep merged AGENTS.md and root app.py from main
* Resolve merge conflict: keep root app.py, remove duplicate app/app.py, restore original tests/test_smoke.py
* consolidate documentation
* refactor Jun 6 22:29 fix: route inference through Modal endpoint instead of HF Router (#15) Carlos Salgado
* test: add E2E tests with real Modal/HF Router inference
- test_cvd_simulation_makes_buttons_indistinguishable: validates CVD pipeline
- test_analyze_with_vlm_produces_wcag_json: real VLM call, WCAG JSON structure
- test_run_analysis_full_pipeline: complete Gradio pipeline function
- 41/41 tests green
* fix: route inference through Modal endpoint instead of HF Router
app_space.py:
- Remove direct HF Router OpenAI calls (no more OpenAI import)
- Add _call_modal_analyze() helper: upload → predict → poll SSE
- analyze_with_vlm() now calls Modal endpoint (https://narwall-tech--color-ux-access-ui.modal.run/)
- Modal returns WCAG JSON via Gradio API: [dict] array → unwrap to plain dict
- HF_TOKEN not needed for inference; Space secrets only need MODAL_URL
tests/:
- tests/test_modal_integration.py: RED tests for Modal call (upload file → predict → poll)
- tests/test_vlm_model_selection.py: rewrite to test Modal path (mock _call_modal_analyze)
- tests/test_e2e.py: removed (tested old HF R… 16:17 fix: remove gradio>=6.0 from requirements_space.txt (SDK forces 5.0.0) socd06
refactor: use gr.__version__ for robust Gradio 5/6 detection
docs: update EVALUATION.md with actual done/pending state 07:54 feat: add color_ux_access package + CVD test fixtures socd06
- color_ux_access/cvd_sim.py: DaltonLens Machado2009 simulation wrapper
Supports: deuteranopia, protanopia, tritanopia, achromatopsia, and anomalous variants
- tests/generate_fixtures.py: synthetic UI screenshot generator
3 fixture pairs (bad/good): form validation, status badges, button panel
Each original has 3 CVD-simulated variants (deuteranopia, protanopia, tritanopia)
- tests/fixtures/: 24 PNG fixtures ready for TDD
- docs/CVD_USER_AUDIT_MODEL.md: user-centric audit framework 07:29 Update EVALUATION.md: NVIDIA Nemotron requirement, model-swappable architecture, task list socd06
Key updates:
- NVIDIA RTX 5080: Discord says Nemotron models required (need organizer confirmation)
- Model-swappable backend pattern: vlm/analyzer.py with MODELS dict
- gr.Dropdown in UI for VLM model selection (enables sponsor prize swaps)
- T3 Architecture: documented as done (screenshot → CVD → VLM → report pipeline)
- Added T1-T12 task list with current status
- Priority: MVP functional first, then polish, then sponsor optimizations
- gr.Workflow: not found in any Gradio release — CARLOS to provide livestream timestamp 07:02 Add sponsor prizes + bonus quests evaluation (docs/EVALUATION.md) socd06
- HF: primary target, 5K cash
- OpenBMB: K/track for MiniCPM-V 4.6 swap (1-line change)
- Modal: 50 credits all + 0K winners (already deployed)
- NVIDIA: RTX 5080 for top projects (no extra work)
- Cohere: already using aya-vision-32b (aligned)
- BFL Klein: skip (image gen, not relevant to VLM task)
- OpenAI: skip (not our model choice)
Bonus priorities: Blog (low effort/high impact), Off-brand theme, MiniCPM swap, Trace sharing
Skip: Fine-tune (high effort), BFL Klein, OpenAI models 06:43 Clean AGENTS.md: update task status, remove local paths, reflect current deployed state socd06
- T1/T2 marked done, T3-T8 active
- HF Router VLM marked deployed + returning JSON
- Removed local path G:\AI\HERMES\color-ux-access- Removed D:\CODE\narwall-selenium\ absolute paths (now generic references)
- Local llama.cpp marked as not wired to UI
- Removed G:\AI\HERMES\skills\ references 06:41 Reorganize repo structure + fix Gradio 6.0 deprecation socd06
Structure:
- app/ ← Gradio app + theme
- color_ux_access/ ← core package (capture, modal_app)
- vlm/ ← VLM inference backends + WCAG report generator
- tests/ ← TDD suite (13 passing)
- docs/ ← AGENTS.md, BONUS_PLAN.md
Deleted (stale/temp):
- *.backup, *.orig, *.before_fix, *.local
- old root-level smoke tests
- __pycache__/, .pytest_cache/, venv/
- personal screenshots (img/, screenshots/)
Added:
- .env.example ← HF_TOKEN template (not committed)
- .gitignore ← excludes .env, llama.cpp/, __pycache__/, etc.
Fixed:
- Gradio 6.0: theme param moved from Blocks() to launch()
- All test assertions synced with actual code (modal_app uses modal.App not Stub, etc.) 
