Spaces:
Sleeping
Sleeping
Commit History
-
Drop audioop-lts: ZeroGPU is Python 3.10, audioop is stdlib
Rishabh Jain07e2351 -
Pre-load models at import; wrap inference with longer-duration @spaces.GPU
Rishabh JainRoot cause of 400s CPU latency: pipeline was constructed lazily inside the GPU slice, so first-call weight download exceeded the 120s slice duration and all models fell back to CPU. Loading at import uses ZeroGPU's import-time CUDA access, and duration=300s covers the heaviest Stage 3 call.9b33910 -
Fix /info endpoint crash: pin gradio_client and guard bool schemas
Rishabh Jaind265d8e -
Merge remote-tracking branch 'origin/main'
Rishabh Jain# Conflicts: # requirements.txtaded968 -
Prepare Space for ZeroGPU deployment
Rishabh Jain- app.py: lazy-load models and wrap inference with @spaces.GPU - vlm.py: CPU-safe device_map and dtype branches for ZeroGPU cold boot - requirements.txt: add spaces, torch, Pillow, numpy; pin gradio>=4.44 - README.md: declare ZeroGPU hardware - .gitattributes: LFS rules for model weights - .gitignore: exclude non-runtime presentation assets - Remove residual CLAUDE.md references from docstrings842d4e9
-
Update requirements.txt
Verified signatureee009bb
-
Initial upload — depth-aware scene description system
Rishabh Jain +1: Claude Sonnet 4.6Stage 2 (VLM + Depth Anything V2) and Stage 3 (+ YOLOv8n) pipelines, evaluation suite, Gradio UI, and ARKitScenes test images. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>5412d82