Spaces:
Paused
Paused
Commit History
-
Use PyTorch cu124 to match system CUDA 12.4
srayuth +1: Claude Opus 4.6Install PyTorch built for CUDA 12.4 instead of CPU-only, matching the nvidia/cuda:12.4.1 base image to avoid symbol conflicts while enabling GPU for YOLO too. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>bd50768 -
Install CPU-only PyTorch to fix CUDA symbol conflict
srayuth +1: Claude Opus 4.6PyTorch's bundled CUDA runtime conflicts with the system CUDA from the NVIDIA base image (undefined symbol: cudaGetDriverEntryPointByVersion). YOLO only needs CPU for lightweight object classification (~3s). rembg still uses GPU via onnxruntime-gpu. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>6a07f6e -
Fix cuDNN version: use cuda:12.4.1-cudnn for cuDNN 9
srayuth +1: Claude Opus 4.6onnxruntime-gpu requires libcudnn.so.9 (cuDNN 9), previous image only had cuDNN 8. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>56927ac -
CUDA base image, pre-cache models, fix double loading
srayuth +1: Claude Opus 4.6- Switch to nvidia/cuda:12.1.1-cudnn8-runtime-ubuntu22.04 for GPU support - Pre-download BiRefNet and ISNet models in Docker build - Use --noreload to prevent Django autoreload double model loading - Fix apps.py preload condition for --noreload mode - Update hardware options table with current HF pricing Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>7bebf4c -
Fix OMP_NUM_THREADS and increase health check timeout
srayuthb917401 -
Deploy Django background removal API
srayuth8836a6d