Spaces:
Sleeping
Sleeping
Commit History
-
Deploy from GitHub - 2026-01-21 11:50:26
github-actions[bot]1dda790 -
Deploy from GitHub - 2026-01-21 10:19:27
github-actions[bot]5de41f0 -
Deploy from GitHub - 2026-01-21 09:52:44
github-actions[bot]37b20e3 -
Deploy from GitHub - 2026-01-21 09:09:33
github-actions[bot]d9cf8c2
-
Deploy from GitHub - 2026-01-19 04:39:51
github-actions[bot]37a2a42 -
Deploy from GitHub - 2026-01-19 04:19:46
github-actions[bot]20cfecf -
info endpoint
Olivia1282ba1
-
info endpoint
Olivia4aca758 -
Add CUDA kernels and backend comparison
OliviaFeatures: - Add custom CUDA kernels (FusedInstanceNorm) - Add backend selection (Auto/CUDA/PyTorch) - Add performance comparison tab with benchmarking - Add interactive backend speedup display - Add CUDA availability badge in header - Add per-backend performance tracking - Add auto-fallback to PyTorch when CUDA unavailable Kernels: - instance_norm.cu - Fused InstanceNorm kernel - cuda_build.py - JIT compilation utilities - instance_norm_wrapper.py - Python wrapper with fallback - kernels/__init__.py - Package initialization The app now: - Detects CUDA availability at startup - Uses custom kernels when available (GPU) - Falls back to PyTorch on CPU or compilation failure - Shows real-time backend comparison in stats - Has dedicated Performance tab for benchmarks3386f25