Spaces:
Runtime error
Runtime error
Commit History
-
Disable torch.compile for timestep_embedding function
The timesteps.device attribute causes ConstantVariable errors with torch.compile. Disabling compilation for this function prevents the error while keeping main UNet compilation benefits intact. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>6cbd0a8 -
Fix timestep_embedding for torch.compile compatibility
Revert to CPU-first tensor creation then .to(device) instead of device= parameter. The device= parameter causes torch.compile to fail with ConstantVariable assertion error for torch.device type. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>ecce6e7 -
Enable torch.compile for 1.5-1.7× speedup
- Fix timestep_embedding to create tensors on target device - Disable compilation for SAG/FreeU to allow attention capture - Clone SAG attention scores to prevent CUDAGraphs overwrite - Replace in-place += with explicit assignment in attention - Use list indexing for dynamic slicing compatibility - Change dynamic=True for variable batch size support - Remove triton_cache.tar.gz (cache approach doesn't work on ZeroGPU) First inference: 60s (compilation), subsequent: 2-10s (cached) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>576412f
-
fix to torch.compile
7062f04 -
enable torch,compile with fixes
96f8c12 -
Remove progress tracking, add essential fixes only
- Fix PyTorch indexing deprecation warning (tuple instead of list) - Suppress torchsde floating-point precision warnings - Add 1-hour automatic cache cleanup (delete_cache) - Add app.queue() for proper Gradio functionality - Remove all ProgressTracker and progress display code for simplicity00847d8
-
Upload 360 files
Verified signature762fbb3