Spaces:
Running on Zero
Running on Zero
Commit History
-
Try sequential CPU offload for even more memory savings Running
a6f96a4 -
Fix: Enable CPU offloading to prevent OOM errors
The PixArt-Sigma model is too large to fit entirely in GPU memory. - Model alone: ~12GB - Peak inference memory: 65GB+ Solution: Use enable_model_cpu_offload() instead of moving entire model to GPU. This keeps weights on CPU and only moves to GPU when needed.e4241ae -
Add comprehensive debug logging and reduce memory usage
- Add memory logging at every stage (before/after loading, generation) - Add step-by-step logging throughout the pipeline - Add try/except blocks with detailed error messages - Enable memory-efficient options (attention slicing, VAE slicing) - Reduce default resolution from 1024x1024 to 512x512 - Reduce default steps from 30 to 20 - Simplify examples to reduce memory usage - Disable resolution binning to avoid classify_height_width_bin issues - Add gc.collect() and torch.cuda.empty_cache() for memory recoveryd1f210d -
Fix: Override __call__ method in PixArtSigmaPipeline
The patched pipeline_pixart_alpha_call function was defined but never assigned to the pipeline class. Added __call__ assignment to use the patched call method that calls self.classify_height_width_bin() instead of self.image_processor.classify_height_width_bin().0023489 -
Fix: Add missing classify_height_width_bin and resize_and_crop_tensor methods
The VaeImageProcessor class in newer diffusers versions no longer has classify_height_width_bin method. Added these methods directly to PixArtSigmaPipeline class to fix AttributeError.097dbb8
-
Fix Gradio 6 cache_examples compatibility
b4c081e -
Remove training-only deps causing Space build failure
2c2fa1b -
Update README.md
Verified signature0cb7315 -
Update requirements.txt
Verified signature56fc061
-
Update app.py (#4)
Verified signature- Update app.py (bfe5ff96e56e985a7f87514973bb8c6ffe6ce910) Co-authored-by: John Smith <John6666@users.noreply.huggingface.co>4d30e65
-
Update README.md
Verified signaturee50dff3
-
adding examples and caching to the app (#1)
Verified signature- adding examples and caching to the app (1efee9a68b2e7c158a9b8e1b4be22206d1eb01de) Co-authored-by: yuvraj sharma <ysharma@users.noreply.huggingface.co>43c5610
-
Update app.py
Verified signature46e6575 -
Update app.py
Verified signaturef0bd898 -
Update app.py
Verified signatureb466771 -
Update app.py
Verified signatureecd6341 -
Hi
eadd7b4 -
Create app.py
Verified signaturef8cfb21 -
Create requirements.txt
Verified signatureee87c51 -
initial commit
Verified signature1a8e8a3