Fix: ensure_on_device before each inference to prevent text_encoder drift d95102e verified samwell commited on 17 days ago
Explicitly move all pipeline components to CUDA, use inference_mode 850ecac verified samwell commited on 17 days ago
Remove generator entirely to fix CUDA/CPU device mismatch 0c995b9 verified samwell commited on 17 days ago
Fix generator - remove device spec, use official example pattern 0e4048f verified samwell commited on 17 days ago
Fix CUDA device mismatch - resize image and add autocast 2b2bf3c verified samwell commited on 17 days ago
Use public pytorch image, install cosmos-guardrail with --no-deps 1c3a936 verified samwell commited on 18 days ago