Image-Morpher/logs/execution_20250309_232833.log

2025-03-09 23:30:11,054 - INFO - Total execution time: 97.80 seconds
2025-03-09 23:30:11,054 - INFO - Model Path: stabilityai/stable-diffusion-2-1-base
2025-03-09 23:30:11,054 - INFO - Image Path 0: ./assets/Trump.jpg
2025-03-09 23:30:11,054 - INFO - Image Path 1: ./assets/Biden.jpg
2025-03-09 23:30:11,055 - INFO - Use LCM: True
2025-03-09 23:30:11,055 - INFO - Number of inference steps: 8
2025-03-09 23:30:11,055 - INFO - Guidance scale: 1


so this was without the 

# Configure compiler settings (safe defaults)
# inductor.config.force_fuse_int_mm_with_mul = True  # Better for diffusion models

it was 0.7 seconds off, so the real benefit (of the real cudnn benchmark OR the set_float32_matmul_precision actually!!, u didnt realise bef) was 
shifting the cudnn benchmark test AFTER pipeline.to("cuda")