Vanilla

src/pipeline.py

@@ -47,11 +47,10 @@ def load_pipeline() -> Pipeline:
         torch_dtype=torch.bfloat16,
     ).to("cuda")
 
-    pipeline.to(memory_format=torch.channels_last)
+    pipeline.transformer.to(memory_format=torch.channels_last)
     pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=False)
     quantize_(pipeline.vae, int8_weight_only())
-    pipeline.vae = torch.compile(pipeline.vae, fullgraph=True, mode="max-autotune")
-    
+        
     PROMPT = 'semiconformity, peregrination, quip, twineless, emotionless, tawa, depickle'
     with torch.inference_mode():  
         for _ in range(4):