Add a line

src/pipeline.py

@@ -54,9 +54,9 @@ def load_pipeline() -> Pipeline:
     ).to("cuda")
 
     pipeline.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")
+    pipeline.transformer = torch.compile(pipeline.transformer, mode="max-autotune", fullgraph=True)
+    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.no_grad():