Extra config

src/pipeline.py

@@ -13,6 +13,10 @@ from transformers import T5EncoderModel, CLIPTextModel
 
 
 Pipeline: TypeAlias = FluxPipeline
+torch._inductor.config.conv_1x1_as_mm = True
+torch._inductor.config.coordinate_descent_tuning = True
+torch._inductor.config.epilogue_fusion = False
+torch._inductor.config.coordinate_descent_check_all_directions = True
 torch.backends.cudnn.benchmark = True
 
 CHECKPOINT = "jokerbit/flux.1-schnell-Robert-int8wo"
@@ -38,7 +42,7 @@ def load_pipeline() -> Pipeline:
         torch_dtype=torch.bfloat16,
     )
 
-    pipeline.to(memory_format=torch.channels_last)
+    pipeline.transformer.to(memory_format=torch.channels_last)
     pipeline.to("cuda")
     quantize_(pipeline.vae, int8_weight_only())
     for _ in range(4):