Advanced compile

src/pipeline.py

@@ -11,6 +11,10 @@ from torch import Generator
 from torchao.quantization import quantize_, int8_weight_only
 from transformers import T5EncoderModel, CLIPTextModel
 
+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
 
 Pipeline: TypeAlias = FluxPipeline
 os.environ['PYTORCH_CUDA_ALLOC_CONF']="expandable_segments:True"