Remove redundant

src/pipeline.py

@@ -10,20 +10,15 @@ from pipelines.models import TextToImageRequest
 from torch import Generator
 from torchao.quantization import quantize_, int8_weight_only
 from transformers import T5EncoderModel, CLIPTextModel, logging 
-import torch._dynamo
-# torch._dynamo.config.suppress_errors = True
+
 
 Pipeline: TypeAlias = FluxPipeline
 
 torch.backends.cudnn.benchmark = True
-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
-
 
 os.environ['PYTORCH_CUDA_ALLOC_CONF']="expandable_segments:True"
 os.environ["TOKENIZERS_PARALLELISM"] = "True"
+
 CHECKPOINT = "jokerbit/flux.1-schnell-Robert-int8wo"
 REVISION = "5ef0012f11a863e5111ec56540302a023bc8587b"
 
@@ -47,11 +42,8 @@ def load_pipeline() -> Pipeline:
         torch_dtype=torch.bfloat16,
     ).to("cuda")
     quantize_(pipeline.vae, int8_weight_only())
-    # pipeline.to(memory_format=torch.channels_last)
-    pipeline.transformer.to(memory_format=torch.channels_last)
-    # pipeline.vae.to(memory_format=torch.channels_last)
- 
-    # quantize_(pipeline.vae, int8_weight_only())
+    pipeline.to(memory_format=torch.channels_last)
+    # pipeline.transformer.to(memory_format=torch.channels_last)
     
     PROMPT = 'semiconformity, peregrination, quip, twineless, emotionless, tawa, depickle'
     with torch.inference_mode():