Upload src/pipeline.py with huggingface_hub

src/pipeline.py

@@ -10,7 +10,6 @@ from pipelines.models import TextToImageRequest
 from torch import Generator
 from torchao.quantization import quantize_, int8_weight_only
 from transformers import T5EncoderModel, CLIPTextModel
-import torch_tensorrt
 
 
 Pipeline: TypeAlias = FluxPipeline
@@ -52,7 +51,7 @@ def load_pipeline() -> Pipeline:
 
     pipeline.transformer.to(memory_format=torch.channels_last)
     #    quantize_(pipeline.vae, int8_weight_only())
-    pipeline.vae = torch.compile(pipeline.vae, backend="tensorrt")
+    pipeline.vae = torch.compile(pipeline.vae)
     pipeline.to("cuda")
     
     for _ in range(2):