Update src/pipeline.py

src/pipeline.py

@@ -131,7 +131,8 @@ def load_pipeline() -> Pipeline:
     text_encoder_2 = T5EncoderModel.from_pretrained(
         "city96/t5-v1_1-xxl-encoder-bf16", torch_dtype=torch.bfloat16
     )
-    vae=AutoencoderKL.from_pretrained(ckpt_id, subfolder="vae", torch_dtype=dtype)
+    # vae=AutoencoderKL.from_pretrained(ckpt_id, subfolder="vae", torch_dtype=dtype)
+    vae = torch.load('/root/.cache/huggingface/hub/compiled_vae.pth')
     # transformer = FluxTransformer2DModel.from_pretrined("manbeast3b/transfomer-flux-schnell-int8") # torch_dtype=dtype
     pipeline = DiffusionPipeline.from_pretrained(
         ckpt_id,
@@ -152,11 +153,11 @@ def load_pipeline() -> Pipeline:
     # pipeline.transformer.save_pretrained("/root/.cache/huggingface/hub/transformer-flux")
     # exit()    
 
-    pipeline.vae.to(memory_format=torch.channels_last)
-    pipeline.vae = torch.compile(pipeline.vae)
-    torch.save(pipeline.vae, '/root/.cache/huggingface/hub/compiled_vae.pth')
-    exit()
-    pipeline.vae = torch.load('/root/.cache/huggingface/hub/compiled_vae.pth')
+    # pipeline.vae.to(memory_format=torch.channels_last)
+    # pipeline.vae = torch.compile(pipeline.vae)
+    # torch.save(pipeline.vae, '/root/.cache/huggingface/hub/compiled_vae.pth')
+    # exit()
+    
     
     pipeline._exclude_from_cpu_offload = ["vae"]
     pipeline.enable_sequential_cpu_offload()