Upload handler.py

handler.py

@@ -24,8 +24,8 @@ class EndpointHandler:
         dtype = torch.bfloat16
         quantization_config = TorchAoConfig("int4dq")
         vae = AutoencoderKL.from_pretrained(repo_id, subfolder="vae", torch_dtype=dtype)
-        #transformer = FluxTransformer2DModel.from_pretrained(repo_id, subfolder="transformer", torch_dtype=dtype).to("cuda")
-        self.pipeline = FluxPipeline.from_pretrained(repo_id, vae=vae, torch_dtype=dtype, quantization_config=quantization_config)
+        transformer = FluxTransformer2DModel.from_pretrained(repo_id, subfolder="transformer", torch_dtype=dtype, quantization_config=quantization_config).to("cuda")
+        self.pipeline = FluxPipeline.from_pretrained(repo_id, transformer=transformer, vae=vae, torch_dtype=dtype)
         if is_compile: self.pipeline = compile_pipeline(self.pipeline)
         self.pipeline.to("cuda")