Update README.md

README.md

@@ -90,20 +90,21 @@ class StableDiffusionTest():
         self.text_encoder, _, preprocess = open_clip.create_model_and_transforms('xlm-roberta-large-ViT-H-14', pretrained=text_encoder_path)
         self.tokenizer = open_clip.get_tokenizer('xlm-roberta-large-ViT-H-14')
         self.text_encoder.text.output_tokens = True
-        self.proj = MLP(1024, 1280, 1024,2048, use_residual=False).to(device,dtype=dtype)
-        self.text_encoder = self.text_encoder.to(device)
+        self.text_encoder = self.text_encoder.to(device,dtype=dtype)
 
         self.vae = AutoencoderKL.from_pretrained(model_id, subfolder="vae").to(device)
         scheduler = DPMSolverMultistepScheduler.from_pretrained(model_id, subfolder="scheduler")
         self.pipe = StableDiffusionXLPipeline.from_pretrained(model_id, scheduler=scheduler,torch_dtype=dtype).to(device) 
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.pipe.vae_scale_factor)
+
+        self.proj = MLP(1024, 1280, 1024,2048, use_residual=False).to(device,dtype=dtype)
         self.proj.load_state_dict(torch.load(proj_path, map_location="cpu"))
 
 
     def encode_prompt(self, prompt, device, num_images_per_prompt, do_classifier_free_guidance, negative_prompt):
         batch_size = len(prompt) if isinstance(prompt, list) else 1
 
-        text_input_ids = self.tokenizer(prompt).to(device,dtype=dtype)
+        text_input_ids = self.tokenizer(prompt).to(device)
         _,text_embeddings = self.text_encoder.encode_text(text_input_ids)
 
         add_text_embeds,text_embeddings_2048 = self.proj(text_embeddings)