Speed up the inference a little bit

handler.py

@@ -23,7 +23,8 @@ class EndpointHandler:
         ).to(device)
 
         self.pipe.load_lora_weights("SvenN/sdxl-emoji", weight_name="lora.safetensors")
-
+        self.pipe.fuse_lora()
+        
         text_encoders = [self.pipe.text_encoder, self.pipe.text_encoder_2]
         tokenizers = [self.pipe.tokenizer, self.pipe.tokenizer_2]
 
@@ -44,10 +45,11 @@ class EndpointHandler:
         inputs = data.pop("inputs", data)
 
         # Automatically add trigger tokens to the beginning of the prompt
-        full_prompt = f"A <s0><s1> {inputs}"
+        full_prompt = f"A <s0><s1> emoji {inputs}"
         images = self.pipe(
             full_prompt,
             cross_attention_kwargs={"scale": 0.8},
+            num_inference_steps=25
         ).images
         image = images[0]