Update app.py

app.py

@@ -5,7 +5,7 @@ import spaces
 """
 For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
 """
-pipe = pipeline("text-generation", model="microsoft/Phi-3-mini-4k-instruct", trust_remote_code=True)
+pipe = pipeline("text-generation", model="microsoft/Phi-3-mini-4k-instruct", framework="pt", trust_remote_code=True)
 
 @spaces.GPU(duration=120)
 def respond(
@@ -28,14 +28,15 @@ def respond(
 
     response = ""
 
-    for message in pipe.text_generation(
+    for message in pipe(
         messages,
-        max_tokens=max_tokens,
-        stream=True,
+        max_length=max_tokens,
+        do_sample=True,
         temperature=temperature,
         top_p=top_p,
+        return_dict_in_generate=True,
     ):
-        token = message.choices[0].delta.content
+        token = message["generated_text"]
 
         response += token
         yield response
@@ -59,6 +60,5 @@ demo = gr.ChatInterface(
     ],
 )
 
-
 if __name__ == "__main__":
     demo.launch()