Update app.py

app.py

@@ -60,7 +60,7 @@ def generate_text(messages):
                 model_id, token=huggingface_token ,torch_dtype=dtype,device_map=device
             )
     streamer = TextStreamer(tokenizer, skip_prompt=True)    
-    text_generator = pipeline("text-generation", model=model, tokenizer=tokenizer,torch_dtype=dtype,device_map=device ,streamer=streamer) #pipeline has not to(device)
+    text_generator = pipeline("text-generation", model=model, tokenizer=tokenizer,torch_dtype=dtype,device_map=device ,streamer=streamer,model_kwargs={"stream": True} ) #pipeline has not to(device)
     result = text_generator(messages, max_new_tokens=256, do_sample=True, temperature=0.7)
     print(f"result={result}")
     generated_output = ""