Update app.py

app.py

@@ -22,6 +22,22 @@ Propose auto-completion to the text. You have several roles:
 Don't suggest anything if there are no good suggestions. 
 Make sure the suggestions are valid completions of the text! Suggest only up to 5 words ahead. The scheme of your answer should be "answer1;answer2;answer3" (return between 0 to 4 answers).
 Answers should be only the completions themselves. 
+Examples: 
+(1)
+User: "Help me write a sentiment analysis pipeline"
+Assistant: "using huggingface;using NLTK;using python"
+
+(2)
+User: "My name is"
+Assistant: "" (nothing much to contribute at this point. return nothing)
+
+(3)
+User: "Help me find a present for my"
+Assistant: "girlfriend;mother;father;friend"
+
+'''
+
+'''
 You will now get a blank message from the user and then after your answer, the user will give you the text to complete: 
 Example:
 
@@ -50,8 +66,8 @@ Assistant: "girlfriend;mother;father;friend"
 
 start_messages = [
     {'role': 'system', 'content': system_prompt}, 
-    {'role': 'user', 'content': '  '},
-    {'role': 'assistant', 'content': '<Waiting for text>'}
+    # {'role': 'user', 'content': '  '},
+    # {'role': 'assistant', 'content': '<Waiting for text>'}
 ]
 
 
@@ -88,7 +104,9 @@ def generate(text, past_key_values):
         *start_messages,
         {'role': 'user', 'content': text}
     ]
-    past_key_values = past_kv_to_device(past_key_values, pipe.model.device, pipe.model.dtype)
+    
+    if past_key_values:
+        past_key_values = past_kv_to_device(past_key_values, pipe.model.device, pipe.model.dtype)
     response = pipe(messages, 
                     past_key_values=past_key_values, 
                     **generate_kwargs)[0]['generated_text']
@@ -97,8 +115,8 @@ def generate(text, past_key_values):
     
 if __name__ == "__main__":
     with torch.no_grad():
-        past_key_values = set_past_key_values()
-        print(f'{past_key_values=}')
-        demo = gr.Interface(partial(generate, past_key_values=past_key_values), 
+        # past_key_values = set_past_key_values()
+        # print(f'{past_key_values=}')
+        demo = gr.Interface(partial(generate, past_key_values=None), 
                             inputs="textbox", outputs="textbox")
         demo.launch()