Update app.py

app.py

@@ -12,18 +12,25 @@ def correct_words(text):
     corrected_words = [spell.correction(word) for word in words]
     corrected_text = ' '.join(corrected_words)
     return corrected_text
+import asyncio
+
+# ... (previous code)
+
+async def generate_async(tokens):
+    for token in tokens:
+        yield token
 
 def generate_response(message):
     global history
-    for _ in range(2):  # Assuming you want two responses, adjust as needed
-        tokens = [ord(char) for char in message]  # Convert characters to ASCII values
-        response = llm.generate(tokens=tokens, top_k=50, top_p=0.95, temperature=1.0, repetition_penalty=1.0, last_n_tokens=1,stream=True)
-        response_text = ' '.join(map(str, response))
+    for _ in range(2):
+        tokens = [ord(char) for char in message]
+        response = llm.generate_async(tokens=tokens, top_k=50, top_p=0.95, temperature=1.0, repetition_penalty=1.0, last_n_tokens=1)
+        response_text = ' '.join(map(str, await asyncio.gather(*response)))
         time.sleep(2)
         corrected_response = correct_words(response_text)
         history.append(corrected_response)
         yield ' '.join(history)
-    # Clear the history list after the last response
+    history = ["Chatbot:"]
     history = ["Chatbot:"]
     # Clear the history list after the last response
     history = ["Chatbot:"]