Update app.py

app.py

@@ -95,11 +95,11 @@ headers = {"Authorization": "Bearer " + os.environ['HF_TOKEN']}
 
 def merlin_chat(message, history):
     chat_text = ""
-    #chat_json = ""
+    chat_list = []
     for turn in history:
         chat_text += f"{turn[0]}\n\n{turn[1]}\n\n"
-        #chat_json += json.dumps({"role": "user", "content": turn[0]})
-        #chat_json += json.dumps({"role": "assistant", "content": turn[1]})
+        chat_list.append({"role": "user", "content": turn[0]})
+        chat_list.append({"role": "assistant", "content": turn[1]})
     chat_text += f"{message}\n"
     doc = nlp(chat_text)
     ents_found = []
@@ -138,20 +138,17 @@ def merlin_chat(message, history):
     system_msg = {
         'role': 'system', 'content': context
     }
+    chat_list.insert(0,system_msg)
     user_msg = {'role': 'user', 'content': message}
-    prompt_data = history
-    prompt_data.append(user_msg)
-    prompt_data.insert(0,system_msg)
-    prompt = json.dumps(prompt_data)[:-1] #+ ",{\"role\": \"assistant\", \"content\": \""
+    chat_list.append(user_msg)
+    prompt = json.dumps(chat_list)[:-1]
     for attempt in range(3):
-        # result = model(prompt, max_new_tokens=250, return_full_text=False, handle_long_generation="hole")
         result = generate_text(prompt, model_path, parameters, headers)
         response = result[0]
         print(response) # so we can see it in logs
         start = 0
         end = 0
         cleanStr = response.lstrip()
-        cleanStr = cleanStr.replace(prompt,"")
         start = cleanStr.find('{')
         if start<=0:
             continue