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_json = ""
     for turn in history:
-        chat_text += f"USER: {turn[0]}\n\nASSISTANT: {turn[1]}\n\n"
-        chat_json += json.dumps({"role": "user", "content": turn[0]})
-        chat_json += json.dumps({"role": "assistant", "content": turn[1]})
+        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_text += f"USER: {message}\n"
     doc = nlp(chat_text)
     ents_found = []
@@ -135,24 +135,23 @@ def merlin_chat(message, history):
                     continue
                 else:
                     context += entsum + '\n\n'
-    context
     system_msg = {
         'role': 'system', 'content': context
     }
     user_msg = {'role': 'user', 'content': message}
-    prompt = "[" + json.dumps(system_msg) + chat_json + json.dumps(user_msg) + "{'role': 'assistant, 'content': '"
+    prompt_data = history
+    prompt_data.append(user_msg)
+    prompt_data.insert(0,system_msg)
+    prompt = json.dumps(system_msg)[:-1] + ",{'role': 'assistant', 'content': '"
     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('{') # this should skip over whatever it recalls to what it says next
-        if start<=0:
-            continue
-        cleanStr = cleanStr[start:]
         end = cleanStr.find('}') + 1
         if end<=0:
             continue