# Ensuring Llamaindex to reply with a predetermined answer with CSV file?

**URL:** <https://discuss.huggingface.co/t/ensuring-llamaindex-to-reply-with-a-predetermined-answer-with-csv-file/39792>\
**Category:** Beginners\
**Created:** [May 14, 2023, 4:35am UTC](https://discuss.huggingface.co/t/ensuring-llamaindex-to-reply-with-a-predetermined-answer-with-csv-file/39792 "2023-05-14T04:35:01Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![kamiyan89](https://avatars.discourse-cdn.com/v4/letter/k/d78d45/32.png) [@kamiyan89](https://discuss.huggingface.co/u/kamiyan89)\
**Post date:** [May 14, 2023, 4:35am UTC](https://discuss.huggingface.co/t/ensuring-llamaindex-to-reply-with-a-predetermined-answer-with-csv-file/39792/1 "2023-05-14T04:35:01Z")

</div>

Hi Everyone.

I am currently following this tutorial “[How to Train an AI Chatbot With Custom Knowledge Base Using ChatGPT API | Beebom](https://beebom.com/how-train-ai-chatbot-custom-knowledge-base-chatgpt-api/)”

I have dome some tests and it works really well with PDFs.

Now, I want to create a Q&A chatbot.  
I Created a CSV file with the common questions, and the answers to these questions

But when I train the model with the CSV file I created, and ask for example question 1, it replies with answer 2.  
 ![image](https://us1.discourse-cdn.com/hellohellohello/original/3X/b/c/bc780bb55feccba3603fae38381c230423bbb996.png)

Is there a way to ensure that Questions are tied to their correspondent answers ? so the model doesn’t mix them together?

My code is as follows ( I think this is more of a conceptual question than a code question but just in case)

````auto
import os
from llama_index import SimpleDirectoryReader, GPTListIndex, GPTVectorStoreIndex, LLMPredictor, PromptHelper, ServiceContext, StorageContext, load_index_from_storage
#from langchain.chat_models import ChatOpenAI
from langchain import OpenAI
import gradio as gr
import random
import time
import sys

os.environ["OPENAI_API_KEY"] = 'my key'
messages = [{"role": "system", "content": "follow the three instructions below for your outputs:"},
{"role": "system", "content": "1. make sure all expressions are compatible with Japanese"},
{"role": "system", "content": "2. replying in English is extrictly forbidden"},
{"role": "system", "content": "3. use Japanese only for outputs"}
]

    
def load_index():
    max_input_size = 4096
    max_chunk_overlap = 20
    chunk_size_limit = 600
    num_outputs = 768
    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)
    llm_predictor = LLMPredictor(llm=OpenAI(temperature=0.4, model_name="text-davinci-003", max_tokens=num_outputs))
    global index
    service_context = ServiceContext.from_defaults(llm_predictor=llm_predictor, prompt_helper=prompt_helper)
    index = load_index_from_storage('index.json',service_context=service_context)
    return ("indexing finished")

def chat(chat_history, user_input):
  bot_response = index.query(user_input,response_mode="compact")
  print("Q:",user_input)
  response = ""
  for letter in ''.join(bot_response.response):
      response += letter + ""
      yield chat_history + [(user_input, response)]
  print("A:",response)

with gr.Blocks() as demo:
    gr.Markdown('AI chat(β 0.1)')
    load_index()
    with gr.Tab("chatbot"):
          chatbot = gr.Chatbot()
          message = gr.Textbox ()
          message.submit(chat, [chatbot, message], chatbot)

demo.queue(max_size=100).launch(share= True)```
````

---

<div class="post-metadata">

**Author:** ![kamiyan89](https://avatars.discourse-cdn.com/v4/letter/k/d78d45/32.png) [@kamiyan89](https://discuss.huggingface.co/u/kamiyan89)\
**Post date:** [May 16, 2023, 10:25am UTC](https://discuss.huggingface.co/t/ensuring-llamaindex-to-reply-with-a-predetermined-answer-with-csv-file/39792/2 "2023-05-16T10:25:04Z")

</div>

Gentle bump

---

<div class="post-metadata">

**Author:** ![kamiyan89](https://avatars.discourse-cdn.com/v4/letter/k/d78d45/32.png) [@kamiyan89](https://discuss.huggingface.co/u/kamiyan89)\
**Post date:** [May 18, 2023, 3:48pm UTC](https://discuss.huggingface.co/t/ensuring-llamaindex-to-reply-with-a-predetermined-answer-with-csv-file/39792/3 "2023-05-18T15:48:58Z")

</div>

Answering my own question, by changing from vector index to panda index it works now  
[https://gpt-index.readthedocs.io/en/latest/examples/index\_structs/struct\_indices/PandasIndexDemo.html](https://gpt-index.readthedocs.io/en/latest/examples/index_structs/struct_indices/PandasIndexDemo.html)