The issue of partial output

API
1

Hi, I’ve created a chatbot with a custom knowledge base by following the instructions on this page: How to Train an AI Chatbot With Custom Knowledge Base Using ChatGPT API | Beebom. However, I’m experiencing an issue where the output is always partial and seems to cut off after a certain length, and I’m not sure what condition is causing it. I have attached an image to show an example of this, along with the code.

Screenshot 2023-03-28 at 11.39.17 PM
Screenshot 2023-03-28 at 11.39.17 PM767×489 48.1 KB 

from gpt_index import SimpleDirectoryReader, GPTListIndex, GPTSimpleVectorIndex, LLMPredictor, PromptHelper
from langchain.chat_models import ChatOpenAI
import gradio as gr
import sys
import os

os.environ["OPENAI_API_KEY"] = '________'

def construct_index(directory_path):
    max_input_size = 4096
    num_outputs = 2000
    max_chunk_overlap = 20
    chunk_size_limit = 1024
    max_length = 5000

    prompt_helper = PromptHelper(max_input_size, num_outputs, max_chunk_overlap, chunk_size_limit=chunk_size_limit)

    llm_predictor = LLMPredictor(llm=ChatOpenAI(temperature=0.7, model_name="gpt-3.5-turbo", max_tokens=num_outputs))

    documents = SimpleDirectoryReader(directory_path).load_data()

    index = GPTSimpleVectorIndex(documents, llm_predictor=llm_predictor, prompt_helper=prompt_helper)

    index.save_to_disk('index.json')

    return index

def chatbot(input_text):
    index = GPTSimpleVectorIndex.load_from_disk('index.json')
    response = index.query(input_text, response_mode="compact")
    return response.response

iface = gr.Interface(fn=chatbot,
                     inputs=gr.inputs.Textbox(lines=7, label="Enter your text"),
                     outputs="text",
                     title="Custom-trained AI Chatbot")

index = construct_index("docs")
iface.launch(share=True)

Does anyone know how to fix this problem of partial output? Thank you very much!

1 Like
2

I’m having responses cut short also. I cannot work out how to get a complete response.

I asked ChatGPT, and was suggested using response_mode=“complete” but for me this errors out and doesn’t seem to be a valid response mode