How to get a full response from chatgpt in a telegram bot written in python?

[dense flax]

Good day, gentlemen and ladies! I need help with a telegram chat bot written in python. The problem is that a cropped response comes to my request sent via telegram. This is due to the restriction of tokens on the response. I want the response to continue as if I had asked to continue chatgpt in the laptop or web version. How can i do this? Is there any other way to get a complete answer?

[gleaming wren]

Can you share your code for the query/request

[dense flax]

sure, check this

@bot.message_handler(content_types=['text'])
def handle_all_messages(message):
response = openai.Completion.create(
model="text-davinci-003",
prompt=message.text,
temperature=0.5,
max_tokens=150,
top_p=1.0,
frequency_penalty=0.5,
presence_penalty=0.0,
)
bot.send_message(chat_id=message.from_user.id, text=response['choices'][0]['text'])

bot.polling(non_stop=True)

on the whole, this is a piece of code taken from playground

[gleaming wren]

Quick note, you can encapsulate code using backticks. Try it next time you send code using three (`) followed directly by the language, in this case, py. Such as this:

@bot.message_handler(content_types=['text'])
def handle_all_messages(message):
    response = openai.Completion.create(
    model="text-davinci-003",
    prompt=message.text,
    temperature=0.5,
    max_tokens=150,
    top_p=1.0,
    frequency_penalty=0.5,
    presence_penalty=0.0,
    )
    bot.send_message(chat_id=message.from_user.id, text=response['choices'][0]['text'])


bot.polling(non_stop=True)```

Your max tokens are only 150, which is going to cut off a LOT. Try increasing it to about 400. This will increase the cost but it won't cut off as much.

My own bot is at 800.

[dense flax]

yes, I experimented with the value of the tokens. And suppose I do not have enough maximum value?

[gleaming wren]

Correct. 150 tokens is only about 120ish words in English. Once you get past that, it will cut off.

The only way to keep that would be to feed previous responses back to the API

You could theoretically feed 3850 tokens worth of previous chats to Davinci-003.

This is likely how ChatGPT is able to "recall" conversations to a certain extent.