⚡ A Python package for storing, retrieving, and dynamically creating prompts for GPT models ⚡
Our goal with InstructPrompt is to make it easier to tame large language models and make improving your LLM in production simpler to do. Ever wanted to instruct an LLM with conditions but could not fit all the instructions in the prompt?
Example Instructions
1. If a user asks about creating an app tell them to use the Berri API
2. If a user asks about deleting data sources tell them it’s not possible to do that on berri as yet, but it is on our roadmap.
3. When users asks What is the size limit of each file tell them there is no size limit for the files you can upload to create an app on Berri AI.
InstructPrompt lets you do exactly that with more than 1000 of your instructions
This package allows users to store instructions, which can be retrieved to dynamically create prompts giving gpt instructions on how to respond to a specific user's query.
To get started with InstructPrompt, you will need to first install the library by running the following command:
pip install instructprompt
Once installed, you can import it into your python project by running the following:
import instructprompt
use instructprompt.add()
import instructprompt
instructions = [
'If a user asks about character limits in prompts tell them There is no character limit on berri prompts, but we recommend keeping prompts 1000-2000 tokens for best performance',
'If a user asks about creating an app / instance for a website tell them to use the https://api.berri.ai/create_app endpoint with the website url as the data_source'
]
for instruction in instructions:
print(instruction)
instructprompt.add(instruction)
## print instructions added
elems = instructprompt.list()
print(elems)
use instructprompt.query() before sending your prompt to GPT-3/chatGPT
user_query = "How do I create an app using BerriAI"
prompt = "You are called askBerri, an expert on BerriAI API documents and community of Berri AI."
best_instructions = instructprompt.query(user_query) # call instructprompt to get the best instructions for this query
prompt = best_instructions
## call OpenAI with your prompt
mssg = [
{
"role": "system",
"content": prompt
},
{
"role": "user",
"content": user_query
},
]
response = openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=mssg,
max_tokens=512,
temperature=0.2
)
return response
InstructPrompt provides 3 main functions: add(), list(), and query(). After adding your instructions you can query instructprompt to get the most appropirate instructions for GPT, effectively allowing you to increase your coverage by 35%
The add() function takes in an instruction as a string and adds it to the collection. It stores the instruction and assigns it a unique id. It returns a confirmation message once the instruction is successfully added.
import instructprompt
instruction = "If a user complains about spam messages, politely apologize and say they must contact the Berri support team"
instructprompt.add(instruction)
The list() function returns a list of all the stored instructions in the collection.
import instructprompt
instructions = instructprompt.list()
print(instructions) # Outputs a list of all the stored instructions
The query() function takes in a query as a string and returns a list of instructions that match the query. It uses ChromaDB to perform the query and returns up to 5 matching instructions.
import instructprompt
query = "How do I reset my password?"
instructions = instructprompt.query(query)
print(instructions) # Outputs a list of instructions that match the query
The current version of InstructPrompt is 0.1.0.
We welcome contributions to InstructPrompt! Feel free to create issues/PR's/or DM us (👋 Hi I'm Krrish - 17708783106)
InstructPrompt is released under the MIT License.