promptengineers 0.0.7

A collection of utilities by Prompt Engineers

Prompt Engineers AI Open Source Package

Build and Publish

## Build Package
bash scripts/build.sh

## Publish Package to PyPi
bash scripts/publish.sh

Development

## In the application directory start your virtual env (this would be the workspace
## where your API server that you would like to install the model)
source .venv/bin/activate

## Then change directory to where your package is, make changes and run the following.
pip install .

## Switch back to the directory of your where your workspace is for you app server.
cd <path>/<app>/<server>
pip install -r requirements.txt

## Make sure your app server has the packages shown in setup.py and run your server...

How to use...

Load Documents from Sources

import os
from promptengineers.retrieval.factories import EmbeddingFactory, LoaderFactory, RetrievalFactory
from promptengineers.retrieval.strategies import VectorstoreContext

OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
PINECONE_KEY = os.environ.get('PINECONE_API_KEY')
PINECONE_ENV = os.environ.get('PINECONE_ENV')
PINECONE_INDEX = os.environ.get('PINECONE_INDEX')

INDEX_PROVIDER = 'pinecone'
INDEX_NAME = 'default'
NAMESPACE = '000000000000000000000000::formio'
EMBEDDING_LLM = 'text-embedding-ada-002'
LOADERS = [
	{
		"type": "copy",
	  	"text": "In a quaint village, an elderly man discovered a mysterious, "
				"ancient book in his attic. Each night, he read a chapter, and the "
				"next day, miracles began happening in the village. Unbeknownst to "
				"the villagers, their fortunes were being written by the man's dreams, "
				"guided by the magical book."
	},
	{
		"type": "web_base",
	  	"urls": ['https://adaptive.biz']
	},
	{
		"type": "website",
	  	"urls": ['https://adaptive.biz'] # Retrieves subpages
	},
	{
		"type": "youtube",
	  	"urls": ['https://www.youtube.com/watch?v=GyllRd2E6fg']
	}
]

# Generate Embeddings
embedding = EmbeddingFactory(EMBEDDING_LLM, OPENAI_API_KEY)

# Choose the appropriate vector search provider strategy for Pinecone
retrieval_provider = RetrievalFactory(
	INDEX_PROVIDER, 
	embedding.create_embedding(), 
	{
		'api_key': PINECONE_KEY, 
		'env': PINECONE_ENV, 
		'index_name': INDEX_NAME, 
		'namespace': NAMESPACE
	}
)

# Create loaders and process each
all_loaders = []
for loader_config in LOADERS:
	try:
		loader = LoaderFactory.create(loader_type=loader_config['type'], loader_config=loader_config)
		all_loaders.append(loader)
	except ValueError as e:
		print(f"Error creating loader: {e}")
	except Exception as e:
		print(f"Unexpected error: {e}")

# Assuming PineconeService and EmbeddingFactory have internal error handling
try:
	# Create a vector store service context
	vectostore_service = VectorstoreContext(retrieval_provider.create_strategy())
	vectorstore = vectostore_service.add(all_loaders)
except Exception as e:
	print(f"Error processing documents with Pinecone: {e}")

Retrieval Augemented Generation (RAG) - HTTP Chat

import os
import asyncio

from promptengineers.chat import langchain_http_retrieval_chat
from promptengineers.retrieval.factories import EmbeddingFactory, RetrievalFactory
from promptengineers.retrieval.strategies import VectorstoreContext

# Environment Variables
OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
PINECONE_KEY = os.environ.get('PINECONE_API_KEY')
PINECONE_ENV = os.environ.get('PINECONE_ENV')
PINECONE_INDEX = os.environ.get('PINECONE_INDEX')

# Retrieval Constants
INDEX_PROVIDER = 'pinecone'
INDEX_NAME = 'default'
NAMESPACE = '000000000000000000000000::formio'
EMBEDDING_LLM = 'text-embedding-ada-002'

# Chat Constants
CHAT_LLM = 'gpt-3.5-turbo'
TEMPERATURE = 0.9
MESSAGES = [
    {
        'role': 'system', 
        'content': 'You are a helpful document retrieval AI, '
                    'use the context to answer the user queries.'
    },
    {
        'role': 'user', 
        'content': 'Can you summarize the context?'
    }
]

# Generate Embeddings
embedding = EmbeddingFactory(EMBEDDING_LLM, OPENAI_API_KEY)

# Choose the appropriate vector search provider strategy for Pinecone
retrieval_provider = RetrievalFactory(
    INDEX_PROVIDER, 
    embedding.create_embedding(), 
    {
        'api_key': PINECONE_KEY, 
        'env': PINECONE_ENV, 
        'index_name': INDEX_NAME, 
        'namespace': NAMESPACE
    }
)

# Create a vector store service context
vectostore_service = VectorstoreContext(retrieval_provider.create_strategy())

# Load the vectorstore using the service context
vectorstore = vectostore_service.load()

# Run the chat
async def main():
    response, cb = await langchain_http_retrieval_chat(
        messages=MESSAGES,         
        model=CHAT_LLM, 
        temperature=TEMPERATURE, 
        vectorstore=vectorstore,
        openai_api_key=OPENAI_API_KEY
    )
    print(response)
    print(cb)

asyncio.run(main())

Retrieval Augemented Generation (RAG) - Stream Chat

import os
import asyncio

from promptengineers.chat import langchain_stream_vectorstore_chat
from promptengineers.retrieval.factories import EmbeddingFactory, RetrievalFactory
from promptengineers.retrieval.strategies import VectorstoreContext

# Environment Variables
OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
PINECONE_KEY = os.environ.get('PINECONE_API_KEY')
PINECONE_ENV = os.environ.get('PINECONE_ENV')
PINECONE_INDEX = os.environ.get('PINECONE_INDEX')

# Retrieval Constants
INDEX_PROVIDER = 'pinecone'
INDEX_NAME = 'default'
NAMESPACE = '000000000000000000000000::formio'
EMBEDDING_LLM = 'text-embedding-ada-002'

# Chat Constants
CHAT_LLM = 'gpt-3.5-turbo'
TEMPERATURE = 0.9
MESSAGES = [
    {
        'role': 'system', 
        'content': 'You are a helpful document retrieval AI, '
                    'use the context to answer the user queries.'
    },
    {
        'role': 'user', 
        'content': 'Can you summarize the context?'
    }
]

# Generate Embeddings
embedding = EmbeddingFactory(EMBEDDING_LLM, OPENAI_API_KEY)

# Choose the appropriate vector search provider strategy for Pinecone
retrieval_provider = RetrievalFactory(
    INDEX_PROVIDER, 
    embedding.create_embedding(), 
    {
        'api_key': PINECONE_KEY, 
        'env': PINECONE_ENV, 
        'index_name': INDEX_NAME, 
        'namespace': NAMESPACE
    }
)

# Create a vector store service context
vectostore_service = VectorstoreContext(retrieval_provider.create_strategy())

# Load the vectorstore using the service context
vectorstore = vectostore_service.load()

# Run the chat
async def main():
    response = langchain_stream_vectorstore_chat(
        messages=MESSAGES,         
        model=CHAT_LLM, 
        temperature=TEMPERATURE, 
        vectorstore=vectorstore,
        openai_api_key=OPENAI_API_KEY
    )
    async for data in response:
        print(data)

asyncio.run(main())

Agent Chat equipped w/ tools - HTTP Chat

import os
import asyncio

from promptengineers.chat import langchain_http_agent_chat
from promptengineers.core.config.tools import AVAILABLE_TOOLS
from promptengineers.llms.utils import gather_tools
from promptengineers.retrieval.factories import EmbeddingFactory, RetrievalFactory
from promptengineers.retrieval.strategies import VectorstoreContext

# Environment Variables
OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
PINECONE_KEY = os.environ.get('PINECONE_API_KEY')
PINECONE_ENV = os.environ.get('PINECONE_ENV')
PINECONE_INDEX = os.environ.get('PINECONE_INDEX')

# Retrieval Constants
INDEX_PROVIDER = 'pinecone'
INDEX_NAME = 'default'
NAMESPACE = '000000000000000000000000::formio'
EMBEDDING_LLM = 'text-embedding-ada-002'

# Chat Constants
CHAT_LLM = 'gpt-3.5-turbo-1106'
TEMPERATURE = 0.1
MESSAGES = [
    {
        'role': 'system', 
        'content': 'You are a powerful AI assistant, you are equipped with tools '
                    'to help you accomplish your tasks. Query the context when you need '
                    'additional information to complete your task. If the user query is not '
                    'related to the context then you can use the tools complete the task.'
    },
    {
        'role': 'user', 
        'content': 'What is 14125 compounded annually for 5 years at 4 percent for 23 years?' # Math Agent
        # 'content': 'Can you provide a react code sample to render a form in Form.io?'           # Retrieval Agent
    }
]

# Generate Embeddings
embedding = EmbeddingFactory(EMBEDDING_LLM, OPENAI_API_KEY)

# Choose the appropriate vector search provider strategy for Pinecone
retrieval_provider = RetrievalFactory(
    INDEX_PROVIDER, 
    embedding.create_embedding(), 
    {
        'api_key': PINECONE_KEY, 
        'env': PINECONE_ENV, 
        'index_name': INDEX_NAME, 
        'namespace': NAMESPACE
    }
)

# Create a vector store service context
vectostore_service = VectorstoreContext(retrieval_provider.create_strategy())

# Load the vectorstore using the service context
vectorstore = vectostore_service.load()

# Gather the tools
tools = gather_tools(tools=['math_tool'],
                    available_tools=AVAILABLE_TOOLS,
                    vectorstore=vectorstore,
                    plugins=[])

# Run the chat
async def main():
    response, cb = await langchain_http_agent_chat(
        messages=MESSAGES,         
        model=CHAT_LLM,
        tools=tools,
        temperature=TEMPERATURE, 
        openai_api_key=OPENAI_API_KEY
    )
    print(response)
    print(cb)

asyncio.run(main())

Agent Chat equipped w/ tools - Stream Chat

import os
import asyncio

from promptengineers.chat import langchain_stream_agent_chat
from promptengineers.core.config.tools import AVAILABLE_TOOLS
from promptengineers.llms.utils import gather_tools
from promptengineers.retrieval.factories import EmbeddingFactory, RetrievalFactory
from promptengineers.retrieval.strategies import VectorstoreContext

# Environment Variables
OPENAI_API_KEY = os.environ.get('OPENAI_API_KEY')
PINECONE_KEY = os.environ.get('PINECONE_API_KEY')
PINECONE_ENV = os.environ.get('PINECONE_ENV')
PINECONE_INDEX = os.environ.get('PINECONE_INDEX')

# Retrieval Constants
INDEX_PROVIDER = 'pinecone'
INDEX_NAME = 'default'
NAMESPACE = '000000000000000000000000::formio'
EMBEDDING_LLM = 'text-embedding-ada-002'

# Chat Constants
CHAT_LLM = 'gpt-4-0125-preview'
TEMPERATURE = 0.1
MESSAGES = [
    {
        'role': 'system', 
        'content': 'You are a powerful AI assistant, you are equipped with tools '
                    'to help you accomplish your tasks. Query the context when you need '
                    'additional information to complete your task. If the user query is not '
                    'related to the context then you can use the tools complete the task.'
    },
    {
        'role': 'user', 
        # 'content': 'What is 14125 compounded annually for 5 years at 4 percent for 23 years?' # Math Agent
        'content': 'Can you provide a react code sample to render a form in Form.io?'           # Retrieval Agent
    }
]

# Generate Embeddings
embedding = EmbeddingFactory(EMBEDDING_LLM, OPENAI_API_KEY)

# Choose the appropriate vector search provider strategy for Pinecone
retrieval_provider = RetrievalFactory(
    INDEX_PROVIDER, 
    embedding.create_embedding(), 
    {
        'api_key': PINECONE_KEY, 
        'env': PINECONE_ENV, 
        'index_name': INDEX_NAME, 
        'namespace': NAMESPACE
    }
)

# Create a vector store service context
vectostore_service = VectorstoreContext(retrieval_provider.create_strategy())

# Load the vectorstore using the service context
vectorstore = vectostore_service.load()

# Gather the tools
tools = gather_tools(tools=['math_tool'],
                    available_tools=AVAILABLE_TOOLS,
                    vectorstore=vectorstore,
                    plugins=[])

# Run the chat
async def main():
    response = langchain_stream_agent_chat(
        messages=MESSAGES,         
        model=CHAT_LLM,
        tools=tools,
        temperature=TEMPERATURE, 
        openai_api_key=OPENAI_API_KEY
    )
    async for data in response:
        print(data)

asyncio.run(main())