Skip to content

๐Ÿค– State-of-the-art, production ready LLM apps made mega-easy, so you don't have to build them from scratch ๐Ÿคฏ Create a bot, now ๐Ÿซต

License

Notifications You must be signed in to change notification settings

momegas/megabots

Folders and files

NameName
Last commit message
Last commit date

Latest commit

ย 

History

84 Commits
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 
ย 

Repository files navigation

๐Ÿค– Megabots

Tests Python Version Code style: black License

๐Ÿค– Megabots provides State-of-the-art, production ready LLM apps made mega-easy, so you don't have to build them from scratch ๐Ÿคฏ Create a bot, now ๐Ÿซต

The Megabots library can be used to create bots that:

  • โŒš๏ธ are production ready, in minutes
  • ๐Ÿ—‚๏ธ can answer questions over documents
  • ๐Ÿ’พ can connect to vector databases
  • ๐ŸŽ–๏ธ automatically expose the bot as a rebust API using FastAPI (early release)
  • ๐Ÿ“ automatically expose the bot as a UI using Gradio

๐Ÿค– Megabots is backed by some of the most famous tools for productionalising AI. It uses LangChain for managing LLM chains, langchain-serve to create a production ready API, Gradio to create a UI. At the moment it uses OpenAI to generate answers, but we plan to support other LLMs in the future.

Getting started

Note: This is a work in progress. The API might change.

pip install megabots
from megabots import bot
import os

os.environ["OPENAI_API_KEY"] = "my key"

# Create a bot ๐Ÿ‘‰ with one line of code. Automatically loads your data from ./index or index.pkl.
# Keep in mind that you need to have one or another.
qnabot = bot("qna-over-docs")

# Ask a question
answer = qnabot.ask("How do I use this bot?")

# Save the index to save costs (GPT is used to create the index)
qnabot.save_index("index.pkl")

# Load the index from a previous run
qnabot = bot("qna-over-docs", index="./index.pkl")

# Or create the index from a directory of documents
qnabot = bot("qna-over-docs", index="./index")

# Change the model
qnabot = bot("qna-over-docs", model="text-davinci-003")

Changing the bot's prompt

You can change the bots promnpt to customize it to your needs. In the qna-over-docs type of bot you will need to pass 2 variables for the context (knwoledge searched from the index) and the question (the human question).

from megabots import bot

prompt = """
Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say that you don't know, don't try to make up an answer.
Answer in the style of Tony Stark.

{context}

Question: {question}
Helpful humorous answer:"""

qnabot = bot("qna-over-docs", index="./index.pkl", prompt=prompt)

qnabot.ask("what was the first roster of the avengers?")

Working with memory

You can easily add memory to your bot using the memory parameter. It accepts a string with the type of the memory to be used. This defaults to some sane dafaults. Should you need more configuration, you can use the memory function and pass the type of memory and the configuration you need.

from megabots import bot

qnabot = bot("qna-over-docs", index="./index.pkl", memory="conversation-buffer")

print(qnabot.ask("who is iron man?"))
print(qnabot.ask("was he in the first roster?"))
# Bot should understand who "he" refers to.

Or using the memoryfactory function

from megabots import bot, memory

mem("conversation-buffer-window", k=5)

qnabot = bot("qna-over-docs", index="./index.pkl", memory=mem)

print(qnabot.ask("who is iron man?"))
print(qnabot.ask("was he in the first roster?"))

NOTE: For the qna-over-docs bot, when using memory and passing your custom prompt, it is important to remember to pass one more variable to your custom prompt to facilitate for chat history. The variable name is history.

from megabots import bot

prompt = """
Use the following pieces of context to answer the question at the end.
If you don't know the answer, just say that you don't know, don't try to make up an answer.

{context}

{history}
Human: {question}
AI:"""

qnabot = bot("qna-over-docs", prompt=prompt, index="./index.pkl", memory="conversation-buffer")

print(qnabot.ask("who is iron man?"))
print(qnabot.ask("was he in the first roster?"))

Using Megabots with Milvus (more DBs comming soon)

Megabots bot can also use Milvus as a backend for its search engine. You can find an example of how to do it below.

In order to run Milvus you need to follow this guide to download a docker compose file and run it. The command is:

wget https://raw.githubusercontent.com/milvus-io/pymilvus/v2.2.7/examples/hello_milvus.py

You can then install Attu as a management tool for Milvus

from megabots import bot

# Attach a vectorstore by passing the name of the database. Default port for milvus is 19530 and default host is localhost
# Point it to your files directory so that it can index the files and add them to the vectorstore
bot = bot("qna-over-docs", index="./examples/files/", vectorstore="milvus")

bot.ask("what was the first roster of the avengers?")

Or use the vectorstore factory function for more customisation

from megabots import bot, vectorstore

milvus = vectorstore("milvus", host="localhost", port=19530)

bot = bot("qna-over-docs", index="./examples/files/", vectorstore=milvus)

Exposing an API with langchain-serve

You can also expose the bot endpoints locally using langchain-serve. A sample file api.py is provided in the megabots folder.

To expose the API locally, you can do

lc-serve deploy local megabots.api

You should then be able to visit http://localhost:8000/docs to see & interact with the API documentation.

To deploy your API to the cloud, you can do and connect to the API using the endpoint provided in the output.

lc-serve deploy jcloud megabots.api
Show command output
โ•ญโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ฌโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ•ฎ
โ”‚ App ID       โ”‚                                 langchain-dec14439a6                                 โ”‚
โ”œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ผโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ค
โ”‚ Phase        โ”‚                                       Serving                                        โ”‚
โ”œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ผโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ค
โ”‚ Endpoint     โ”‚                      https://langchain-dec14439a6.wolf.jina.ai                       โ”‚
โ”œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ผโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ค
โ”‚ App logs     โ”‚                               dashboards.wolf.jina.ai                                โ”‚
โ”œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ผโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ค
โ”‚ Swagger UI   โ”‚                    https://langchain-dec14439a6.wolf.jina.ai/docs                    โ”‚
โ”œโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ผโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ค
โ”‚ OpenAPI JSON โ”‚                https://langchain-dec14439a6.wolf.jina.ai/openapi.json                โ”‚
โ•ฐโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”ดโ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ”€โ•ฏ

You can read more about langchain-serve here.

Exposing a Gradio chat-like interface

You can expose a gradio UI for the bot using create_interface function. Assuming your file is called ui.py run gradio qnabot/ui.py to run the UI locally. You should then be able to visit http://127.0.0.1:7860 to see the API documentation.

from megabots import bot, create_interface

demo = create_interface(bot("qna-over-docs"))

Customising bot

The bot function should serve as the starting point for creating and customising your bot. Below is a list of the available arguments in bot.

Argument Description
task The type of bot to create. Available options: qna-over-docs. More comming soon
index Specifies the index to use for the bot. It can either be a saved index file (e.g., index.pkl) or a directory of documents (e.g., ./index). In the case of the directory the index will be automatically created. If no index is specified bot will look for index.pkl or ./index
model The name of the model to use for the bot. You can specify a different model by providing its name, like "text-davinci-003". Supported models: gpt-3.5-turbo (default),text-davinci-003 More comming soon.
prompt A string template for the prompt, which defines the format of the question and context passed to the model. The template should include placeholder variables like so: context, {question} and in the case of using memory history.
memory The type of memory to be used by the bot. Can be a string with the type of the memory or you can use memory factory function. Supported memories: conversation-buffer, conversation-buffer-window
vectorstore The vectorstore to be used for the index. Can be a string with the name of the databse or you can use vectorstore factory function. Supported DBs: milvus.

| sources | When sources is True the bot will also include sources in the response. A known issue exists, where if you pass a custom prompt with sources the code breaks. |

How QnA bot works

Large language models (LLMs) are powerful, but they can't answer questions about documents they haven't seen. If you want to use an LLM to answer questions about documents it was not trained on, you have to give it information about those documents. To solve this, we use "retrieval augmented generation."

In simple terms, when you have a question, you first search for relevant documents. Then, you give the documents and the question to the language model to generate an answer. To make this work, you need your documents in a searchable format (an index). This process involves two main steps: (1) preparing your documents for easy querying, and (2) using the retrieval augmented generation method.

qna-over-docs uses FAISS to create an index of documents and GPT to generate answers.

sequenceDiagram
    actor User
    participant API
    participant LLM
    participant Vectorstore
    participant IngestionEngine
    participant DataLake
    autonumber

    Note over API, DataLake: Ingestion phase
    loop Every X time
    IngestionEngine ->> DataLake: Load documents
    DataLake -->> IngestionEngine: Return data
    IngestionEngine -->> IngestionEngine: Split documents and Create embeddings
    IngestionEngine ->> Vectorstore: Store documents and embeddings
    end

    Note over API, DataLake: Generation phase

    User ->> API: Receive user question
    API ->> Vectorstore: Lookup documents in the index relevant to the question
    API ->> API: Construct a prompt from the question and any relevant documents
    API ->> LLM: Pass the prompt to the model
    LLM -->> API: Get response from model
    API -->> User: Return response

Loading

How to contribute?

We welcome any suggestions, problem reports, and contributions! For any changes you would like to make to this project, we invite you to submit an issue.

For more information, see CONTRIBUTING instructions.