QuantGPT v2 is the next iteration of the QuantGPT project, originally created by @rnikitin. This version aims to improve on the original by utilizing a more stable front-end framework, Next.js, and the newly updated OpenAI Assistants API v2.
.png)
- AI-Powered Documentation Search: Query the full breadth of VectorBT PRO documentation using natural language, powered by OpenAI Assistants v2.
- Efficient Strategy Development: Quickly find the functions, parameters, and modules you need, saving time and streamlining the process from conception to execution.
git clone git@github.com:BaileySimrell/QuantGPT-v2.git
cd QuantGPT-v2- Visit OpenAI API to create an account.
- Ensure you add billing information to avoid having API requests rejected.
3. Create a project and set your OpenAI API key
export OPENAI_API_KEY="sk_..."(or in .env.example and rename it to .env).
Create an assistant on the OpenAI platform with the following configuration:
- Description: A helpful assistant with a knowledge base of VectorBT PRO documentation.
- Capabilities: Ensure that the assistant has capabilities like file search enabled.
- Settings:
- File Search:
Enabled - Temperature:
0.24 - Top P:
1
- File Search:
Use the following instruction for the assistant:
You are a helpful assistant that has a knowledge base uploaded to you containing information on how the closed-source VectorBT (PRO) Python library and its modules work for building financial backtests and simulations.
VectorBT PRO (vectorbtpro) is a next-generation engine for backtesting, algorithmic trading, and research. It's a high-performance, actively-developed, proprietary successor to the vectorbt library, one of the world's most innovative open-source backtesting packages. The PRO version extends the open-source package with new impressive features and useful enhancements.
You are an expert at reading through the provided VectorBT (PRO) documentation and coming up with clear, accurate answers to users' queries.
You have been given a massive index to search through which contains all of the text from VBT (PRO)'s documentation. If you cannot find/retrieve the answer in your vector store, you do NOT make anything up. Respond saying that you can't find any information on that topic specifically.
Also, FYI, VectorBT (PRO) can also be referred to in this context as VBT, so if VBT is mentioned in the messages, assume the user is referring to this closed source version, NOT the open source `vectorbt`. VectorBT PRO has been completely refactored to improve performance and enable new groundbreaking features, such as parallelization support, so many things are different from how the older, open source version worked.
```shell
python -m venv venv
source venv/bin/activate
```
```shell
pip install requests
pip install beautifulsoup4
pip install html2text
pip install readability-lxml
pip install lxml_html_clean
```
Open up the notebook located at /docs-scraper/notebook.ipynb and run the cells to scrape the VectorBT (PRO) documentation. This will create a series of markdown files that can be uploaded to the assistant. Huge thanks to @rnikitin for creating all of this original scraper logic.
Go to your assistant's settings page on the OpenAI platform and upload the markdown files created by the scraper. This will allow the assistant to search through the documentation when answering user queries.
NOTE: OpenAI's front-end for uploading files is a bit finicky. As of right now, there is a known limitation where you may only upload 100 files at a time. Since the scraper creates just under 300 files, you will need to upload the files in three separate batches. I highly suggest sorting your files by name and uploading them in alphabetical order to avoid any issues and stay organized. Eventually, I will write a script to automate this process.
npm installnpm run dev11. Navigate to http://localhost:3000.
Once you are on the main page, you will be prompted to select an interface type. For now, only the basic chat interface is functioning properly. The file search interface is still a work in progress as it is lacking pagination and a few other features.
You can deploy this project to Vercel or any other platform that supports Next.js.
The main logic for chat will be found in the Chat component in app/components/chat.tsx, and the handlers starting with api/assistants/threads (found in api/assistants/threads/...). Feel free to start your own project and copy some of this logic in! The Chat component itself can be copied and used directly, provided you copy the styling from app/components/chat.module.css as well.
- Basic Chat Example: http://localhost:3000/chat
- Basic Chat with File Interface Example (WIP): http://localhost:3000/chat-with-file-search
app/components/chat.tsx- handles chat rendering, streaming, and function call forwardingapp/components/file-viewer.tsx- handles uploading, fetching, and deleting files for file search
api/assistants-POST: create assistant (only used at startup)api/assistants/threads-POST: create new threadapi/assistants/threads/[threadId]/messages-POST: send message to assistantapi/assistants/threads/[threadId]/actions-POST: inform assistant of the result of a function it decided to callapi/assistants/files-GET/POST/DELETE: fetch, upload, and delete assistant files for file search
A heartfelt thank you goes out to the individuals and teams whose work has been fundamental to the development of QuantGPT:
- @polakowo, for creating vectorbt.pro, a library that has significantly democratized quantitative trading.
- @rnikitin, for creating the original QuantGPT project, which has served as the foundation for this project.
- @openai, for the GPT models that have redefined our interaction with machine learning and data analysis.
MIT License – To view the full license, see the LICENSE file in the GitHub repository.
Copyright (c) 2024 Bailey Simrell