BitDB

BitDB is a lightweight skeleton database, very similar to SQLite. Stemming from my interest in learning more about both databases and memory operations in C, I created this project by watching Computer Architecture lectures and following cstack's database tutorial. After, I used ruby for testing different database scenarios and edge cases. Then, being a native mac user myself, I used Apple ML Research's brand new MLX library to implement a Mistral-7B model finetuned with lora to be a natural language to code generation assistant, named Ada. I took heavy inspiration from the llama example from the mlx repository, and adjusted the code to use mistral. (As of Dec 16 a mistral lora example has been added to the repo.) I wrote this setup guide for Mac because of the MLX usage in this project.

Contents

• Demo
• Setup
• Run
• Syntax
• Examples
• Version
• References

Demo

Setup

First step is to ensure that you can compile the C program. Navigate to terminal and run the following command which installs Xcode which contains gcc, the compiler I used for this project. Follow the steps prompted after running this command.

xcode-select --install

Next, navigate to the model directory. Follow the steps in the readme that show you how to setup the Mistral-7B model.

Run

Running this project is quite a simple process. First navigate to the project directory and run the following command to compile the C program.

gcc -o database db.c

To put it simply, hen you run the command above, the gcc compiler converts the C code in db.c into an executable file named database. It compiles the code, turns it into machine code, and links any necessary libraries to create a runnable program.

When you run the command below, you're executing the database program that was previously compiled, and passing mydatabase.db as an argument to it. The program database will then perform various database actions as defined in its code, and it uses mydatabase.db as a save file.

./database mydatabase.db

Note: You only need to compile the database the first time, after that you can run the save file using the second command and all your work will be saved.

Syntax

The database is very simple. I built most of it from scratch so there is not a whole lot of functionality, yet. There are select and insert operations, and a Natural Language Processing (NLP) assistant named Ada.

To select, just simply write:

select

Currently the database is setup to take a Name, ID, and email. To insert write:

insert [Name] [ID] [email]

To use the Ada assistant just write Ada to start the line (Case Sensitive):

Ada [Insert Natural Language Query Here]

Also there are a couple of Meta commands that can be used. First there is

.exit

which will exit the database and save your work to the save file. There is also

.btree

which will print the btree structure. This database is organized using a BTree for efficient time efficiency. Finally there is the

.constants

which will print out all the constant variables.

Examples

Here is an example of the insert and select statements:






Here is an example of the Ada assistant functionality:

Version

• 0.2

  • Added Language Model Capability
  • Fixed Duplicate keys bug
  • Fixed database saving bug

• 0.1

  • Initial Release
  • Finished database functionality
    

• Add in Update and Delete Functionality
• Finish LLM assistant
• Fix bug where the database doesn't save

References

1. Database Tutorial
2. MLX Library
3. MLX Examples
4. Mistral 7B Paper
5. LoRA Paper