Llama 3 open sourced "GitHub hot spots"

[Live broadcast preview] Will large models replace programmers? "

Recently, Meta (formerly Facebook) has open sourced their company's new generation large model Llama 3. Although only 8B and 70B versions have been released so far, its evaluation results are already better than Claude 3 Sonnet, Mistral Medium and GPT-3.5. Large model. Meta officials also stated that these are just appetizers. A stronger 400B parameter model is already being trained and is expected to be available to everyone in a few months (it is not clear whether it will be open source or not).

Talking back to last week's popular open source projects, open source projects based on LLM to build knowledge bases are very popular recently, but I didn't include any of them. Because the effect of using large models locally is not good, and the API for requesting large models is not free, I found an OpenAI API reverse proxy open source project that can be used to implement free OpenAI APIs. Content is also the key to the knowledge base, and Reader can convert web content into LLM-friendly text. For those who have no programming foundation, here is a 30-Days-Of-Python open source tutorial, learn it! Of course, you can also have fun while studying, such as trying 'Unknown Kill', which is an open source card game similar to Three Kingdoms Kill.

Contents of this article
- 1. Open Source News
  - 1.1 Meta open source Llama 3 major models
- 2. Open source hot search projects
  - 2.1 OpenAI API free reverse proxy: ChatGPT
  - 2.2 An emerging system-level programming language: Zig
  - 2.3 Convert web page content into LLM-friendly text: Reader
  - 2.4 Three Kingdoms type card game: noname
  - 2.5 30-Days-Of-Python Programming Challenge: 30-Days-Of-Python
- 3. HelloGitHub hot reviews
  - 3.1 Gadget to customize Windows taskbar transparency: TranslucentTB
  - 3.2 Cross-platform handwriting note-taking and drawing application: Rnote
- 4. Ending

1. Open Source News

1.1 Meta open source Llama 3 major models

At the beginning of this year, Zuckerberg publicly explained: Why did Meta open source the Llama model?

Improved models: Open source can leverage the power of the community to continuously improve the quality of models, because community feedback and review can help improve security and operational efficiency, which benefits everyone.
Product development: Although open source does not rule out the possibility of turning models into commercial products, leaders of open source models can integrate community innovation into their own products to improve product competitiveness.
Industry standards: Open source software has the potential to become industry standards, thereby promoting technology development and uniformity.
Attracting talent: Since developers and researchers are more likely to participate in open source projects, an open source strategy can help companies attract and recruit the best talent in the industry.

Compared with Llama 2, the newly released Llama 3 has improved in terms of parameter size, training data set, model architecture (GQA), performance, multi-language support, inference and code generation, but it does not support Chinese very well now. And meta.ai still uses Llama 2.

GitHub address: github.com/meta-llama/llama3

2. Open source hot search projects

2.1 OpenAI API free reverse proxy: ChatGPT

Main language: TypeScript , Star: 3.7k , weekly growth: 1k

Although you can now use ChatGPT (gpt-3.5-turbo model) for free without logging in, there is a fee if you want to use the interface. This project is based on the free ChatGPT website service and converts it into a free ChatGPT API. The interface returns are consistent with the official one and supports Docker deployment. It should be noted that the deployed server must be in a country and region supported by the OpenAI service.

GitHub address→ github.com/PawanOsman/ChatGPT

2.2 An emerging system-level programming language: Zig

Main language: Zig , Star: 30k , weekly growth: 300

It is an imperative, general-purpose, statically typed, compiled systems programming language focused on performance, safety, and readability. It supports compile-time generics and reflection, cross-compilation and manual memory management. It aims to improve the C language and can easily work with C language code bases. Zig is simple and direct, with no implicit control flow, no implicit memory allocation, no preprocessor, and no macros. It is especially suitable for developing compilers, operating system kernels, desktop applications, performance-sensitive applications, embedded systems, etc.

const std = @import("std");
const parseInt = std.fmt.parseInt;

test "parse integers" {
    const input = "123 67 89,99";
    const ally = std.testing.allocator;

    var list = std.ArrayList(u32).init(ally);
    // Ensure the list is freed at scope exit.
    // Try commenting out this line!
    defer list.deinit();

    var it = std.mem.tokenizeAny(u8, input, " ,");
    while (it.next()) |num| {
        const n = try parseInt(u32, num, 10);
        try list.append(n);
    }

    const expected = [_]u32{ 123, 67, 89, 99 };

    for (expected, list.items) |exp, actual| {
        try std.testing.expectEqual(exp, actual);
    }
}

GitHub address→ github.com/ziglang/zig

2.3 Convert web page content into LLM-friendly text: Reader

Main language: TypeScript , Star: 3k , weekly growth: 2.5k

This project can convert the specified URL content into clean, LLM-friendly text, thereby improving the input quality of the Agent and RAG systems, and can be used as a part of building a knowledge base. I tried it and the content extraction effect is good, but it does not work for URLs that require login to access.

GitHub address→ github.com/jina-ai/reader

2.4 Three Kingdoms type card game: noname

Main language: JavaScript , Star: 1.9k , weekly growth: 100

"Unknown Kill" is a card strategy game based on the Three Kingdoms. It is completely free and has no ads. The gameplay is the same as Three Kingdoms Kill, but it has a much higher degree of freedom. There are a large number of generals for players to choose from, and it also supports self-made generals and skills. It provides game modes such as identity, national war, landlord war, tower defense, one-on-one, and online.

GitHub address→ github.com/libccy/noname

2.5 30-Days-Of-Python Programming Challenge: 30-Days-Of-Python

Main language: Python , Star: 31k , weekly growth: 600

This project is to help people learn the Python programming language in 30 days. Through daily practice and study, they can gradually master the basic and advanced knowledge of Python. It may take more than 100 days to complete. This tutorial is suitable for beginners who want to get started with Python quickly, or developers who have a certain understanding of Python and want to gain a deeper understanding of Python through practice.

GitHub address→ github.com/Asabeneh/30-Days-Of-Python

3. HelloGitHub hot reviews

In this chapter, we will share the popular open source projects on the HelloGitHub website this week. You are welcome to share with us your experience in using these open source projects.

3.1 Gadget to customize Windows taskbar transparency: TranslucentTB

Main language: C++

This project is a tool developed in C++ for adjusting the transparency of the Windows taskbar. It is small, free, easy to use, and supports 5 taskbar states, 6 dynamic modes, and the Windows 10/11 operating system.

Project details → hellogithub.com/repository/48c9ce6373cd4e108d74bcb4f8ac7a41

3.2 Cross-platform handwriting note-taking and drawing application: Rnote

Main language: Rust

This is a drawing application written in Rust and GTK4, which can be used for sketching, handwritten notes, annotating documents, etc. It supports import/export of PDF and image files, as well as unlimited canvas, drag and drop, auto-save and other functions. It is suitable for Windows, Linux and macOS systems and needs to be used with a handwriting tablet.

Project details → hellogithub.com/repository/1d768cfa742c4fcd9c0a0bd4374cb425

4. Ending

After concluding the exciting content of this week's "GitHub Hotspot Express", I hope these open source projects can inspire everyone and help you find new tools, learning resources, or entertainment projects. If you are not satisfied after reading this, you can find more popular open source projects by reading the content of "Previous Review".

Review of past issues

The above is the entire content of this week's "GitHub Hotspot Express". If you find other fun and interesting GitHub projects, come to HelloGitHub to share them with everyone.