Handcrafting Various AI Algorithms and Models with Excel. This project implements and demonstrates the core algorithms and concepts of artificial intelligence and deep learning using Excel, allowing beginners to operate and understand the principles of AI, including matrix multiplication, MLP, RNN, Transformer, ResNet, etc., in a unique and easy-to-understand form, reducing the threshold for learning AI.

Easy-to-Use Multi-platform LLM Chatbot. This project is an LLM chatbot and development framework built with Python, supporting integration with various messaging platforms and large language models. It offers a visual management panel and a flexible plugin extension mechanism, with features like rate limiting, whitelists, keyword filtering, image understanding, and speech-to-text.

Open-source Low-code RAG Platform. This is an RAG platform built based on LangChain and LlamaIndex, supporting one-click Docker deployment. It features an easy-to-use web interface that offers data upload, model management, internet connectivity, and prompt template functionalities, allowing users to effortlessly build, debug, and publish RAG applications without writing any code.

Open Source LLM Evaluation Framework. This framework is designed to evaluate Large Language Models (LLMs), capable of testing model performance across various tasks. It offers over 60 academic benchmarks, supports multiple model frameworks, local models, cloud services (like OpenAI), hardware acceleration, and the capability to customize tasks.

AI Tool for Removing Image Watermarks. This is an open-source image restoration tool based on TensorFlow that can remove watermarks from images with just one command.

Integrating 'Old' Devices into the Digital World. This project utilizes inexpensive hardware like ESP32 (less than 10 euros) and the TensorFlow Lite framework to achieve automatic recognition of meter readings and data transmission, making it easy to retrofit traditional devices (water meters, gas meters, electric meters) into smart devices.

Python Library for Structuring LLM Outputs. This project is a Python library designed for handling structured outputs from Large Language Models (LLMs). It leverages Pydantic for data validation and type annotation, enabling the conversion of LLM results (natural language) into structured data. It supports multiple LLM services, as well as features like automatic retries and streaming responses.

import instructor
from pydantic import BaseModel
from openai import OpenAI


# Define your desired output structure
class UserInfo(BaseModel):
    name: str
    age: int


# Patch the OpenAI client
client = instructor.from_openai(OpenAI())

# Extract structured data from natural language
user_info = client.chat.completions.create(
    model="gpt-4o-mini",
    response_model=UserInfo,
    messages=[{"role": "user", "content": "John Doe is 30 years old."}],
)

print(user_info.name)
#> John Doe
print(user_info.age)
#> 30

Lightweight C++ AI Toolkit. This is an AI toolkit written in C++, which includes more than 100 AI models, covering areas such as object detection, facial recognition, segmentation, and background removal. It supports mainstream inference engines like ONNXRuntime, MNN, NCNN, TNN, and TensorRT, helping developers quickly deploy and utilize AI models.

#include "lite/lite.h"

int main(int argc, char *argv[]) {
  std::string onnx_path = "yolov5s.onnx";
  std::string test_img_path = "test_yolov5.jpg";
  std::string save_img_path = "test_results.jpg";

  auto *yolov5 = new lite::cv::detection::YoloV5(onnx_path); 
  std::vector<lite::types::Boxf> detected_boxes;
  cv::Mat img_bgr = cv::imread(test_img_path);
  yolov5->detect(img_bgr, detected_boxes);
  
  lite::utils::draw_boxes_inplace(img_bgr, detected_boxes);
  cv::imwrite(save_img_path, img_bgr);  
  delete yolov5;
  return 0;
}

Train a Small Language Model from Scratch. This is not only an implementation of a mini-language model, but also an introductory tutorial for LLMs, aimed at lowering the barrier to learning and getting started with LLMs. It provides the full process code and tutorials from data preprocessing to model training, fine-tuning, and inference. The smallest model has a parameter count of only 0.02B, which can be easily run on a regular GPU.

Open Source LLMs Vulnerability Scanner. This project is an open-source tool developed by NVIDIA for scanning vulnerabilities in Large Language Models (LLMs). It supports detecting potential security issues and adverse behaviors from LLMs, including hallucination, data leakage, prompt injection, erroneous messages, generation of toxic content, and jailbreaking.