MyLLM

• Developed a scalable LLM pipeline in PyTorch, covering data preprocessing, tokenization, training using multi-GPU training, , and fine-tuning with SFT and RLHF (PPO/DPO). • Optimized for real-time inference quantization, and KV_cache while maintaining low-level control for efficiency. • Currently building Meta_Bot, a chatbot that explains LLM concepts, integrating Gradio for real time interaction.

PAPER2CODE

PAPER2CODE repository! This repository is dedicated to the implementation of state-of-the-art research papers in machine learning and deep learning from scratch. Our goal is to bridge the gap between theoretical research and practical implementation, providing clear, educational, and reproducible implementations of the models and methods described in these papers. We implement these models using TensorFlow and PyTorch, depending on the specific requirements of each project.

Automated Object Detection

This project provides an **automated pipeline** for **inference** and **fine-tuning** using TensorFlow 2's **Object Detection API**, aimed at simplifying and speeding up the object detection process. ### **Inference Pipeline**: 1. **Download Pre-trained Model**: Quickly access pre-trained models to start inference. 2. **Run Detection**: - **Images**: Detect objects in static images. - **Videos**: Perform detection on video files. - **Webcam**: Enable real-time detection via webcam feed. ### **Fine-Tuning Pipeline**: 1. **Download Pre-trained Model**: Start with a pre-trained model for fine-tuning. 2. **Data Pipeline**: - **Load Data**: Import images and annotations. - **Annotate**: Label images with bounding boxes and class info. - **Convert Annotations**: Transform annotations from XML to CSV. - **Create TFRecord**: Generate TFRecord files for large datasets. 3. **Retraining**: - **Load Data**: Convert data into NumPy arrays for model training. - **Model Config**: Adjust model configuration for fine-tuning. - **Device Config**: Select CPU, GPU, or TPU for training. - **Training Loops**: Fine-tune the model with iterative training. ### **Application Development**: Utilize the fine-tuned model to build custom applications in the **App** directory, enhancing workflow efficiency. ### **Efficiency and Automation**: This project automates key processes of object detection and model training, minimizing manual effort and maximizing efficiency, ultimately providing a seamless workflow for better results.

SilvaNet

**SilvaNet** is a lightweight Python library designed to simplify deep learning concepts and model building. It features **autograd-enabled tensors** for seamless backpropagation, an **intuitive API** for easy neural network construction, and support for **element-wise operations, activation functions (ReLU, sigmoid, tanh), and loss functions (softmax cross-entropy).** SilvaNet is **flexible, extensible, and supports model saving/loading**, making it an ideal tool for both beginners and educators. 🚀