This repository guides you through the process of building a GPT-style Large Language Model (LLM) from scratch using PyTorch. The structure and approach are inspired by the book Build a Large Language Model (From Scratch) by Sebastian Raschka.

• Title: Build a Large Language Model (From Scratch)
• Author: Sebastian Raschka
• Publisher: Manning Publications
• Link: manning.com/books/build-a-large-language-model-from-scratch
• Free Version: Github Gist
• Download PDF: PDF Version
• Resourse: Check the resource from where I learned.

• Set up the environment and tools.
• Learn about tokenization, embeddings, and the idea of a "language model".
• Encode input/output sequences and build basic forward models.
• Understand unidirectional processing and causal language modeling.

• Explore Transformer components: attention, multi-head attention, and positional encoding.
• Implement residual connections, normalization, and feedforward layers.
• Build a GPT-style decoder-only transformer architecture.

• Load and preprocess datasets like TinyShakespeare.
• Implement batch creation, context windows, and training routines.
• Use cross-entropy loss, optimizers, and learning rate schedulers.
• Monitor perplexity and improve generalization.

• Generate text using greedy, top-k, top-p, and temperature sampling.
• Evaluate and tune generation.
• Export and convert model for Hugging Face compatibility.
• Deploy via Hugging Face Hub and Gradio Space.

FaseehGPT is an advanced pipeline for training a GPT-style language model specifically designed for the Arabic language. FaseehGPT

• Python 3.8+
• PyTorch
• NumPy
• Matplotlib
• JupyterLab or Notebooks
• Hugging Face libraries: transformers, datasets, huggingface_hub
• gradio for deployment

git clone https://github.com/codewithdark-git/Building-LLMs-from-scratch.git
cd Building-LLMs-from-scratch
pip install -r requirements.txt

Building-LLMs-from-scratch/
├── notebooks/            # Weekly learning notebooks
├── models/               # Model architectures & checkpoints
├── data/                 # Preprocessing and datasets
├── hf_deploy/            # Hugging Face config & deployment scripts
├── theoretical/          # Podcast & theoretical discussions
├── utils/                # Helper scripts
├── requirements.txt
└── README.md

This project includes:

• Scripts to convert the model for 🤗 Transformers compatibility
• Uploading to Hugging Face Hub
• Launching an interactive demo on Hugging Face Spaces using Gradio

You’ll find detailed instructions inside the hf_deploy/ folder.

• Transformers Docs
• Hugging Face

MIT License — see the LICENSE file for details.