|
1 | | -# Building-LLMs-from-scratch |
2 | | -Building LLMs from scratch |
| 1 | +# 🧠 Building LLMs from Scratch – A 30-Day Journey |
| 2 | + |
| 3 | +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**. |
| 4 | + |
| 5 | +--- |
| 6 | + |
| 7 | +## 📘 Reference Book |
| 8 | + |
| 9 | +* **Title**: *Build a Large Language Model (From Scratch)* |
| 10 | +* **Author**: Sebastian Raschka |
| 11 | +* **Publisher**: Manning Publications |
| 12 | +* **Link**: [manning.com/books/build-a-large-language-model-from-scratch](https://www.manning.com/books/build-a-large-language-model-from-scratch) |
| 13 | + |
| 14 | +--- |
| 15 | + |
| 16 | +## 🗓️ Weekly Curriculum Overview |
| 17 | + |
| 18 | +### 🔹 Week 1: Core Concepts of Language Modeling |
| 19 | + |
| 20 | +* Set up your development environment and explore foundational concepts in NLP and tokenization. |
| 21 | +* Learn how to numerically encode language, build vocabularies, and understand token embeddings. |
| 22 | +* Grasp the importance of attention mechanisms and understand how to implement them manually. |
| 23 | + |
| 24 | +--- |
| 25 | + |
| 26 | +### 🔹 Week 2: Building the Transformer |
| 27 | + |
| 28 | +* Dive into the architecture of Transformer models from the ground up. |
| 29 | +* Learn about positional encoding, residual connections, normalization, and multi-head attention. |
| 30 | +* Construct and test a decoder-style Transformer (like GPT) with causal masking. |
| 31 | + |
| 32 | +--- |
| 33 | + |
| 34 | +### 🔹 Week 3: Training and Optimization |
| 35 | + |
| 36 | +* Prepare and preprocess datasets such as TinyShakespeare or WikiText. |
| 37 | +* Create efficient data pipelines and define model training loops. |
| 38 | +* Apply optimizer strategies, monitor model perplexity, and manage model checkpoints. |
| 39 | + |
| 40 | +--- |
| 41 | + |
| 42 | +### 🔹 Week 4: Evaluation and Hugging Face Deployment |
| 43 | + |
| 44 | +* Implement text generation methods including greedy and top-k sampling. |
| 45 | +* Evaluate the model's outputs and compare them with other LLMs. |
| 46 | +* Learn how to convert your model for Hugging Face Hub and push it live. |
| 47 | +* Create a Hugging Face Space using Gradio to serve your model with an interactive UI. |
| 48 | + |
| 49 | +--- |
| 50 | + |
| 51 | +## 🛠️ Getting Started |
| 52 | + |
| 53 | +### Prerequisites |
| 54 | + |
| 55 | +* Python 3.8+ |
| 56 | +* PyTorch |
| 57 | +* NumPy |
| 58 | +* Matplotlib |
| 59 | +* JupyterLab or Notebooks |
| 60 | +* Hugging Face libraries: `transformers`, `datasets`, `huggingface_hub` |
| 61 | +* `gradio` for deployment |
| 62 | + |
| 63 | +### Installation |
| 64 | + |
| 65 | +```bash |
| 66 | +git clone https://github.com/codewithdark-git/Building-LLMs-from-scratch.git |
| 67 | +cd Building-LLMs-from-scratch |
| 68 | +pip install -r requirements.txt |
| 69 | +``` |
| 70 | + |
| 71 | +--- |
| 72 | + |
| 73 | +## 📁 Project Structure |
| 74 | + |
| 75 | +``` |
| 76 | +Building-LLMs-from-scratch/ |
| 77 | +├── notebooks/ # Weekly learning notebooks |
| 78 | +├── models/ # Model architectures & checkpoints |
| 79 | +├── data/ # Preprocessing and datasets |
| 80 | +├── hf_deploy/ # Hugging Face config & deployment scripts |
| 81 | +├── utils/ # Helper scripts |
| 82 | +├── requirements.txt |
| 83 | +└── README.md |
| 84 | +``` |
| 85 | + |
| 86 | +--- |
| 87 | + |
| 88 | +## 🚀 Hugging Face Deployment |
| 89 | + |
| 90 | +This project includes: |
| 91 | + |
| 92 | +* Scripts to convert the model for 🤗 Transformers compatibility |
| 93 | +* Uploading to Hugging Face Hub |
| 94 | +* Launching an interactive demo on Hugging Face Spaces using Gradio |
| 95 | + |
| 96 | +You’ll find detailed instructions inside the `hf_deploy/` folder. |
| 97 | + |
| 98 | +--- |
| 99 | + |
| 100 | +## 📚 Resources |
| 101 | + |
| 102 | +* [Transformers Docs](https://huggingface.co/docs/transformers) |
| 103 | +* [Hugging Face](https://huggingface.co) |
| 104 | + |
| 105 | +--- |
| 106 | + |
| 107 | +## 📄 License |
| 108 | + |
| 109 | +MIT License — see the `LICENSE` file for details. |
0 commit comments