🚀 SwiftLet

SwiftLet is a lightweight Python framework for running open-source Large Language Models (LLMs) locally using safetensors.

Swiftlet is a lightweight and educational framework that provides reimplementations of various Large Language Models (LLMs).
It is designed for learning, experimentation, and local execution of models — all without relying on external LLM libraries.

---

🚀 Features

• ✅ Local execution of supported LLMs
• ✅ Minimal dependencies, easy to understand
• 📦 Clean architecture for adding more models
• 🔍 Designed for research, prototyping, and educational use
• 🛠️ Open for contributions and experiments

---

🧠 Implemented Models

Model	Status	Notes
Gemma 1	✅ Working	Text-only
Gemma 2	✅ Working	Text-only
Gemma 3	✅ Partially Working	Vision support not implemented
Qwen 2	✅ Working	Text-only (MoE support not implemented)

ℹ️ More models will be added soon!

---

Planned Features

SwiftLet is under active development. The following features are planned for future releases:

• Integration with Native Runtimes
  Support for running LLMs via optimized backends like llama.cpp for improved performance on local machines.

• File Interaction Support
  Enable LLMs to read and process local documents, files, and structured data formats.

• Modular Tool Integration
  Easily connect models to external tools, functions, or APIs to extend their utility.

• Enhanced Model Management
  Tools to manage multiple models, switch between them, and handle custom configurations.

• Prompt Templates and Inference APIs
  Built-in support for structured prompting, templates, and customizable inference pipelines.

• Extensibility and Plugins
  A modular architecture that allows developers to add new capabilities with simple plugin hooks.

• Performance Monitoring and Debugging
  Tools for logging, inspecting model behavior, and optimizing local performance.

---

Codes

1. Gemma: View on Kaggle