DQN Labs

DQN Labs is an independent AI research project focused on building, training, and experimenting with Large Language Models (LLMs).

Our goal is to push the limits of small, efficient AI models and make powerful AI systems accessible to everyone.

Try our models at dqnlabsai.web.app

🔬 Current Focus

DQN Labs is currently focused on LLM development and fine-tuning, including:

• 🧠 Training performant, cutting-edge models for local inference.
• 🛜 Hosting our models at dqnlabsai.web.app!
• 💻 Improving code generation and reasoning ability.
• 📊 Evaluating models on benchmarks such as MMLU, HumanEval, GSM8k, etc.
• ⚡ Publishing models on LM Studio and HuggingFace for consumer hardware.

📦 What You'll Find Here

This organization hosts:

• 🤖 Fine-tuned language models
• 🛜 100% free inference at dqnlabsai.web.app!
• 📊 Evaluation results and benchmarks

Most experiments focus on efficient training methods and lightweight models.

Everybody is free to contribute to the organizations, whether you choose to do so through providing new data, sharing your research through papers, or even fine tuning a few models yoruself!

🛠 Tech Stack

• Python
• Hugging Face Transformers
• MLX (Apple Silicon optimization)
• LoRA / parameter-efficient fine-tuning

🎯 Vision

DQN Labs is exploring how capable small AI models can become through better data, smarter training, and efficient infrastructure. We believe that in order to achieve powerful and small local AI, we need specialization for models. Our current model offerings include:

• dqnGPT 3B (general assistant)

• dqnCode v1 4B (flagship model, powerful coding-focused model with tool use compatibility)

• dqnMath v1 4B (efficient, low-token math solving model for daily use)

• dqnScience BETA 4B (heavy reasoning model, super performant for size and excels in scientific reasoning)

🌐 Links

• 🌐 Website: dqnlabsai.web.app
• 🤗 Hugging Face: DQN Labs (you're here already!)
• 🎥 YouTube: DQN Labs

⚡ Motto

Local AI for everyone.