Distributed Llama

Connect home devices into a powerful cluster to accelerate LLM inference. More devices mean faster performance, leveraging tensor parallelism and high-speed synchronization over Ethernet.

Supports Linux, macOS, and Windows. Optimized for ARM and x86_64 AVX2 CPUs.

How to Run

• 💻 How to Run on Linux, MacOS or Windows
• 🍓 How to Run on Raspberry Pi
• 🧠 How to Run on GPU

News

• 16 Sep 2025 - Qwen 3 MoE models are now supported on Vulkan.
• 5 Sep 2025 - Qwen 3 MoE models are now supported on CPU.
• 3 Aug 2025 - Qwen 3 0.6B, 1.7B, 8B and 14B models are now supported.
• 23 Mar 2025 - 🌋 Experimental Vulkan support
• 12 Feb 2025 - 🚧 Merged the fundamental codebase refactor
• 9 Jan 2025 - 🍎 Llama 3.3 70B on 4 x Mac Mini M4 Pro 24GB RAM

🔥 Setup Root Node by Single Command

Python 3 and C++ compiler required. The command will download the model and the tokenizer.

🛠️ Convert Model Manually

• 🤗 How to Convert Hugging Face Model

🚧 Known Limitations

• You can run Distributed Llama only on 1, 2, 4... 2^n nodes.
• The maximum number of nodes is equal to the number of KV heads in the model #70.
• Only the following quantizations are supported #183:
  • q40 model with q80 buffer-float-type
  • f32 model with f32 buffer-float-type

👷 Architecture

[🔀 SWITCH OR ROUTER]
      | | | |
      | | | |_______ 🔸 device1 (ROOT)     10.0.0.1
      | | |_________ 🔹 device2 (WORKER 1) 10.0.0.2:9999
      | |___________ 🔹 device3 (WORKER 2) 10.0.0.3:9999
      |_____________ 🔹 device4 (WORKER 3) 10.0.0.4:9999
                        ...

The project is split up into two parts:

• 🔸 Root node - it's responsible for loading the model and weights and forward them to workers. Also, it synchronizes the state of the neural network. The root node is also a worker, it processes own slice of the neural network.
• 🔹 Worker node - it processes own slice of the neural network. It doesn't require any configuration related to the model.

You always need the root node and you can add 2^n - 1 worker nodes to speed up the inference. The RAM usage of the neural network is split up across all nodes. The root node requires a bit more RAM than worker nodes.

🎹 Commands

• dllama inference - run the inference with a simple benchmark,
• dllama chat - run the CLI chat,
• dllama worker - run the worker node,
• dllama-api - run the API server.

🎹 Supported Arguments

Inference, Chat, API

Argument	Description	Example
--model <path>	Path to model.	dllama_model_meta-llama-3-8b_q40.m
--tokenizer <path>	Tokenizer to model.	dllama_tokenizer_llama3.t
--buffer-float-type <type>	Float precision of synchronization.	q80
--workers <workers>	Addresses of workers (ip:port), separated by space.	10.0.0.1:9999 10.0.0.2:9999
--max-seq-len <n>	The maximum sequence length, it helps to reduce the RAM usage.	4096

Inference, Chat, Worker, API

Argument	Description	Example
--nthreads <n>	Amount of threads. Don't set a higher value than number of CPU cores.	4

Worker, API

Argument	Description	Example
--host <addr>	Binding address.	127.0.0.1
--port <port>	Binding port.	9999

Inference

Argument	Description	Example
--prompt <prompt>	Initial prompt.	"Hello World"
--steps <steps>	Number of tokens to generate.	256

📊 Measurements

Please check the discussions section, where many measurements were published on different configurations.

✋ Contribution

Feel free to contribute to this project. For small changes, simply create a new merge request. For larger changes, please create an issue to discuss your plans. Please follow these guidelines when contributing:

• Make only minimal changes and avoid modifying files that are not necessary.
• Ensure the code is compatible across all supported systems and CPUs.
• This repository is maintained in English.

💡 License

This project is released under the MIT license.

📖 Citation

@misc{dllama,
  author = {Bartłomiej Tadych},
  title = {Distributed Llama},
  year = {2024},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/b4rtaz/distributed-llama}},
  commit = {7eb77ca93ec0d502e28d36b6fb20039b449cbea4}
}