Distributed Llama

Tensor parallelism is all you need. Run LLMs on weak devices or make powerful devices even more powerful by distributing the workload and dividing the RAM usage. This project proves that it's possible split the workload of LLMs across multiple devices and achieve a significant speedup. Distributed Llama allows you to run huge LLMs in-house. The project uses TCP sockets to synchronize the state. You can easily configure your AI cluster by using a home router.

Tip

Check out the new article: 🌳 How to Run Llama 3.1 405B on Home Devices? Build AI Cluster!

🔥 Setup Root Node by Single Command

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

Model	Purpose	Size	Command
TinyLlama 1.1B 3T Q40	Benchmark	844 MB	python launch.py tinyllama_1_1b_3t_q40
Llama 3 8B Q40	Benchmark	6.32 GB	python launch.py llama3_8b_q40
Llama 3 8B Instruct Q40	Chat, API	6.32 GB	python launch.py llama3_8b_instruct_q40
Llama 3.1 8B Instruct Q40	Chat, API	6.32 GB	python launch.py llama3_1_8b_instruct_q40
Llama 3.1 405B Instruct Q40	Chat, API	238 GB	python launch.py llama3_1_405b_instruct_q40
Llama 3.2 1B Instruct Q40	Chat, API	1.7 GB	python launch.py llama3_2_1b_instruct_q40
Llama 3.2 3B Instruct Q40	Chat, API	3.4 GB	python launch.py llama3_2_3b_instruct_q40

🛠️ Convert Model Manually

Supported architectures: Llama, Mixtral

• How to Convert Llama 2, Llama 3, Llama 3.1
• 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.
• CPU support only, GPU support is planned, optimized for (weights format × buffer format):
  • ARM CPUs
    • ✅ F32 × F32
    • ❌ F16 × F32
    • ✅ Q40 × F32
    • ✅ Q40 × Q80
  • x86_64 AVX2 CPUs
    • ✅ F32 × F32
    • ❌ F16 × F32
    • ✅ Q40 × F32
    • ✅ Q40 × Q80

👷 Architecture

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.

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.	10.0.0.1:9991 10.0.0.2:9991
--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
--port <port>	Binding port.	9999

Inference

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

📊 Measurements

Average Token Generation Time

I - inference time of the root node, T - network transfer time of the root node.

Raspberry Pi 5 8GB

_{^{Weights = Q40, Buffer = Q80, nSamples = 16, switch = TP-Link LS1008G, tested on 0.3.1 version}}

Model	1 x RasPi 5 8 GB	2 x RasPi 5 8 GB	4 x RasPi 5 8 GB
Llama 2 7B	441.09 ms, 2.26 t/s I: 434.84 ms, T: 5.25 ms	341.46 ms, 2.92 t/s I: 257.78 ms, T: 83.27 ms	219.08 ms, 4.56 t/s 🔥 I: 163.42 ms, T: 55.25 ms
Llama 3 8B	564.31 ms, 1.77 t/s I: 556.67 ms, T: 6.17 ms	444.27 ms, 2.25 t/s I: 362.73 ms, T: 80.11 ms	331.47 ms, 3.01 t/s 🔥 I: 267.62 ms, T: 62.34 ms

Raspberry Pi 4B 8 GB

_{^{8 x Raspberry Pi 4B 8GB}}

_{^{Distributed Llama running Llama 2 70B Q40 on 8 Raspberry Pi 4B devices}}

_{^{Weights = Q40, Buffer = Q80, nSamples = 16, switch = TP-Link LS1008G, tested on 0.1.0 version}}

Model	1 x RasPi 4B 8 GB	2 x RasPi 4B 8 GB	4 x RasPi 4B 8 GB	8 x RasPi 4B 8 GB
Llama 2 7B	1312.50 ms I: 1307.94 ms, T: 1.81 ms	793.69 ms I: 739.00 ms, T: 52.50 ms	494.00 ms 🔥 I: 458.81 ms, T: 34.06 ms	588.19 ms I: 296.69 ms, T: 289.75 ms
Llama 2 13B	Not enough RAM	1497.19 ms I: 1465.06 ms, T: 30.88 ms	848.19 ms 🔥 I: 746.88 ms, T: 99.50 ms	1114.88 ms I: 460.8 ms, T: 652.88 ms
Llama 2 70B	Not enough RAM	Not enough RAM	Not enough RAM	4842.81 ms 🔥 I: 2121.94 ms, T: 2719.62 ms

x86_64 CPU Cloud Server

_{^{Weights = Q40, Buffer = Q80, nSamples = 16, VMs = c3d-highcpu-30, tested on 0.1.0 version}}

Model	1 x VM	2 x VM	4 x VM
Llama 2 7B	101.81 ms I: 101.06 ms, T: 0.19 ms	69.69 ms I: 61.50 ms, T: 7.62 ms	53.69 ms 🔥 I: 40.25 ms, T: 12.81 ms
Llama 2 13B	184.19 ms I: 182.88 ms, T: 0.69 ms	115.38 ms I: 107.12 ms, T: 7.81 ms	86.81 ms 🔥 I: 66.25 ms, T: 19.94 ms
Llama 2 70B	909.69 ms I: 907.25 ms, T: 1.75 ms	501.38 ms I: 475.50 ms, T: 25.00 ms	293.06 ms 🔥 I: 264.00 ms, T: 28.50 ms

Network Transfer for Generating Token

F32 Buffer

Model	2 devices	4 devices	8 devices
Llama 3 8B	2048 kB	6144 kB	14336 kB

Q80 Buffer

Model	2 devices	4 devices	8 devices
Llama 3 8B	544 kB	1632 kB	3808 kB

📟 Setup Raspberry Pi Devices

1. Install Raspberry Pi OS Lite (64 bit) on your Raspberry Pi devices. This OS doesn't have desktop environment.
2. Connect all devices to your switch or router.
3. Connect to all devices via SSH.

ssh user@raspberrypi1.local
ssh user@raspberrypi2.local

4. Install Git:

sudo apt install git

5. Clone this repository and compile Distributed Llama on all devices:

git clone https://github.com/b4rtaz/distributed-llama.git
make dllama
make dllama-api

6. Transfer weights and the tokenizer file to the root device.
7. Optional: assign static IP addresses.

sudo ip addr add 10.0.0.1/24 dev eth0 # 1th device
sudo ip addr add 10.0.0.2/24 dev eth0 # 2th device

8. Run worker nodes on worker devices:

sudo nice -n -20 ./dllama worker --port 9998 --nthreads 4

9. Run root node on the root device:

sudo nice -n -20 ./dllama inference --model dllama_model_meta-llama-3-8b_q40.m --tokenizer dllama_tokenizer_llama3.t --buffer-float-type q80 --prompt "Hello world" --steps 16 --nthreads 4 --workers 10.0.0.2:9998

To add more worker nodes, just add more addresses to the --workers argument.

./dllama inference ... --workers 10.0.0.2:9998 10.0.0.3:9998 10.0.0.4:9998

💻 Setup computers with MacOS, Linux, or Windows

You need x86_64 AVX2 CPUs or ARM CPUs. Different devices may have different CPUs.

MacOS or Linux

The below instructions are for Debian-based distributions but you can easily adapt them to your distribution, macOS.

1. Install Git and GCC:

sudo apt install git build-essential

2. Clone this repository and compile Distributed Llama on all computers:

git clone https://github.com/b4rtaz/distributed-llama.git
make dllama
make dllama-api

Continue to point 3.

Windows

1. Install Git and Mingw (via Chocolatey):

choco install mingw

2. Clone this repository and compile Distributed Llama on all computers:

git clone https://github.com/b4rtaz/distributed-llama.git
make dllama
make dllama-api

Continue to point 3.

Run Cluster

3. Transfer weights and the tokenizer file to the root computer.
4. Run worker nodes on worker computers:

./dllama worker --port 9998 --nthreads 4

5. Run root node on the root computer:

./dllama inference --model dllama_model_meta-llama-3-8b_q40.m --tokenizer dllama_tokenizer_llama3.t --buffer-float-type q80 --prompt "Hello world" --steps 16 --nthreads 4 --workers 192.168.0.1:9998

To add more worker nodes, just add more addresses to the --workers argument.

./dllama inference ... --workers 192.168.0.1:9998 192.168.0.2:9998 192.168.0.3:9998

✋ 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}
}