Run large language models at home, BitTorrent-style.
Fine-tuning and inference up to 10x faster than offloading


Generate text with distributed **LLaMA 2 (70B)**, **Stable Beluga 2**, **Guanaco-65B** or **BLOOM-176B** and fine‑tune them for your own tasks — right from your desktop computer or Google Colab: ```python from transformers import AutoTokenizer from petals import AutoDistributedModelForCausalLM model_name = "stabilityai/StableBeluga2" # You can also use "meta-llama/Llama-2-70b-hf", "meta-llama/Llama-2-70b-chat-hf", # repos with LLaMA-65B, "bigscience/bloom", or "bigscience/bloomz" tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoDistributedModelForCausalLM.from_pretrained(model_name) # Embeddings & prompts are on your device, transformer blocks are distributed across the Internet inputs = tokenizer("A cat sat", return_tensors="pt")["input_ids"] outputs = model.generate(inputs, max_new_tokens=5) print(tokenizer.decode(outputs[0])) # A cat sat on a mat... ```

πŸš€  Try now in Colab

πŸ¦™ **Want to run LLaMA 2?** Request access to its weights at the ♾️ [Meta AI website](https://ai.meta.com/resources/models-and-libraries/llama-downloads/) and πŸ€— [Model Hub](https://huggingface.co/meta-llama/Llama-2-70b-hf), then run `huggingface-cli login` in the terminal before loading the model. Or just try it in our [chatbot app](https://chat.petals.dev). πŸ“‹ **Terms of use.** Make sure you follow the model license (see [LLaMA 2](https://bit.ly/llama2-license), [Stable Beluga 2](https://huggingface.co/stabilityai/StableBeluga2/blob/main/LICENSE.txt), [LLaMA](https://bit.ly/llama-license), and [BLOOM](https://bit.ly/bloom-license)). πŸ” **Privacy.** Your data will be processed by other people in the public swarm. Learn more about privacy [here](https://github.com/bigscience-workshop/petals/wiki/Security,-privacy,-and-AI-safety). For sensitive data, you can set up a [private swarm](https://github.com/bigscience-workshop/petals/wiki/Launch-your-own-swarm) among people you trust. πŸ’¬ **Any questions?** Ping us in [our Discord](https://discord.gg/KdThf2bWVU)! ### Connect your GPU and increase Petals capacity Petals is a community-run system — we rely on people sharing their GPUs. You can check out available servers on our [swarm monitor](https://health.petals.dev) and connect your GPU to help serving one of the models! 🐍 **Linux + Anaconda.** Run these commands: ```bash conda install pytorch pytorch-cuda=11.7 -c pytorch -c nvidia pip install git+https://github.com/bigscience-workshop/petals python -m petals.cli.run_server stabilityai/StableBeluga2 ``` πŸͺŸ **Windows + WSL.** Follow the guide on our [Wiki](https://github.com/bigscience-workshop/petals/wiki/Run-Petals-server-on-Windows). πŸ‹ **Any OS + Docker.** Run our [Docker](https://www.docker.com) image: ```bash sudo docker run -p 31330:31330 --ipc host --gpus all --volume petals-cache:/cache --rm learningathome/petals:main \ python -m petals.cli.run_server --port 31330 stabilityai/StableBeluga2 ``` These commands will host a part of [Stable Beluga 2](https://huggingface.co/stabilityai/StableBeluga2) on your machine. You can also host `meta-llama/Llama-2-70b-hf`, `meta-llama/Llama-2-70b-chat-hf`, repos with LLaMA-65B, `bigscience/bloom`, `bigscience/bloomz`, and other compatible models from πŸ€— [Model Hub](https://huggingface.co/models), or [add support](https://github.com/bigscience-workshop/petals/wiki/Run-a-custom-model-with-Petals) for new model architectures. πŸ¦™ **Want to host LLaMA 2?** Request access to its weights at the ♾️ [Meta AI website](https://ai.meta.com/resources/models-and-libraries/llama-downloads/) and πŸ€— [Model Hub](https://huggingface.co/meta-llama/Llama-2-70b-hf), generate an πŸ”‘ [access token](https://huggingface.co/settings/tokens), then use this command for `petals.cli.run_server`: ```bash python -m petals.cli.run_server meta-llama/Llama-2-70b-chat-hf --token YOUR_TOKEN_HERE ``` πŸ’¬ **FAQ.** Check out our [Wiki](https://github.com/bigscience-workshop/petals/wiki/FAQ:-Frequently-asked-questions#running-a-server) to learn how to use multple GPUs, restart the server on reboot, etc. If you have any issues, ping us in [our Discord](https://discord.gg/X7DgtxgMhc)! πŸ”’ **Security.** Hosting a server does not allow others to run custom code on your computer. Learn more [here](https://github.com/bigscience-workshop/petals/wiki/Security,-privacy,-and-AI-safety). πŸ† **Thank you!** Once you load and host 10+ blocks, we can show your name or link on the [swarm monitor](https://health.petals.dev) as a way to say thanks. You can specify them with `--public_name YOUR_NAME`. ### Check out tutorials, examples, and more Basic tutorials: - Getting started: [tutorial](https://colab.research.google.com/drive/1uCphNY7gfAUkdDrTx21dZZwCOUDCMPw8?usp=sharing) - Prompt-tune LLaMA-65B for text semantic classification: [tutorial](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-sst2.ipynb) - Prompt-tune BLOOM to create a personified chatbot: [tutorial](https://colab.research.google.com/github/bigscience-workshop/petals/blob/main/examples/prompt-tuning-personachat.ipynb) Useful tools and advanced guides: - [Chatbot web app](https://chat.petals.dev) (connects to Petals via an HTTP/WebSocket endpoint): [source code](https://github.com/petals-infra/chat.petals.dev) - [Monitor](https://health.petals.dev) for the public swarm: [source code](https://github.com/petals-infra/health.petals.dev) - Launch your own swarm: [guide](https://github.com/bigscience-workshop/petals/wiki/Launch-your-own-swarm) - Run a custom foundation model: [guide](https://github.com/bigscience-workshop/petals/wiki/Run-a-custom-model-with-Petals) Learning more: - Frequently asked questions: [FAQ](https://github.com/bigscience-workshop/petals/wiki/FAQ:-Frequently-asked-questions) - In-depth system description: [paper](https://arxiv.org/abs/2209.01188) ## How does it work? - Petals runs large language models like [LLaMA](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) and [BLOOM](https://huggingface.co/bigscience/bloom) **collaboratively** β€” you load a small part of the model, then join people serving the other parts to run inference or fine-tuning. - Single-batch inference runs at **up to 6 steps/sec** for **LLaMA 2** (70B) and ≈ 1 step/sec for BLOOM-176B. This is [up to 10x faster](https://github.com/bigscience-workshop/petals#benchmarks) than offloading, enough to build [chatbots](https://chat.petals.dev) and other interactive apps. Parallel inference reaches hundreds of tokens/sec. - Beyond classic language model APIs β€” you can employ any fine-tuning and sampling methods, execute custom paths through the model, or see its hidden states. You get the comforts of an API with the flexibility of PyTorch.

πŸ“š  See FAQ            πŸ“œ  Read paper

## Installation Here's how to install Petals with [Anaconda](https://www.anaconda.com/products/distribution) on Linux: ```bash conda install pytorch pytorch-cuda=11.7 -c pytorch -c nvidia pip install git+https://github.com/bigscience-workshop/petals ``` If you don't use Anaconda, you can install PyTorch in [any other way](https://pytorch.org/get-started/locally/). If you want to run models with 8-bit weights, please install PyTorch with CUDA 11.x or newer for compatility with [bitsandbytes](https://github.com/timDettmers/bitsandbytes). See the instructions for macOS and Windows, the full requirements, and troubleshooting advice in our [FAQ](https://github.com/bigscience-workshop/petals/wiki/FAQ:-Frequently-asked-questions#running-a-client). ## Benchmarks The benchmarks below are for BLOOM-176B:
Network Single-batch inference
(steps/s)
Parallel forward
(tokens/s)
Bandwidth Round-trip
latency
Sequence length Batch size
128 2048 1 64
Offloading, max. possible speed on 1x A100 1
256 Gbit/s 0.18 0.18 2.7 170.3
128 Gbit/s 0.09 0.09 2.4 152.8
Petals on 14 heterogeneous servers across Europe and North America 2
Real world 0.83 0.79 32.6 179.4
Petals on 3 servers, with one A100 each 3
1 Gbit/s < 5 ms 1.71 1.54 70.0 253.6
100 Mbit/s < 5 ms 1.66 1.49 56.4 182.0
100 Mbit/s 100 ms 1.23 1.11 19.7 112.2
1 **An upper bound for offloading performance.** We base our offloading numbers on the best possible hardware setup for offloading: CPU RAM offloading via PCIe 4.0 with 16 PCIe lanes per GPU and PCIe switches for pairs of GPUs. We assume zero latency for the upper bound estimation. In 8-bit, the model uses 1 GB of memory per billion parameters. PCIe 4.0 with 16 lanes has a throughput of 256 Gbit/s, so offloading 176B parameters takes 5.5 seconds. The throughput is twice as slow (128 Gbit/s) if we have two GPUs behind the same PCIe switch. 2 **A real-world distributed setting** with 14 servers holding 2Γ— RTX 3060, 4Γ— 2080Ti, 2Γ— 3090, 2Γ— A4000, and 4Γ— A5000 GPUs. These are personal servers and servers from university labs, spread across Europe and North America and connected to the Internet at speeds of 100–1000 Mbit/s. 4 servers operate from under firewalls. 3 **An optimistic setup** that requires least communication. The client nodes have 8 CPU cores and no GPU. We provide more evaluations and discuss these results in more detail in **Section 3.3** of our [paper](https://arxiv.org/pdf/2209.01188.pdf). ## πŸ› οΈ Contributing Please see our [FAQ](https://github.com/bigscience-workshop/petals/wiki/FAQ:-Frequently-asked-questions#contributing) on contributing. ## πŸ“œ Citation Alexander Borzunov, Dmitry Baranchuk, Tim Dettmers, Max Ryabinin, Younes Belkada, Artem Chumachenko, Pavel Samygin, and Colin Raffel. [Petals: Collaborative Inference and Fine-tuning of Large Models.](https://arxiv.org/abs/2209.01188) _arXiv preprint arXiv:2209.01188,_ 2022. ```bibtex @article{borzunov2022petals, title = {Petals: Collaborative Inference and Fine-tuning of Large Models}, author = {Borzunov, Alexander and Baranchuk, Dmitry and Dettmers, Tim and Ryabinin, Max and Belkada, Younes and Chumachenko, Artem and Samygin, Pavel and Raffel, Colin}, journal = {arXiv preprint arXiv:2209.01188}, year = {2022}, url = {https://arxiv.org/abs/2209.01188} } ``` --------------------------------------------------------------------------------

This project is a part of the BigScience research workshop.