🦾 OpenLLM

An open platform for operating large language models (LLMs) in production.
Fine-tune, serve, deploy, and monitor any LLMs with ease.

📖 Introduction

OpenLLM is an open-source platform designed to facilitate the deployment and operation of large language models (LLMs) in real-world applications. With OpenLLM, you can run inference on any open-source LLM, deploy them on the cloud or on-premises, and build powerful AI applications.

Key features include:

🚂 State-of-the-art LLMs: Integrated support for a wide range of open-source LLMs and model runtimes, including but not limited to Llama 2, StableLM, Falcon, Dolly, Flan-T5, ChatGLM, and StarCoder.

🔥 Flexible APIs: Serve LLMs over a RESTful API or gRPC with a single command. You can interact with the model using a Web UI, CLI, Python/JavaScript clients, or any HTTP client of your choice.

⛓️ Freedom to build: First-class support for LangChain, BentoML, OpenAI endpoints, and Hugging Face, allowing you to easily create your own AI applications by composing LLMs with other models and services.

🎯 Streamline deployment: Automatically generate your LLM server Docker images or deploy as serverless endpoints via ☁️ BentoCloud, which effortlessly manages GPU resources, scales according to traffic, and ensures cost-effectiveness.

🤖️ Bring your own LLM: Fine-tune any LLM to suit your needs. You can load LoRA layers to fine-tune models for higher accuracy and performance for specific tasks. A unified fine-tuning API for models (LLM.tuning()) is coming soon.

⚡ Quantization: Run inference with less computational and memory costs with quantization techniques such as LLM.int8, SpQR (int4), AWQ, GPTQ, and SqueezeLLM.

📡 Streaming: Support token streaming through server-sent events (SSE). You can use the /v1/generate_stream endpoint for streaming responses from LLMs.

🔄 Continuous batching: Support continuous batching via vLLM for increased total throughput.

OpenLLM is designed for AI application developers working to build production-ready applications based on LLMs. It delivers a comprehensive suite of tools and features for fine-tuning, serving, deploying, and monitoring these models, simplifying the end-to-end deployment workflow for LLMs.

💾 TL/DR

For starter, we provide two ways to quickly try out OpenLLM:

Jupyter Notebooks

Try this OpenLLM tutorial in Google Colab: Serving Llama 2 with OpenLLM.

Docker

We provide a docker container that helps you start running OpenLLM:

docker run --rm -it -p 3000:3000 ghcr.io/bentoml/openllm start facebook/opt-1.3b --backend pt

Note

Given you have access to GPUs and have setup nvidia-docker, you can additionally pass in --gpus to use GPU for faster inference and optimization

docker run --rm --gpus all -p 3000:3000 -it ghcr.io/bentoml/openllm start HuggingFaceH4/zephyr-7b-beta --backend vllm

🏃 Get started

The following provides instructions for how to get started with OpenLLM locally.

Prerequisites

You have installed Python 3.8 (or later) and pip. We highly recommend using a Virtual Environment to prevent package conflicts.

Install OpenLLM

Install OpenLLM by using pip as follows:

pip install openllm

To verify the installation, run:

$ openllm -h

Usage: openllm [OPTIONS] COMMAND [ARGS]...

   ██████╗ ██████╗ ███████╗███╗   ██╗██╗     ██╗     ███╗   ███╗
  ██╔═══██╗██╔══██╗██╔════╝████╗  ██║██║     ██║     ████╗ ████║
  ██║   ██║██████╔╝█████╗  ██╔██╗ ██║██║     ██║     ██╔████╔██║
  ██║   ██║██╔═══╝ ██╔══╝  ██║╚██╗██║██║     ██║     ██║╚██╔╝██║
  ╚██████╔╝██║     ███████╗██║ ╚████║███████╗███████╗██║ ╚═╝ ██║
   ╚═════╝ ╚═╝     ╚══════╝╚═╝  ╚═══╝╚══════╝╚══════╝╚═╝     ╚═╝.

  An open platform for operating large language models in production.
  Fine-tune, serve, deploy, and monitor any LLMs with ease.

Options:
  -v, --version  Show the version and exit.
  -h, --help     Show this message and exit.

Commands:
  build       Package a given models into a BentoLLM.
  import      Setup LLM interactively.
  models      List all supported models.
  prune       Remove all saved models, (and optionally bentos) built with OpenLLM locally.
  query       Query a LLM interactively, from a terminal.
  start       Start a LLMServer for any supported LLM.
  start-grpc  Start a gRPC LLMServer for any supported LLM.

Extensions:
  build-base-container  Base image builder for BentoLLM.
  dive-bentos           Dive into a BentoLLM.
  get-containerfile     Return Containerfile of any given Bento.
  get-prompt            Get the default prompt used by OpenLLM.
  list-bentos           List available bentos built by OpenLLM.
  list-models           This is equivalent to openllm models...
  playground            OpenLLM Playground.

Start a LLM server

OpenLLM allows you to quickly spin up an LLM server using openllm start. For example, to start an OPT server, run the following:

openllm start facebook/opt-1.3b

This starts the server at http://0.0.0.0:3000/. OpenLLM downloads the model to the BentoML local Model Store if they have not been registered before. To view your local models, run bentoml models list.

To interact with the server, you can visit the web UI at http://0.0.0.0:3000/ or send a request using curl. You can also use OpenLLM’s built-in Python client to interact with the server:

import openllm

client = openllm.client.HTTPClient('http://localhost:3000')
client.query('Explain to me the difference between "further" and "farther"')

Alternatively, use the openllm query command to query the model:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'Explain to me the difference between "further" and "farther"'

OpenLLM seamlessly supports many models and their variants. You can specify different variants of the model to be served by providing the --model-id option. For example:

openllm start facebook/opt-2.7b

Note

OpenLLM supports specifying fine-tuning weights and quantized weights for any of the supported models as long as they can be loaded with the model architecture. Use the openllm models command to see the complete list of supported models, their architectures, and their variants.

🧩 Supported models

OpenLLM currently supports the following models. By default, OpenLLM doesn't include dependencies to run all models. The extra model-specific dependencies can be installed with the instructions below.

Baichuan

Quickstart

Note: Baichuan requires to install with:
pip install "openllm[baichuan]"

Run the following command to quickly spin up a Baichuan server:

TRUST_REMOTE_CODE=True openllm start baichuan-inc/baichuan-7b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Baichuan variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Baichuan-compatible models.

Supported models

You can specify any of the following Baichuan models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

TRUST_REMOTE_CODE=True openllm start baichuan-inc/baichuan2-7b-base --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

TRUST_REMOTE_CODE=True openllm start baichuan-inc/baichuan2-7b-base --backend pt

ChatGLM

Quickstart

Note: ChatGLM requires to install with:
pip install "openllm[chatglm]"

Run the following command to quickly spin up a ChatGLM server:

TRUST_REMOTE_CODE=True openllm start thudm/chatglm-6b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any ChatGLM variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more ChatGLM-compatible models.

Supported models

You can specify any of the following ChatGLM models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

TRUST_REMOTE_CODE=True openllm start thudm/chatglm-6b --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

TRUST_REMOTE_CODE=True openllm start thudm/chatglm-6b --backend pt

DollyV2

Quickstart

Run the following command to quickly spin up a DollyV2 server:

openllm start databricks/dolly-v2-3b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any DollyV2 variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more DollyV2-compatible models.

Supported models

You can specify any of the following DollyV2 models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start databricks/dolly-v2-3b --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start databricks/dolly-v2-3b --backend pt

Falcon

Quickstart

Note: Falcon requires to install with:
pip install "openllm[falcon]"

Run the following command to quickly spin up a Falcon server:

openllm start tiiuae/falcon-7b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Falcon variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Falcon-compatible models.

Supported models

You can specify any of the following Falcon models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start tiiuae/falcon-7b --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start tiiuae/falcon-7b --backend pt

FlanT5

Quickstart

Run the following command to quickly spin up a FlanT5 server:

openllm start google/flan-t5-large

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any FlanT5 variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more FlanT5-compatible models.

Supported models

You can specify any of the following FlanT5 models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

PyTorch:

openllm start google/flan-t5-small --backend pt

GPTNeoX

Quickstart

Run the following command to quickly spin up a GPTNeoX server:

openllm start eleutherai/gpt-neox-20b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any GPTNeoX variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more GPTNeoX-compatible models.

Supported models

You can specify any of the following GPTNeoX models via openllm start:

eleutherai/gpt-neox-20b

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start eleutherai/gpt-neox-20b --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start eleutherai/gpt-neox-20b --backend pt

Llama

Quickstart

Run the following command to quickly spin up a Llama server:

openllm start NousResearch/llama-2-7b-hf

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Llama variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Llama-compatible models.

Supported models

You can specify any of the following Llama models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start meta-llama/Llama-2-70b-chat-hf --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start meta-llama/Llama-2-70b-chat-hf --backend pt

Mistral

Quickstart

Run the following command to quickly spin up a Mistral server:

openllm start mistralai/Mistral-7B-Instruct-v0.1

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Mistral variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Mistral-compatible models.

Supported models

You can specify any of the following Mistral models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start HuggingFaceH4/zephyr-7b-alpha --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start HuggingFaceH4/zephyr-7b-alpha --backend pt

Mixtral

Quickstart

Run the following command to quickly spin up a Mixtral server:

openllm start mistralai/Mixtral-8x7B-Instruct-v0.1

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Mixtral variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Mixtral-compatible models.

Supported models

You can specify any of the following Mixtral models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start mistralai/Mixtral-8x7B-Instruct-v0.1 --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start mistralai/Mixtral-8x7B-Instruct-v0.1 --backend pt

MPT

Quickstart

Note: MPT requires to install with:
pip install "openllm[mpt]"

Run the following command to quickly spin up a MPT server:

TRUST_REMOTE_CODE=True openllm start mosaicml/mpt-7b-instruct

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any MPT variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more MPT-compatible models.

Supported models

You can specify any of the following MPT models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

TRUST_REMOTE_CODE=True openllm start mosaicml/mpt-7b --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

TRUST_REMOTE_CODE=True openllm start mosaicml/mpt-7b --backend pt

OPT

Quickstart

Run the following command to quickly spin up a OPT server:

openllm start facebook/opt-1.3b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any OPT variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more OPT-compatible models.

Supported models

You can specify any of the following OPT models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start facebook/opt-125m --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start facebook/opt-125m --backend pt

Phi

Quickstart

Run the following command to quickly spin up a Phi server:

TRUST_REMOTE_CODE=True openllm start microsoft/phi-1_5

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Phi variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Phi-compatible models.

Supported models

You can specify any of the following Phi models via openllm start:

microsoft/phi-1_5

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

TRUST_REMOTE_CODE=True openllm start microsoft/phi-1_5 --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

TRUST_REMOTE_CODE=True openllm start microsoft/phi-1_5 --backend pt

Qwen

Quickstart

Note: Qwen requires to install with:
pip install "openllm[qwen]"

Run the following command to quickly spin up a Qwen server:

TRUST_REMOTE_CODE=True openllm start qwen/Qwen-7B-Chat

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Qwen variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Qwen-compatible models.

Supported models

You can specify any of the following Qwen models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

TRUST_REMOTE_CODE=True openllm start qwen/Qwen-7B-Chat --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

TRUST_REMOTE_CODE=True openllm start qwen/Qwen-7B-Chat --backend pt

StableLM

Quickstart

Run the following command to quickly spin up a StableLM server:

openllm start stabilityai/stablelm-tuned-alpha-3b

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any StableLM variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more StableLM-compatible models.

Supported models

You can specify any of the following StableLM models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start stabilityai/stablelm-tuned-alpha-3b --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start stabilityai/stablelm-tuned-alpha-3b --backend pt

StarCoder

Quickstart

Note: StarCoder requires to install with:
pip install "openllm[starcoder]"

Run the following command to quickly spin up a StarCoder server:

openllm start bigcode/starcoder

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any StarCoder variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more StarCoder-compatible models.

Supported models

You can specify any of the following StarCoder models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

openllm start bigcode/starcoder --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

openllm start bigcode/starcoder --backend pt

Quickstart

Run the following command to quickly spin up a Yi server:

TRUST_REMOTE_CODE=True openllm start 01-ai/Yi-6B

In a different terminal, run the following command to interact with the server:

export OPENLLM_ENDPOINT=http://localhost:3000
openllm query 'What are large language models?'

Note: Any Yi variants can be deployed with OpenLLM. Visit the HuggingFace Model Hub to see more Yi-compatible models.

Supported models

You can specify any of the following Yi models via openllm start:

Supported backends

OpenLLM will support vLLM and PyTorch as default backend. By default, it will use vLLM if vLLM is available, otherwise fallback to PyTorch.

Important: We recommend user to explicitly specify --backend to choose the desired backend to run the model. If you have access to a GPU, always use --backend vllm.

vLLM (Recommended):

To install vLLM, run pip install "openllm[vllm]"

TRUST_REMOTE_CODE=True openllm start 01-ai/Yi-6B --backend vllm

Important: Using vLLM requires a GPU that has architecture newer than 8.0 to get the best performance for serving. It is recommended that for all serving usecase in production, you should choose vLLM for serving.

Note: Currently, adapters are yet to be supported with vLLM.

PyTorch:

TRUST_REMOTE_CODE=True openllm start 01-ai/Yi-6B --backend pt

More models will be integrated with OpenLLM and we welcome your contributions if you want to incorporate your custom LLMs into the ecosystem. Check out Adding a New Model Guide to learn more.

💻 Run your model on multiple GPUs

OpenLLM allows you to start your model server on multiple GPUs and specify the number of workers per resource assigned using the --workers-per-resource option. For example, if you have 4 available GPUs, you set the value as one divided by the number as only one instance of the Runner server will be spawned.

openllm start facebook/opt-2.7b --workers-per-resource 0.25

Note

The amount of GPUs required depends on the model size itself. You can use the Model Memory Calculator from Hugging Face to calculate how much vRAM is needed to train and perform big model inference on a model and then plan your GPU strategy based on it.

When using the --workers-per-resource option with the openllm build command, the environment variable is saved into the resulting Bento.

For more information, see Resource scheduling strategy.

🛞 Runtime implementations

Different LLMs may support multiple runtime implementations. Models that have vLLM (vllm) supports will use vLLM by default, otherwise it fallback to use PyTorch (pt).

To specify a specific runtime for your chosen model, use the --backend option. For example:

openllm start meta-llama/Llama-2-7b-chat-hf --backend vllm

Note:

To use the vLLM backend, you need a GPU with at least the Ampere architecture or newer and CUDA version 11.8.
To see the backend options of each model supported by OpenLLM, see the Supported models section or run openllm models.

📐 Quantization

Quantization is a technique to reduce the storage and computation requirements for machine learning models, particularly during inference. By approximating floating-point numbers as integers (quantized values), quantization allows for faster computations, reduced memory footprint, and can make it feasible to deploy large models on resource-constrained devices.

OpenLLM supports the following quantization techniques

PyTorch backend

With PyTorch backend, OpenLLM supports int8, int4, gptq

For using int8 and int4 quantization through bitsandbytes, you can use the following command:

openllm start facebook/opt-6.7b --quantize int8

To run inference with gptq, simply pass --quantize gptq:

openllm start TheBloke/Llama-2-7B-Chat-GPTQ --quantize gptq

Note

In order to run GPTQ, make sure you run pip install "openllm[gptq]" first to install the dependency. From the GPTQ paper, it is recommended to quantized the weights before serving. See AutoGPTQ for more information on GPTQ quantization.

vLLM backend

With vLLM backend, OpenLLM supports awq, squeezellm

To run inference with awq, simply pass --quantize awq:

openllm start TheBloke/zephyr-7B-alpha-AWQ --quantize awq

To run inference with squeezellm, simply pass --quantize squeezellm:

openllm start squeeze-ai-lab/sq-llama-2-7b-w4-s0 --quantize squeezellm --serialization legacy

Important

Since both squeezellm and awq are weight-aware quantization methods, meaning the quantization is done during training, all pre-trained weights needs to get quantized before inference time. Make sure to fine compatible weights on HuggingFace Hub for your model of choice.

🛠️ Serving fine-tuning layers

PEFT, or Parameter-Efficient Fine-Tuning, is a methodology designed to fine-tune pre-trained models more efficiently. Instead of adjusting all model parameters, PEFT focuses on tuning only a subset, reducing computational and storage costs. LoRA (Low-Rank Adaptation) is one of the techniques supported by PEFT. It streamlines fine-tuning by using low-rank decomposition to represent weight updates, thereby drastically reducing the number of trainable parameters.

With OpenLLM, you can take advantage of the fine-tuning feature by serving models with any PEFT-compatible layers using the --adapter-id option. For example:

openllm start facebook/opt-6.7b --adapter-id aarnphm/opt-6-7b-quotes:default

OpenLLM also provides flexibility by supporting adapters from custom file paths:

openllm start facebook/opt-6.7b --adapter-id /path/to/adapters:local_adapter

To use multiple adapters, use the following format:

openllm start facebook/opt-6.7b --adapter-id aarnphm/opt-6.7b-lora:default --adapter-id aarnphm/opt-6.7b-french:french_lora

By default, all adapters will be injected into the models during startup. Adapters can be specified per request via adapter_name:

curl -X 'POST' \
  'http://localhost:3000/v1/generate' \
  -H 'accept: application/json' \
  -H 'Content-Type: application/json' \
  -d '{
  "prompt": "What is the meaning of life?",
  "stop": [
    "philosopher"
  ],
  "llm_config": {
    "max_new_tokens": 256,
    "temperature": 0.75,
    "top_k": 15,
    "top_p": 1
  },
  "adapter_name": "default"
}'

To include this into the Bento, you can specify the --adapter-id option when using the openllm build command:

openllm build facebook/opt-6.7b --adapter-id ...

If you use a relative path for --adapter-id, you need to add --build-ctx.

openllm build facebook/opt-6.7b --adapter-id ./path/to/adapter_id --build-ctx .

Important

Fine-tuning support is still experimental and currently only works with PyTorch backend. vLLM support is coming soon.

🐍 Python SDK

Each LLM can be instantiated with openllm.LLM:

import openllm

llm = openllm.LLM('facebook/opt-2.7b')

The main inference API is the streaming generate_iterator method:

async for generation in llm.generate_iterator('What is the meaning of life?'):
  print(generation.outputs[0].text)

Note

The motivation behind making llm.generate_iterator an async generator is to provide support for Continuous batching with vLLM backend. By having the async endpoints, each prompt will be added correctly to the request queue to process with vLLM backend.

There is also a one-shot generate method:

await llm.generate('What is the meaning of life?')

This method is easy to use for one-shot generation use case, but merely served as an example how to use llm.generate_iterator as it uses generate_iterator under the hood.

Important

If you need to call your code in a synchronous context, you can use asyncio.run that wraps an async function:

import asyncio
async def generate(prompt, **attrs): return await llm.generate(prompt, **attrs)
asyncio.run(generate("The meaning of life is", temperature=0.23))

⚙️ Integrations

OpenLLM is not just a standalone product; it's a building block designed to integrate with other powerful tools easily. We currently offer integration with BentoML, OpenAI's Compatible Endpoints, LlamaIndex, LangChain, and Transformers Agents.

OpenAI Compatible Endpoints

OpenLLM Server can be used as a drop-in replacement for OpenAI's API. Simply specify the base_url to llm-endpoint/v1 and you are good to go:

import openai

client = openai.OpenAI(
  base_url='http://localhost:3000/v1', api_key='na'
)  # Here the server is running on localhost:3000

completions = client.completions.create(
  prompt='Write me a tag line for an ice cream shop.', model=model, max_tokens=64, stream=stream
)

The compatible endpoints supports /completions, /chat/completions, and /models

Note

You can find out OpenAI example clients under the examples folder.

BentoML

OpenLLM LLM can be integrated as a Runner in your BentoML service. Simply call await llm.generate to generate text. Note that llm.generate uses runner under the hood:

import bentoml
import openllm

llm = openllm.LLM('facebook/opt-2.7b')

svc = bentoml.Service(name='llm-opt-service', runners=[llm.runner])


@svc.api(input=bentoml.io.Text(), output=bentoml.io.Text())
async def prompt(input_text: str) -> str:
  generation = await llm.generate(input_text)
  return generation.outputs[0].text

LlamaIndex

To start a local LLM with llama_index, simply use llama_index.llms.openllm.OpenLLM:

import asyncio
from llama_index.llms.openllm import OpenLLM

llm = OpenLLM('HuggingFaceH4/zephyr-7b-alpha')

llm.complete('The meaning of life is')


async def main(prompt, **kwargs):
  async for it in llm.astream_chat(prompt, **kwargs):
    print(it)


asyncio.run(main('The time at San Francisco is'))

If there is a remote LLM Server running elsewhere, then you can use llama_index.llms.openllm.OpenLLMAPI:

from llama_index.llms.openllm import OpenLLMAPI

Note

All synchronous and asynchronous API from llama_index.llms.LLM are supported.

LangChain

To quickly start a local LLM with langchain, simply do the following:

from langchain.llms import OpenLLM

llm = OpenLLM(model_name='llama', model_id='meta-llama/Llama-2-7b-hf')

llm('What is the difference between a duck and a goose? And why there are so many Goose in Canada?')

Important

By default, OpenLLM use safetensors format for saving models. If the model doesn't support safetensors, make sure to pass serialisation="legacy" to use the legacy PyTorch bin format.

langchain.llms.OpenLLM has the capability to interact with remote OpenLLM Server. Given there is an OpenLLM server deployed elsewhere, you can connect to it by specifying its URL:

from langchain.llms import OpenLLM

llm = OpenLLM(server_url='http://44.23.123.1:3000', server_type='http')
llm('What is the difference between a duck and a goose? And why there are so many Goose in Canada?')

To integrate a LangChain agent with BentoML, you can do the following:

llm = OpenLLM(model_id='google/flan-t5-large', embedded=False, serialisation='legacy')
tools = load_tools(['serpapi', 'llm-math'], llm=llm)
agent = initialize_agent(tools, llm, agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION)
svc = bentoml.Service('langchain-openllm', runners=[llm.runner])


@svc.api(input=Text(), output=Text())
def chat(input_text: str):
  return agent.run(input_text)

Note

You can find out more examples under the examples folder.

Transformers Agents

OpenLLM seamlessly integrates with Transformers Agents.

Warning

The Transformers Agent is still at an experimental stage. It is recommended to install OpenLLM with pip install -r nightly-requirements.txt to get the latest API update for HuggingFace agent.

import transformers

agent = transformers.HfAgent('http://localhost:3000/hf/agent')  # URL that runs the OpenLLM server

agent.run('Is the following `text` positive or negative?', text="I don't like how this models is generate inputs")

🚀 Deploying models to production

There are several ways to deploy your LLMs:

🐳 Docker container

Building a Bento: With OpenLLM, you can easily build a Bento for a specific model, like mistralai/Mistral-7B-Instruct-v0.1, using the build command.:
```
openllm build mistralai/Mistral-7B-Instruct-v0.1
```
A Bento, in BentoML, is the unit of distribution. It packages your program's source code, models, files, artefacts, and dependencies.
Containerize your Bento
```
bentoml containerize <name:version>
```
This generates a OCI-compatible docker image that can be deployed anywhere docker runs. For best scalability and reliability of your LLM service in production, we recommend deploy with BentoCloud。

☁️ BentoCloud

Deploy OpenLLM with BentoCloud, the serverless cloud for shipping and scaling AI applications.

Create a BentoCloud account: sign up here for early access

Log into your BentoCloud account:

bentoml cloud login --api-token <your-api-token> --endpoint <bento-cloud-endpoint>

Note

Replace <your-api-token> and <bento-cloud-endpoint> with your specific API token and the BentoCloud endpoint respectively.

Bulding a Bento: With OpenLLM, you can easily build a Bento for a specific model, such as mistralai/Mistral-7B-Instruct-v0.1:
```
openllm build mistralai/Mistral-7B-Instruct-v0.1
```
Pushing a Bento: Push your freshly-built Bento service to BentoCloud via the push command:
```
bentoml push <name:version>
```
Deploying a Bento: Deploy your LLMs to BentoCloud with a single bentoml deployment create command following the deployment instructions.

👥 Community

Engage with like-minded individuals passionate about LLMs, AI, and more on our Discord!

OpenLLM is actively maintained by the BentoML team. Feel free to reach out and join us in our pursuit to make LLMs more accessible and easy to use 👉 Join our Slack community!

🎁 Contributing

We welcome contributions! If you're interested in enhancing OpenLLM's capabilities or have any questions, don't hesitate to reach out in our discord channel.

Checkout our Developer Guide if you wish to contribute to OpenLLM's codebase.

🍇 Telemetry

OpenLLM collects usage data to enhance user experience and improve the product. We only report OpenLLM's internal API calls and ensure maximum privacy by excluding sensitive information. We will never collect user code, model data, or stack traces. For usage tracking, check out the code.

You can opt out of usage tracking by using the --do-not-track CLI option:

openllm [command] --do-not-track

Or by setting the environment variable OPENLLM_DO_NOT_TRACK=True:

export OPENLLM_DO_NOT_TRACK=True

📔 Citation

If you use OpenLLM in your research, we provide a citation to use:

@software{Pham_OpenLLM_Operating_LLMs_2023,
author = {Pham, Aaron and Yang, Chaoyu and Sheng, Sean and  Zhao, Shenyang and Lee, Sauyon and Jiang, Bo and Dong, Fog and Guan, Xipeng and Ming, Frost},
license = {Apache-2.0},
month = jun,
title = {{OpenLLM: Operating LLMs in production}},
url = {https://github.com/bentoml/OpenLLM},
year = {2023}
}

tahslim/OpenLLM

🦾 OpenLLM

📖 Introduction

💾 TL/DR

Jupyter Notebooks

Docker

🏃 Get started

Prerequisites

Install OpenLLM

Start a LLM server

🧩 Supported models

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

Quickstart

Supported models

Supported backends

💻 Run your model on multiple GPUs

🛞 Runtime implementations

📐 Quantization

PyTorch backend

vLLM backend

🛠️ Serving fine-tuning layers

🐍 Python SDK

⚙️ Integrations

OpenAI Compatible Endpoints

BentoML

LlamaIndex

LangChain

Transformers Agents

🚀 Deploying models to production

🐳 Docker container

☁️ BentoCloud

👥 Community

🎁 Contributing

🍇 Telemetry

📔 Citation