**LocalAI** is a drop-in replacement REST API that's compatible with OpenAI API specifications for local inferencing. It allows you to run models locally or on-prem with consumer grade hardware, supporting multiple model families that are compatible with the ggml format.
For a list of the supported model families, please see [the model compatibility table below](https://github.com/go-skynet/LocalAI#model-compatibility-table).
- NO GPU required. NO Internet access is required either. Optional, GPU Acceleration is available in `llama.cpp`-compatible LLMs. [See building instructions](https://github.com/go-skynet/LocalAI#cublas).
LocalAI is a community-driven project, focused on making the AI accessible to anyone. Any contribution, feedback and PR is welcome! It was initially created by [mudler](https://github.com/mudler/) at the [SpectroCloud OSS Office](https://github.com/spectrocloud).
LocalAI is an API written in Go that serves as an OpenAI shim, enabling software already developed with OpenAI SDKs to seamlessly integrate with LocalAI. It can be effortlessly implemented as a substitute, even on consumer-grade hardware. This capability is achieved by employing various C++ backends, including [ggml](https://github.com/ggerganov/ggml), to perform inference on LLMs using both CPU and, if desired, GPU.
LocalAI uses C++ bindings for optimizing speed. It is based on [llama.cpp](https://github.com/ggerganov/llama.cpp), [gpt4all](https://github.com/nomic-ai/gpt4all), [rwkv.cpp](https://github.com/saharNooby/rwkv.cpp), [ggml](https://github.com/ggerganov/ggml), [whisper.cpp](https://github.com/ggerganov/whisper.cpp) for audio transcriptions, [bert.cpp](https://github.com/skeskinen/bert.cpp) for embedding and [StableDiffusion-NCN](https://github.com/EdVince/Stable-Diffusion-NCNN) for image generation. See [the model compatibility table](https://github.com/go-skynet/LocalAI#model-compatibility-table) to learn about all the components of LocalAI.
- 17-05-2023: __v1.12.0__ released! 🔥🔥 Minor fixes, plus CUDA (https://github.com/go-skynet/LocalAI/pull/258) support for `llama.cpp`-compatible models and image generation (https://github.com/go-skynet/LocalAI/pull/272).
- 16-05-2023: 🔥🔥🔥 Experimental support for CUDA (https://github.com/go-skynet/LocalAI/pull/258) in the `llama.cpp` backend and Stable diffusion CPU image generation (https://github.com/go-skynet/LocalAI/pull/272) in `master`.
- 13-05-2023: __v1.11.0__ released! 🔥 Updated `llama.cpp` bindings: This update includes a breaking change in the model files ( https://github.com/ggerganov/llama.cpp/pull/1405 ) - old models should still work with the `gpt4all-llama` backend.
- 12-05-2023: __v1.10.0__ released! 🔥🔥 Updated `gpt4all` bindings. Added support for GPTNeox (experimental), RedPajama (experimental), Starcoder (experimental), Replit (experimental), MosaicML MPT. Also now `embeddings` endpoint supports tokens arrays. See the [langchain-chroma](https://github.com/go-skynet/LocalAI/tree/master/examples/langchain-chroma) example! Note - this update does NOT include https://github.com/ggerganov/llama.cpp/pull/1405 which makes models incompatible.
- 11-05-2023: __v1.9.0__ released! 🔥 Important whisper updates ( https://github.com/go-skynet/LocalAI/pull/233 https://github.com/go-skynet/LocalAI/pull/229 ) and extended gpt4all model families support ( https://github.com/go-skynet/LocalAI/pull/232 ). Redpajama/dolly experimental ( https://github.com/go-skynet/LocalAI/pull/214 )
- 10-05-2023: __v1.8.0__ released! 🔥 Added support for fast and accurate embeddings with `bert.cpp` ( https://github.com/go-skynet/LocalAI/pull/222 )
- [Question Answering on Documents locally with LangChain, LocalAI, Chroma, and GPT4All](https://mudler.pm/posts/localai-question-answering/) by Ettore Di Giacinto
- [Tutorial to use k8sgpt with LocalAI](https://medium.com/@tyler_97636/k8sgpt-localai-unlock-kubernetes-superpowers-for-free-584790de9b65) - excellent usecase for localAI, using AI to analyse Kubernetes clusters. by Tyller Gillson
- If you have technological skills and want to contribute to development, have a look at the open issues. If you are new you can have a look at the [good-first-issue](https://github.com/go-skynet/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3A%22good+first+issue%22) and [help-wanted](https://github.com/go-skynet/LocalAI/issues?q=is%3Aissue+is%3Aopen+label%3A%22help+wanted%22) labels.
- If you don't have technological skills you can still help improving documentation or add examples or share your user-stories with our community, any help and contribution is welcome!
It is compatible with the models supported by [llama.cpp](https://github.com/ggerganov/llama.cpp) supports also [GPT4ALL-J](https://github.com/nomic-ai/gpt4all) and [cerebras-GPT with ggml](https://huggingface.co/lxe/Cerebras-GPT-2.7B-Alpaca-SP-ggml).
Note: You might need to convert some models from older models to the new format, for indications, see [the README in llama.cpp](https://github.com/ggerganov/llama.cpp#using-gpt4all) for instance to run `gpt4all`.
Note: rwkv models needs to specify the backend `rwkv` in the YAML config files and have an associated tokenizer along that needs to be provided with it:
It should also be compatible with StableLM and GPTNeoX ggml models (untested).
### Hardware requirements
Depending on the model you are attempting to run might need more RAM or CPU resources. Check out also [here](https://github.com/ggerganov/llama.cpp#memorydisk-requirements) for `ggml` based backends. `rwkv` is less expensive on resources.
| [llama](https://github.com/ggerganov/llama.cpp) ([binding](https://github.com/go-skynet/go-llama.cpp)) | Vicuna, Alpaca, LLaMa | yes | no | yes (doesn't seem to be accurate) | yes |
| [gpt4all-llama](https://github.com/nomic-ai/gpt4all) | Vicuna, Alpaca, LLaMa | yes | no | no | yes |
| [gpt4all-mpt](https://github.com/nomic-ai/gpt4all) | MPT | yes | no | no | yes |
| [gpt4all-j](https://github.com/nomic-ai/gpt4all) | GPT4ALL-J | yes | no | no | yes |
| [gpt2](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | GPT/NeoX, Cerebras | yes | no | no | no |
| [dolly](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | Dolly | yes | no | no | no |
| [redpajama](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | RedPajama | yes | no | no | no |
| [stableLM](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | StableLM GPT/NeoX | yes | no | no | no |
| [replit](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | Replit | yes | no | no | no |
| [gptneox](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | GPT NeoX | yes | no | no | no |
| [starcoder](https://github.com/ggerganov/ggml) ([binding](https://github.com/go-skynet/go-gpt2.cpp)) | Starcoder | yes | no | no | no |
| [bloomz](https://github.com/NouamaneTazi/bloomz.cpp) ([binding](https://github.com/go-skynet/bloomz.cpp)) | Bloom | yes | no | no | no |
| [rwkv](https://github.com/saharNooby/rwkv.cpp) ([binding](https://github.com/donomii/go-rw)) | rwkv | yes | no | no | yes |
| [bert](https://github.com/skeskinen/bert.cpp) ([binding](https://github.com/go-skynet/go-bert.cpp) | bert | no | no | yes | no |
| [whisper](https://github.com/ggerganov/whisper.cpp) | whisper | no | Audio | no | no |
| [stablediffusion](https://github.com/EdVince/Stable-Diffusion-NCNN) ([binding](https://github.com/mudler/go-stable-diffusion)) | stablediffusion | no | Image | no | no |
> `LocalAI` comes by default as a container image. You can check out all the available images with corresponding tags [here](https://quay.io/repository/go-skynet/local-ai?tab=tags&tag=latest).
The easiest way to run LocalAI is by using `docker-compose` (to build locally, see [building LocalAI](https://github.com/go-skynet/LocalAI/tree/master#setup)):
Instead of installing models manually, you can use the LocalAI API endpoints and a model definition to install programmatically via API models in runtime.
<details>
A curated collection of model files is in the [model-gallery](https://github.com/go-skynet/model-gallery) (work in progress!).
To install for example `gpt4all-j`, you can send a POST call to the `/models/apply` endpoint with the model definition url (`url`) and the name of the model should have in LocalAI (`name`, optional):
To see other examples on how to integrate with other projects for instance for question answering or for using it with chatbot-ui, see: [examples](https://github.com/go-skynet/LocalAI/tree/master/examples/).
The API doesn't inject a default prompt for talking to the model. You have to use a prompt similar to what's described in the standford-alpaca docs: https://github.com/tatsu-lab/stanford_alpaca#data-release.
You can use a default template for every model present in your model path, by creating a corresponding file with the `.tmpl` suffix next to your model. For instance, if the model is called `foo.bin`, you can create a sibling file, `foo.bin.tmpl` which will be used as a default prompt and can be used with alpaca:
See the [prompt-templates](https://github.com/go-skynet/LocalAI/tree/master/prompt-templates) directory in this repository for templates for some of the most popular models.
For the edit endpoint, an example template for alpaca-based models can be:
```yaml
Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction:
{{.Instruction}}
### Input:
{{.Input}}
### Response:
```
</details>
### CLI
You can control LocalAI with command line arguments, to specify a binding address, or the number of threads.
Currently LocalAI comes as a container image and can be used with docker or a container engine of choice. You can check out all the available images with corresponding tags [here](https://quay.io/repository/go-skynet/local-ai?tab=tags&tag=latest).
Note: the binary inside the image is rebuild at the start of the container to enable CPU optimizations for the execution environment, you can set the environment variable `REBUILD` to `false` to prevent this behavior.
For example, to generate a chat completion, you can send a POST request to the `/v1/chat/completions` endpoint with the instruction as the request body:
Note: embeddings is supported only with `llama.cpp` compatible models and `bert` models. bert is more performant and available independently of the LLM model.
Note: requires ffmpeg in the container image, which is currently not shipped due to licensing issues. We will prepare separated images with ffmpeg. (stay tuned!)
Download one of the models from https://huggingface.co/ggerganov/whisper.cpp/tree/main in the `models` folder, and create a YAML file for your model:
{"text":"My fellow Americans, this day has brought terrible news and great sadness to our country.At nine o'clock this morning, Mission Control in Houston lost contact with our Space ShuttleColumbia.A short time later, debris was seen falling from the skies above Texas.The Columbia's lost.There are no survivors.One board was a crew of seven.Colonel Rick Husband, Lieutenant Colonel Michael Anderson, Commander Laurel Clark, Captain DavidBrown, Commander William McCool, Dr. Kultna Shavla, and Elon Ramon, a colonel in the IsraeliAir Force.These men and women assumed great risk in the service to all humanity.In an age when spaceflight has come to seem almost routine, it is easy to overlook thedangers of travel by rocket and the difficulties of navigating the fierce outer atmosphere ofthe Earth.These astronauts knew the dangers, and they faced them willingly, knowing they had a highand noble purpose in life.Because of their courage and daring and idealism, we will miss them all the more.All Americans today are thinking as well of the families of these men and women who havebeen given this sudden shock and grief.You're not alone.Our entire nation agrees with you, and those you loved will always have the respect andgratitude of this country.The cause in which they died will continue.Mankind has led into the darkness beyond our world by the inspiration of discovery andthe longing to understand.Our journey into space will go on.In the skies today, we saw destruction and tragedy.As farther than we can see, there is comfort and hope.In the words of the prophet Isaiah, \"Lift your eyes and look to the heavens who createdall these, he who brings out the starry hosts one by one and calls them each by name.\"Because of his great power and mighty strength, not one of them is missing.The same creator who names the stars also knows the names of the seven souls we mourntoday.The crew of the shuttle Columbia did not return safely to Earth yet we can pray that all aresafely home.May God bless the grieving families and may God continue to bless America.[BLANK_AUDIO]"}
Note: image generator supports images up to 512x512. You can use other tools however to upscale the image, for instance: https://github.com/upscayl/upscayl.
Besides the OpenAI endpoints, there are additional LocalAI-only API endpoints.
### Applying a model - `/models/apply`
This endpoint can be used to install a model in runtime.
<details>
LocalAI will create a batch process that downloads the required files from a model definition and automatically reload itself to include the new model.
Here are answers to some of the most common questions.
### How do I get models?
<details>
Most ggml-based models should work, but newer models may require additions to the API. If a model doesn't work, please feel free to open up issues. However, be cautious about downloading models from the internet and directly onto your machine, as there may be security vulnerabilities in lama.cpp or ggml that could be maliciously exploited. Some models can be found on Hugging Face: https://huggingface.co/models?search=ggml, or models from gpt4all should also work: https://github.com/nomic-ai/gpt4all.
</details>
### What's the difference with Serge, or XXX?
<details>
LocalAI is a multi-model solution that doesn't focus on a specific model type (e.g., llama.cpp or alpaca.cpp), and it handles all of these internally for faster inference, easy to set up locally and deploy to Kubernetes.
</details>
### Can I use it with a Discord bot, or XXX?
<details>
Yes! If the client uses OpenAI and supports setting a different base URL to send requests to, you can use the LocalAI endpoint. This allows to use this with every application that was supposed to work with OpenAI, but without changing the application!
There is the availability of localai-webui and chatbot-ui in the examples section and can be setup as per the instructions. However as LocalAI is an API you can already plug it into existing projects that provides are UI interfaces to OpenAI's APIs. There are several already on github, and should be compatible with LocalAI already (as it mimics the OpenAI API)
LocalAI is a community-driven project. It was initially created by [Ettore Di Giacinto](https://github.com/mudler/) at the [SpectroCloud OSS Office](https://github.com/spectrocloud).
