chat-ui / README.md
nsarrazin's picture
nsarrazin HF staff
Add support for cohere endpoints (#976)
a99cca3 unverified
|
raw
history blame
28.9 kB
metadata
title: chat-ui
emoji: 🔥
colorFrom: purple
colorTo: purple
sdk: docker
pinned: false
license: apache-2.0
base_path: /chat
app_port: 3000
failure_strategy: rollback

Chat UI

Chat UI repository thumbnail

A chat interface using open source models, eg OpenAssistant or Llama. It is a SvelteKit app and it powers the HuggingChat app on hf.co/chat.

  1. No Setup Deploy
  2. Setup
  3. Launch
  4. Web Search
  5. Text Embedding Models
  6. Extra parameters
  7. Deploying to a HF Space
  8. Building

No Setup Deploy

If you don't want to configure, setup, and launch your own Chat UI yourself, you can use this option as a fast deploy alternative.

You can deploy your own customized Chat UI instance with any supported LLM of your choice on Hugging Face Spaces. To do so, use the chat-ui template available here.

Set HF_TOKEN in Space secrets to deploy a model with gated access or a model in a private repository. It's also compatible with Inference for PROs curated list of powerful models with higher rate limits. Make sure to create your personal token first in your User Access Tokens settings.

Read the full tutorial here.

Setup

The default config for Chat UI is stored in the .env file. You will need to override some values to get Chat UI to run locally. This is done in .env.local.

Start by creating a .env.local file in the root of the repository. The bare minimum config you need to get Chat UI to run locally is the following:

MONGODB_URL=<the URL to your MongoDB instance>
HF_TOKEN=<your access token>

Database

The chat history is stored in a MongoDB instance, and having a DB instance available is needed for Chat UI to work.

You can use a local MongoDB instance. The easiest way is to spin one up using docker:

docker run -d -p 27017:27017 --name mongo-chatui mongo:latest

In which case the url of your DB will be MONGODB_URL=mongodb://localhost:27017.

Alternatively, you can use a free MongoDB Atlas instance for this, Chat UI should fit comfortably within their free tier. After which you can set the MONGODB_URL variable in .env.local to match your instance.

Hugging Face Access Token

If you use a remote inference endpoint, you will need a Hugging Face access token to run Chat UI locally. You can get one from your Hugging Face profile.

Launch

After you're done with the .env.local file you can run Chat UI locally with:

npm install
npm run dev

Web Search

Chat UI features a powerful Web Search feature. It works by:

  1. Generating an appropriate search query from the user prompt.
  2. Performing web search and extracting content from webpages.
  3. Creating embeddings from texts using a text embedding model.
  4. From these embeddings, find the ones that are closest to the user query using a vector similarity search. Specifically, we use inner product distance.
  5. Get the corresponding texts to those closest embeddings and perform Retrieval-Augmented Generation (i.e. expand user prompt by adding those texts so that an LLM can use this information).

Text Embedding Models

By default (for backward compatibility), when TEXT_EMBEDDING_MODELS environment variable is not defined, transformers.js embedding models will be used for embedding tasks, specifically, Xenova/gte-small model.

You can customize the embedding model by setting TEXT_EMBEDDING_MODELS in your .env.local file. For example:

TEXT_EMBEDDING_MODELS = `[
  {
    "name": "Xenova/gte-small",
    "displayName": "Xenova/gte-small",
    "description": "locally running embedding",
    "chunkCharLength": 512,
    "endpoints": [
      {"type": "transformersjs"}
    ]
  },
  {
    "name": "intfloat/e5-base-v2",
    "displayName": "intfloat/e5-base-v2",
    "description": "hosted embedding model",
    "chunkCharLength": 768,
    "preQuery": "query: ", # See https://huggingface.co/intfloat/e5-base-v2#faq
    "prePassage": "passage: ", # See https://huggingface.co/intfloat/e5-base-v2#faq
    "endpoints": [
      {
        "type": "tei",
        "url": "http://127.0.0.1:8080/",
        "authorization": "TOKEN_TYPE TOKEN" // optional authorization field. Example: "Basic VVNFUjpQQVNT"
      }
    ]
  }
]`

The required fields are name, chunkCharLength and endpoints. Supported text embedding backends are: transformers.js, TEI and OpenAI. transformers.js models run locally as part of chat-ui, whereas TEI models run in a different environment & accessed through an API endpoint. openai models are accessed through the OpenAI API.

When more than one embedding models are supplied in .env.local file, the first will be used by default, and the others will only be used on LLM's which configured embeddingModel to the name of the model.

Extra parameters

OpenID connect

The login feature is disabled by default and users are attributed a unique ID based on their browser. But if you want to use OpenID to authenticate your users, you can add the following to your .env.local file:

OPENID_CONFIG=`{
  PROVIDER_URL: "<your OIDC issuer>",
  CLIENT_ID: "<your OIDC client ID>",
  CLIENT_SECRET: "<your OIDC client secret>",
  SCOPES: "openid profile",
  TOLERANCE: // optional
  RESOURCE: // optional
}`

These variables will enable the openID sign-in modal for users.

Theming

You can use a few environment variables to customize the look and feel of chat-ui. These are by default:

PUBLIC_APP_NAME=ChatUI
PUBLIC_APP_ASSETS=chatui
PUBLIC_APP_COLOR=blue
PUBLIC_APP_DESCRIPTION="Making the community's best AI chat models available to everyone."
PUBLIC_APP_DATA_SHARING=
PUBLIC_APP_DISCLAIMER=
  • PUBLIC_APP_NAME The name used as a title throughout the app.
  • PUBLIC_APP_ASSETS Is used to find logos & favicons in static/$PUBLIC_APP_ASSETS, current options are chatui and huggingchat.
  • PUBLIC_APP_COLOR Can be any of the tailwind colors.
  • PUBLIC_APP_DATA_SHARING Can be set to 1 to add a toggle in the user settings that lets your users opt-in to data sharing with models creator.
  • PUBLIC_APP_DISCLAIMER If set to 1, we show a disclaimer about generated outputs on login.

Web Search config

You can enable the web search through an API by adding YDC_API_KEY (docs.you.com) or SERPER_API_KEY (serper.dev) or SERPAPI_KEY (serpapi.com) or SERPSTACK_API_KEY (serpstack.com) to your .env.local.

You can also simply enable the local google websearch by setting USE_LOCAL_WEBSEARCH=true in your .env.local or specify a SearXNG instance by adding the query URL to SEARXNG_QUERY_URL.

Custom models

You can customize the parameters passed to the model or even use a new model by updating the MODELS variable in your .env.local. The default one can be found in .env and looks like this :

MODELS=`[
  {
    "name": "mistralai/Mistral-7B-Instruct-v0.2",
    "displayName": "mistralai/Mistral-7B-Instruct-v0.2",
    "description": "Mistral 7B is a new Apache 2.0 model, released by Mistral AI that outperforms Llama2 13B in benchmarks.",
    "websiteUrl": "https://mistral.ai/news/announcing-mistral-7b/",
    "preprompt": "",
    "chatPromptTemplate" : "<s>{{#each messages}}{{#ifUser}}[INST] {{#if @first}}{{#if @root.preprompt}}{{@root.preprompt}}\n{{/if}}{{/if}}{{content}} [/INST]{{/ifUser}}{{#ifAssistant}}{{content}}</s>{{/ifAssistant}}{{/each}}",
    "parameters": {
      "temperature": 0.3,
      "top_p": 0.95,
      "repetition_penalty": 1.2,
      "top_k": 50,
      "truncate": 3072,
      "max_new_tokens": 1024,
      "stop": ["</s>"]
    },
    "promptExamples": [
      {
        "title": "Write an email from bullet list",
        "prompt": "As a restaurant owner, write a professional email to the supplier to get these products every week: \n\n- Wine (x10)\n- Eggs (x24)\n- Bread (x12)"
      }, {
        "title": "Code a snake game",
        "prompt": "Code a basic snake game in python, give explanations for each step."
      }, {
        "title": "Assist in a task",
        "prompt": "How do I make a delicious lemon cheesecake?"
      }
    ]
  }
]`

You can change things like the parameters, or customize the preprompt to better suit your needs. You can also add more models by adding more objects to the array, with different preprompts for example.

chatPromptTemplate

When querying the model for a chat response, the chatPromptTemplate template is used. messages is an array of chat messages, it has the format [{ content: string }, ...]. To identify if a message is a user message or an assistant message the ifUser and ifAssistant block helpers can be used.

The following is the default chatPromptTemplate, although newlines and indentiation have been added for readability. You can find the prompts used in production for HuggingChat here.

{{preprompt}}
{{#each messages}}
  {{#ifUser}}{{@root.userMessageToken}}{{content}}{{@root.userMessageEndToken}}{{/ifUser}}
  {{#ifAssistant}}{{@root.assistantMessageToken}}{{content}}{{@root.assistantMessageEndToken}}{{/ifAssistant}}
{{/each}}
{{assistantMessageToken}}

Multi modal model

We currently only support IDEFICS as a multimodal model, hosted on TGI. You can enable it by using the following config (if you have a PRO HF Api token):

    {
      "name": "HuggingFaceM4/idefics-80b-instruct",
      "multimodal" : true,
      "description": "IDEFICS is the new multimodal model by Hugging Face.",
      "preprompt": "",
      "chatPromptTemplate" : "{{#each messages}}{{#ifUser}}User: {{content}}{{/ifUser}}<end_of_utterance>\nAssistant: {{#ifAssistant}}{{content}}\n{{/ifAssistant}}{{/each}}",
      "parameters": {
        "temperature": 0.1,
        "top_p": 0.95,
        "repetition_penalty": 1.2,
        "top_k": 12,
        "truncate": 1000,
        "max_new_tokens": 1024,
        "stop": ["<end_of_utterance>", "User:", "\nUser:"]
      }
    }

Running your own models using a custom endpoint

If you want to, instead of hitting models on the Hugging Face Inference API, you can run your own models locally.

A good option is to hit a text-generation-inference endpoint. This is what is done in the official Chat UI Spaces Docker template for instance: both this app and a text-generation-inference server run inside the same container.

To do this, you can add your own endpoints to the MODELS variable in .env.local, by adding an "endpoints" key for each model in MODELS.

{
// rest of the model config here
"endpoints": [{
  "type" : "tgi",
  "url": "https://HOST:PORT",
  }]
}

If endpoints are left unspecified, ChatUI will look for the model on the hosted Hugging Face inference API using the model name.

OpenAI API compatible models

Chat UI can be used with any API server that supports OpenAI API compatibility, for example text-generation-webui, LocalAI, FastChat, llama-cpp-python, and ialacol.

The following example config makes Chat UI works with text-generation-webui, the endpoint.baseUrl is the url of the OpenAI API compatible server, this overrides the baseUrl to be used by OpenAI instance. The endpoint.completion determine which endpoint to be used, default is chat_completions which uses v1/chat/completions, change to endpoint.completion to completions to use the v1/completions endpoint.

MODELS=`[
  {
    "name": "text-generation-webui",
    "id": "text-generation-webui",
    "parameters": {
      "temperature": 0.9,
      "top_p": 0.95,
      "repetition_penalty": 1.2,
      "top_k": 50,
      "truncate": 1000,
      "max_new_tokens": 1024,
      "stop": []
    },
    "endpoints": [{
      "type" : "openai",
      "baseURL": "http://localhost:8000/v1"
    }]
  }
]`

The openai type includes official OpenAI models. You can add, for example, GPT4/GPT3.5 as a "openai" model:

OPENAI_API_KEY=#your openai api key here
MODELS=`[{
      "name": "gpt-4",
      "displayName": "GPT 4",
      "endpoints" : [{
        "type": "openai"
      }]
},
      {
      "name": "gpt-3.5-turbo",
      "displayName": "GPT 3.5 Turbo",
      "endpoints" : [{
        "type": "openai"
      }]
}]`

You may also consume any model provider that provides compatible OpenAI API endpoint. For example, you may self-host Portkey gateway and experiment with Claude or GPTs offered by Azure OpenAI. Example for Claude from Anthropic:

MODELS=`[{
  "name": "claude-2.1",
  "displayName": "Claude 2.1",
  "description": "Anthropic has been founded by former OpenAI researchers...",
  "parameters": {
      "temperature": 0.5,
      "max_new_tokens": 4096,
  },
  "endpoints": [
      {
          "type": "openai",
          "baseURL": "https://gateway.example.com/v1",
          "defaultHeaders": {
              "x-portkey-config": '{"provider":"anthropic","api_key":"sk-ant-abc...xyz"}'
          }
      }
  ]
}]`

Example for GPT 4 deployed on Azure OpenAI:

MODELS=`[{
  "id": "gpt-4-1106-preview",
  "name": "gpt-4-1106-preview",
  "displayName": "gpt-4-1106-preview",
  "parameters": {
      "temperature": 0.5,
      "max_new_tokens": 4096,
  },
  "endpoints": [
      {
          "type": "openai",
          "baseURL": "https://{resource-name}.openai.azure.com/openai/deployments/{deployment-id}",
          "defaultHeaders": {
              "api-key": "{api-key}"
          },
          "defaultQuery": {
              "api-version": "2023-05-15"
          }
      }
  ]
}]`

Or try Mistral from Deepinfra:

Note, apiKey can either be set custom per endpoint, or globally using OPENAI_API_KEY variable.

MODELS=`[{
  "name": "mistral-7b",
  "displayName": "Mistral 7B",
  "description": "A 7B dense Transformer, fast-deployed and easily customisable. Small, yet powerful for a variety of use cases. Supports English and code, and a 8k context window.",
  "parameters": {
      "temperature": 0.5,
      "max_new_tokens": 4096,
  },
  "endpoints": [
      {
          "type": "openai",
          "baseURL": "https://api.deepinfra.com/v1/openai",
          "apiKey": "abc...xyz"
      }
  ]
}]`
Llama.cpp API server

chat-ui also supports the llama.cpp API server directly without the need for an adapter. You can do this using the llamacpp endpoint type.

If you want to run chat-ui with llama.cpp, you can do the following, using Zephyr as an example model:

  1. Get the weights from the hub
  2. Run the server with the following command: ./server -m models/zephyr-7b-beta.Q4_K_M.gguf -c 2048 -np 3
  3. Add the following to your .env.local:
MODELS=`[
  {
      "name": "Local Zephyr",
      "chatPromptTemplate": "<|system|>\n{{preprompt}}</s>\n{{#each messages}}{{#ifUser}}<|user|>\n{{content}}</s>\n<|assistant|>\n{{/ifUser}}{{#ifAssistant}}{{content}}</s>\n{{/ifAssistant}}{{/each}}",
      "parameters": {
        "temperature": 0.1,
        "top_p": 0.95,
        "repetition_penalty": 1.2,
        "top_k": 50,
        "truncate": 1000,
        "max_new_tokens": 2048,
        "stop": ["</s>"]
      },
      "endpoints": [
        {
         "url": "http://127.0.0.1:8080",
         "type": "llamacpp"
        }
      ]
  }
]`

Start chat-ui with npm run dev and you should be able to chat with Zephyr locally.

Ollama

We also support the Ollama inference server. Spin up a model with

ollama run mistral

Then specify the endpoints like so:

MODELS=`[
  {
      "name": "Ollama Mistral",
      "chatPromptTemplate": "<s>{{#each messages}}{{#ifUser}}[INST] {{#if @first}}{{#if @root.preprompt}}{{@root.preprompt}}\n{{/if}}{{/if}} {{content}} [/INST]{{/ifUser}}{{#ifAssistant}}{{content}}</s> {{/ifAssistant}}{{/each}}",
      "parameters": {
        "temperature": 0.1,
        "top_p": 0.95,
        "repetition_penalty": 1.2,
        "top_k": 50,
        "truncate": 3072,
        "max_new_tokens": 1024,
        "stop": ["</s>"]
      },
      "endpoints": [
        {
         "type": "ollama",
         "url" : "http://127.0.0.1:11434",
         "ollamaName" : "mistral"
        }
      ]
  }
]`

Anthropic

We also support Anthropic models through the official SDK. You may provide your API key via the ANTHROPIC_API_KEY env variable, or alternatively, through the endpoints.apiKey as per the following example.

MODELS=`[
  {
      "name": "claude-3-sonnet-20240229",
      "displayName": "Claude 3 Sonnet",
      "description": "Ideal balance of intelligence and speed",
      "parameters": {
        "max_new_tokens": 4096,
      },
      "endpoints": [
        {
          "type": "anthropic",
          // optionals
          "apiKey": "sk-ant-...",
          "baseURL": "https://api.anthropic.com",
          defaultHeaders: {},
          defaultQuery: {}
        }
      ]
  },
  {
      "name": "claude-3-opus-20240229",
      "displayName": "Claude 3 Opus",
      "description": "Most powerful model for highly complex tasks",
      "parameters": {
         "max_new_tokens": 4096
      },
      "endpoints": [
        {
          "type": "anthropic",
          // optionals
          "apiKey": "sk-ant-...",
          "baseURL": "https://api.anthropic.com",
          defaultHeaders: {},
          defaultQuery: {}
        }
      ]
  }
]`

Amazon

You can also specify your Amazon SageMaker instance as an endpoint for chat-ui. The config goes like this:

"endpoints": [
    {
      "type" : "aws",
      "service" : "sagemaker"
      "url": "",
      "accessKey": "",
      "secretKey" : "",
      "sessionToken": "",
      "region": "",

      "weight": 1
    }
]

You can also set "service" : "lambda" to use a lambda instance.

You can get the accessKey and secretKey from your AWS user, under programmatic access.

Cloudflare Workers AI

You can also use Cloudflare Workers AI to run your own models with serverless inference.

You will need to have a Cloudflare account, then get your account ID as well as your API token for Workers AI.

You can either specify them directly in your .env.local using the CLOUDFLARE_ACCOUNT_ID and CLOUDFLARE_API_TOKEN variables, or you can set them directly in the endpoint config.

You can find the list of models available on Cloudflare here.

  {
  "name" : "nousresearch/hermes-2-pro-mistral-7b",
  "tokenizer": "nousresearch/hermes-2-pro-mistral-7b",
  "parameters": {
    "stop": ["<|im_end|>"]
  },
  "endpoints" : [
    {
      "type" : "cloudflare"
      <!-- optionally specify these
      "accountId": "your-account-id",
      "authToken": "your-api-token"
      -->
    }
  ]
}

Cloudlare Workers AI currently do not support custom sampling parameters like temperature, top_p, etc.

Cohere

You can also use Cohere to run their models directly from chat-ui. You will need to have a Cohere account, then get your API token. You can either specify it directly in your .env.local using the COHERE_API_TOKEN variable, or you can set it in the endpoint config.

Here is an example of a Cohere model config. You can set which model you want to use by setting the id field to the model name.

  {
    "name" : "CohereForAI/c4ai-command-r-v01",
    "id": "command-r",
    "description": "C4AI Command-R is a research release of a 35 billion parameter highly performant generative model",
    "endpoints": [
      {
        "type": "cohere",
        <!-- optionally specify these, or use COHERE_API_TOKEN
        "apiKey": "your-api-token"
        -->
      }
    ]
  }
Google Vertex models

Chat UI can connect to the google Vertex API endpoints (List of supported models).

To enable:

  1. Select or create a Google Cloud project.
  2. Enable billing for your project.
  3. Enable the Vertex AI API.
  4. Set up authentication with a service account so you can access the API from your local workstation.

The service account credentials file can be imported as an environmental variable:

    GOOGLE_APPLICATION_CREDENTIALS = clientid.json

Make sure docker has access to the file. Afterwards Google Vertex endpoints can be configured as following:

MODELS=`[
//...
    {
       "name": "gemini-1.0-pro", //model-name
       "displayName": "Vertex Gemini Pro 1.0",
       "location": "europe-west3",
       "apiEndpoint": "", //alternative api endpoint url
       "endpoints" : [{
         "type": "vertex"
       }]
     },
]`

Custom endpoint authorization

Basic and Bearer

Custom endpoints may require authorization, depending on how you configure them. Authentication will usually be set either with Basic or Bearer.

For Basic we will need to generate a base64 encoding of the username and password.

echo -n "USER:PASS" | base64

VVNFUjpQQVNT

For Bearer you can use a token, which can be grabbed from here.

You can then add the generated information and the authorization parameter to your .env.local.

"endpoints": [
  {
    "url": "https://HOST:PORT",
    "authorization": "Basic VVNFUjpQQVNT",
  }
]

Please note that if HF_TOKEN is also set or not empty, it will take precedence.

Models hosted on multiple custom endpoints

If the model being hosted will be available on multiple servers/instances add the weight parameter to your .env.local. The weight will be used to determine the probability of requesting a particular endpoint.

"endpoints": [
  {
    "url": "https://HOST:PORT",
    "weight": 1
  },
  {
    "url": "https://HOST:PORT",
    "weight": 2
  }
  ...
]

Client Certificate Authentication (mTLS)

Custom endpoints may require client certificate authentication, depending on how you configure them. To enable mTLS between Chat UI and your custom endpoint, you will need to set the USE_CLIENT_CERTIFICATE to true, and add the CERT_PATH and KEY_PATH parameters to your .env.local. These parameters should point to the location of the certificate and key files on your local machine. The certificate and key files should be in PEM format. The key file can be encrypted with a passphrase, in which case you will also need to add the CLIENT_KEY_PASSWORD parameter to your .env.local.

If you're using a certificate signed by a private CA, you will also need to add the CA_PATH parameter to your .env.local. This parameter should point to the location of the CA certificate file on your local machine.

If you're using a self-signed certificate, e.g. for testing or development purposes, you can set the REJECT_UNAUTHORIZED parameter to false in your .env.local. This will disable certificate validation, and allow Chat UI to connect to your custom endpoint.

Specific Embedding Model

A model can use any of the embedding models defined in .env.local, (currently used when web searching), by default it will use the first embedding model, but it can be changed with the field embeddingModel:

TEXT_EMBEDDING_MODELS = `[
  {
    "name": "Xenova/gte-small",
    "chunkCharLength": 512,
    "endpoints": [
      {"type": "transformersjs"}
    ]
  },
  {
    "name": "intfloat/e5-base-v2",
    "chunkCharLength": 768,
    "endpoints": [
      {"type": "tei", "url": "http://127.0.0.1:8080/", "authorization": "Basic VVNFUjpQQVNT"},
      {"type": "tei", "url": "http://127.0.0.1:8081/"}
    ]
  }
]`

MODELS=`[
  {
      "name": "Ollama Mistral",
      "chatPromptTemplate": "...",
      "embeddingModel": "intfloat/e5-base-v2"
      "parameters": {
        ...
      },
      "endpoints": [
        ...
      ]
  }
]`

Deploying to a HF Space

Create a DOTENV_LOCAL secret to your HF space with the content of your .env.local, and they will be picked up automatically when you run.

Building

To create a production version of your app:

npm run build

You can preview the production build with npm run preview.

To deploy your app, you may need to install an adapter for your target environment.

Config changes for HuggingChat

The config file for HuggingChat is stored in the .env.template file at the root of the repository. It is the single source of truth that is used to generate the actual .env.local file using our CI/CD pipeline. See updateProdEnv for more details.

If you want to make changes to model config for HuggingChat, you should do so against .env.template.

We currently use the following secrets for deploying HuggingChat in addition to the .env.template above:

  • MONGODB_URL
  • HF_TOKEN
  • OPENID_CONFIG
  • SERPER_API_KEY

They are defined as secrets in the repository.

Testing config changes locally

You can test the config changes locally by first creating an .env.SECRET_CONFIG file with the secrets defined above. Then you can run the following command to generate the .env.local file:

npm run updateLocalEnv

This will replace your .env.local file with the one that will be used in prod (simply taking .env.template + .env.SECRET_CONFIG).

Populate database

The MONGODB_URL used for this script will be fetched from .env.local. Make sure it's correct! The command runs directly on the database.

You can populate the database using faker data using the populate script:

npm run populate <flags here>

At least one flag must be specified, the following flags are available:

  • reset - resets the database
  • all - populates all tables
  • users - populates the users table
  • settings - populates the settings table for existing users
  • assistants - populates the assistants table for existing users
  • conversations - populates the conversations table for existing users

For example, you could use it like so:

npm run populate reset

to clear out the database. Then login in the app to create your user and run the following command:

npm run populate users settings assistants conversations

to populate the database with fake data, including fake conversations and assistants for your user.