mirror of
https://github.com/danny-avila/LibreChat.git
synced 2026-01-19 16:56:12 +01:00
🅰️ feat: Azure Config to Allow Different Deployments per Model (#1863)
* wip: first pass for azure endpoint schema * refactor: azure config to return groupMap and modelConfigMap * wip: naming and schema changes * refactor(errorsToString): move to data-provider * feat: rename to azureGroups, add additional tests, tests all expected outcomes, return errors * feat(AppService): load Azure groups * refactor(azure): use imported types, write `mapModelToAzureConfig` * refactor: move `extractEnvVariable` to data-provider * refactor(validateAzureGroups): throw on duplicate groups or models; feat(mapModelToAzureConfig): throw if env vars not present, add tests * refactor(AppService): ensure each model is properly configured on startup * refactor: deprecate azureOpenAI environment variables in favor of librechat.yaml config * feat: use helper functions to handle and order enabled/default endpoints; initialize azureOpenAI from config file * refactor: redefine types as well as load azureOpenAI models from config file * chore(ci): fix test description naming * feat(azureOpenAI): use validated model grouping for request authentication * chore: bump data-provider following rebase * chore: bump config file version noting significant changes * feat: add title options and switch azure configs for titling and vision requests * feat: enable azure plugins from config file * fix(ci): pass tests * chore(.env.example): mark `PLUGINS_USE_AZURE` as deprecated * fix(fetchModels): early return if apiKey not passed * chore: fix azure config typing * refactor(mapModelToAzureConfig): return baseURL and headers as well as azureOptions * feat(createLLM): use `azureOpenAIBasePath` * feat(parsers): resolveHeaders * refactor(extractBaseURL): handle invalid input * feat(OpenAIClient): handle headers and baseURL for azureConfig * fix(ci): pass `OpenAIClient` tests * chore: extract env var for azureOpenAI group config, baseURL * docs: azureOpenAI config setup docs * feat: safe check of potential conflicting env vars that map to unique placeholders * fix: reset apiKey when model switches from originally requested model (vision or title) * chore: linting * docs: CONFIG_PATH notes in custom_config.md
This commit is contained in:
Danny Avila
GitHub
parent
7a55132e42
commit
097a978e5b
37 changed files with 2066 additions and 394 deletions
|
|
@ -236,221 +236,9 @@ Note: Using Gemini models through Vertex AI is possible but not yet supported.
|
|||
|
||||
## Azure OpenAI
|
||||
|
||||
In order to use Azure OpenAI with this project, specific environment variables must be set in your `.env` file. These variables will be used for constructing the API URLs.
|
||||
### Please see the dedicated [Azure OpenAI Setup Guide.](./azure_openai.md)
|
||||
|
||||
The variables needed are outlined below:
|
||||
|
||||
### Required Variables
|
||||
|
||||
These variables construct the API URL for Azure OpenAI.
|
||||
|
||||
* `AZURE_API_KEY`: Your Azure OpenAI API key.
|
||||
* `AZURE_OPENAI_API_INSTANCE_NAME`: The instance name of your Azure OpenAI API.
|
||||
* `AZURE_OPENAI_API_DEPLOYMENT_NAME`: The deployment name of your Azure OpenAI API.
|
||||
* `AZURE_OPENAI_API_VERSION`: The version of your Azure OpenAI API.
|
||||
|
||||
For example, with these variables, the URL for chat completion would look something like:
|
||||
```plaintext
|
||||
https://{AZURE_OPENAI_API_INSTANCE_NAME}.openai.azure.com/openai/deployments/{AZURE_OPENAI_API_DEPLOYMENT_NAME}/chat/completions?api-version={AZURE_OPENAI_API_VERSION}
|
||||
```
|
||||
You should also consider changing the `AZURE_OPENAI_MODELS` variable to the models available in your deployment.
|
||||
|
||||
```bash
|
||||
# .env file
|
||||
AZURE_OPENAI_MODELS=gpt-4-1106-preview,gpt-4,gpt-3.5-turbo,gpt-3.5-turbo-1106,gpt-4-vision-preview
|
||||
```
|
||||
|
||||
Overriding the construction of the API URL will be possible but is not yet implemented. Follow progress on this feature here: **[Issue #1266](https://github.com/danny-avila/LibreChat/issues/1266)**
|
||||
|
||||
### Model Deployments
|
||||
|
||||
> Note: a change will be developed to improve current configuration settings, to allow multiple deployments/model configurations setup with ease: **[#1390](https://github.com/danny-avila/LibreChat/issues/1390)**
|
||||
|
||||
As of 2023-12-18, the Azure API allows only one model per deployment.
|
||||
|
||||
**It's highly recommended** to name your deployments *after* the model name (e.g., "gpt-3.5-turbo") for easy deployment switching.
|
||||
|
||||
When you do so, LibreChat will correctly switch the deployment, while associating the correct max context per model, if you have the following environment variable set:
|
||||
|
||||
```bash
|
||||
AZURE_USE_MODEL_AS_DEPLOYMENT_NAME=TRUE
|
||||
```
|
||||
|
||||
For example, when you have set `AZURE_USE_MODEL_AS_DEPLOYMENT_NAME=TRUE`, the following deployment configuration provides the most seamless, error-free experience for LibreChat, including Vision support and tracking the correct max context tokens:
|
||||
|
||||
|
||||
|
||||
|
||||
Alternatively, you can use custom deployment names and set `AZURE_OPENAI_DEFAULT_MODEL` for expected functionality.
|
||||
|
||||
- **`AZURE_OPENAI_MODELS`**: List the available models, separated by commas without spaces. The first listed model will be the default. If left blank, internal settings will be used. Note that deployment names can't have periods, which are removed when generating the endpoint.
|
||||
|
||||
Example use:
|
||||
|
||||
```bash
|
||||
# .env file
|
||||
AZURE_OPENAI_MODELS=gpt-3.5-turbo,gpt-4,gpt-5
|
||||
|
||||
```
|
||||
|
||||
- **`AZURE_USE_MODEL_AS_DEPLOYMENT_NAME`**: Enable using the model name as the deployment name for the API URL.
|
||||
|
||||
Example use:
|
||||
|
||||
```bash
|
||||
# .env file
|
||||
AZURE_USE_MODEL_AS_DEPLOYMENT_NAME=TRUE
|
||||
|
||||
```
|
||||
|
||||
### Setting a Default Model for Azure
|
||||
|
||||
This section is relevant when you are **not** naming deployments after model names as shown above.
|
||||
|
||||
**Important:** The Azure OpenAI API does not use the `model` field in the payload but is a necessary identifier for LibreChat. If your deployment names do not correspond to the model names, and you're having issues with the model not being recognized, you should set this field to explicitly tell LibreChat to treat your Azure OpenAI API requests as if the specified model was selected.
|
||||
|
||||
If AZURE_USE_MODEL_AS_DEPLOYMENT_NAME is enabled, the model you set with `AZURE_OPENAI_DEFAULT_MODEL` will **not** be recognized and will **not** be used as the deployment name; instead, it will use the model selected by the user as the "deployment" name.
|
||||
|
||||
- **`AZURE_OPENAI_DEFAULT_MODEL`**: Override the model setting for Azure, useful if using custom deployment names.
|
||||
|
||||
Example use:
|
||||
|
||||
```bash
|
||||
# .env file
|
||||
# MUST be a real OpenAI model, named exactly how it is recognized by OpenAI API (not Azure)
|
||||
AZURE_OPENAI_DEFAULT_MODEL=gpt-3.5-turbo # do include periods in the model name here
|
||||
|
||||
```
|
||||
|
||||
### Using a Specified Base URL with Azure
|
||||
|
||||
The base URL for Azure OpenAI API requests can be dynamically configured. This is useful for proxying services such as [Cloudflare AI Gateway](https://developers.cloudflare.com/ai-gateway/providers/azureopenai/), or if you wish to explicitly override the baseURL handling of the app.
|
||||
|
||||
LibreChat will use the `AZURE_OPENAI_BASEURL` environment variable, which can include placeholders for the Azure OpenAI API instance and deployment names.
|
||||
|
||||
In the application's environment configuration, the base URL is set like this:
|
||||
|
||||
```bash
|
||||
# .env file
|
||||
AZURE_OPENAI_BASEURL=https://example.azure-api.net/${INSTANCE_NAME}/${DEPLOYMENT_NAME}
|
||||
|
||||
# OR
|
||||
AZURE_OPENAI_BASEURL=https://${INSTANCE_NAME}.openai.azure.com/openai/deployments/${DEPLOYMENT_NAME}
|
||||
|
||||
# Cloudflare example
|
||||
AZURE_OPENAI_BASEURL=https://gateway.ai.cloudflare.com/v1/ACCOUNT_TAG/GATEWAY/azure-openai/${INSTANCE_NAME}/${DEPLOYMENT_NAME}
|
||||
```
|
||||
|
||||
The application replaces `${INSTANCE_NAME}` and `${DEPLOYMENT_NAME}` in the `AZURE_OPENAI_BASEURL`, processed according to the other settings discussed in the guide.
|
||||
|
||||
**You can also omit the placeholders completely and simply construct the baseURL with your credentials:**
|
||||
|
||||
```bash
|
||||
# .env file
|
||||
AZURE_OPENAI_BASEURL=https://instance-1.openai.azure.com/openai/deployments/deployment-1
|
||||
|
||||
# Cloudflare example
|
||||
AZURE_OPENAI_BASEURL=https://gateway.ai.cloudflare.com/v1/ACCOUNT_TAG/GATEWAY/azure-openai/instance-1/deployment-1
|
||||
```
|
||||
|
||||
Setting these values will override all of the application's internal handling of the instance and deployment names and use your specified base URL.
|
||||
|
||||
**Notes:**
|
||||
- You should still provide the `AZURE_OPENAI_API_VERSION` and `AZURE_API_KEY` via the .env file as they are programmatically added to the requests.
|
||||
- When specifying instance and deployment names in the `AZURE_OPENAI_BASEURL`, their respective environment variables can be omitted (`AZURE_OPENAI_API_INSTANCE_NAME` and `AZURE_OPENAI_API_DEPLOYMENT_NAME`) except for use with Plugins.
|
||||
- Specifying instance and deployment names in the `AZURE_OPENAI_BASEURL` instead of placeholders creates conflicts with "plugins," "vision," "default-model," and "model-as-deployment-name" support.
|
||||
- Due to the conflicts that arise with other features, it is recommended to use placeholder for instance and deployment names in the `AZURE_OPENAI_BASEURL`
|
||||
|
||||
### Enabling Auto-Generated Titles with Azure
|
||||
|
||||
The default titling model is set to `gpt-3.5-turbo`.
|
||||
|
||||
If you're using `AZURE_USE_MODEL_AS_DEPLOYMENT_NAME` and have "gpt-35-turbo" setup as a deployment name, this should work out-of-the-box.
|
||||
|
||||
In any case, you can adjust the title model as such: `OPENAI_TITLE_MODEL=your-title-model`
|
||||
|
||||
### Using GPT-4 Vision with Azure
|
||||
|
||||
Currently, the best way to setup Vision is to use your deployment names as the model names, as [shown here](#model-deployments)
|
||||
|
||||
This will work seamlessly as it does with the [OpenAI endpoint](#openai) (no need to select the vision model, it will be switched behind the scenes)
|
||||
|
||||
Alternatively, you can set the [required variables](#required-variables) to explicitly use your vision deployment, but this may limit you to exclusively using your vision deployment for all Azure chat settings.
|
||||
|
||||
|
||||
**Notes:**
|
||||
|
||||
- If using `AZURE_OPENAI_BASEURL`, you should not specify instance and deployment names instead of placeholders as the vision request will fail.
|
||||
- As of December 18th, 2023, Vision models seem to have degraded performance with Azure OpenAI when compared to [OpenAI](#openai)
|
||||
|
||||
|
||||
|
||||
|
||||
> Note: a change will be developed to improve current configuration settings, to allow multiple deployments/model configurations setup with ease: **[#1390](https://github.com/danny-avila/LibreChat/issues/1390)**
|
||||
|
||||
### Generate images with Azure OpenAI Service (DALL-E)
|
||||
|
||||
| Model ID | Feature Availability | Max Request (characters) |
|
||||
|----------|----------------------|-------------------------|
|
||||
| dalle2 | East US | 1000 |
|
||||
| dalle3 | Sweden Central | 4000 |
|
||||
|
||||
- First you need to create an Azure resource that hosts DALL-E
|
||||
- At the time of writing, dall-e-3 is available in the `SwedenCentral` region, dall-e-2 in the `EastUS` region.
|
||||
- Then, you need to deploy the image generation model in one of the above regions.
|
||||
- Read the [Azure OpenAI Image Generation Quickstart Guide](https://learn.microsoft.com/en-us/azure/ai-services/openai/dall-e-quickstart) for further assistance
|
||||
- Configure your environment variables based on Azure credentials:
|
||||
|
||||
**- For DALL-E-3:**
|
||||
|
||||
```bash
|
||||
DALLE3_AZURE_API_VERSION=the-api-version # e.g.: 2023-12-01-preview
|
||||
DALLE3_BASEURL=https://<AZURE_OPENAI_API_INSTANCE_NAME>.openai.azure.com/openai/deployments/<DALLE3_DEPLOYMENT_NAME>/
|
||||
DALLE3_API_KEY=your-azure-api-key-for-dall-e-3
|
||||
```
|
||||
|
||||
**- For DALL-E-2:**
|
||||
|
||||
```bash
|
||||
DALLE2_AZURE_API_VERSION=the-api-version # e.g.: 2023-12-01-preview
|
||||
DALLE2_BASEURL=https://<AZURE_OPENAI_API_INSTANCE_NAME>.openai.azure.com/openai/deployments/<DALLE2_DEPLOYMENT_NAME>/
|
||||
DALLE2_API_KEY=your-azure-api-key-for-dall-e-2
|
||||
```
|
||||
|
||||
**DALL-E Notes:**
|
||||
|
||||
- For DALL-E-3, the default system prompt has the LLM prefer the ["vivid" style](https://platform.openai.com/docs/api-reference/images/create#images-create-style) parameter, which seems to be the preferred setting for ChatGPT as "natural" can sometimes produce lackluster results.
|
||||
- See official prompt for reference: **[DALL-E System Prompt](https://github.com/spdustin/ChatGPT-AutoExpert/blob/main/_system-prompts/dall-e.md)**
|
||||
- You can adjust the system prompts to your liking:
|
||||
|
||||
```bash
|
||||
DALLE3_SYSTEM_PROMPT="Your DALL-E-3 System Prompt here"
|
||||
DALLE2_SYSTEM_PROMPT="Your DALL-E-2 System Prompt here"
|
||||
```
|
||||
|
||||
- The `DALLE_REVERSE_PROXY` environment variable is ignored when Azure credentials (DALLEx_AZURE_API_VERSION and DALLEx_BASEURL) for DALL-E are configured.
|
||||
|
||||
### Optional Variables
|
||||
|
||||
*These variables are currently not used by LibreChat*
|
||||
|
||||
* `AZURE_OPENAI_API_COMPLETIONS_DEPLOYMENT_NAME`: The deployment name for completion. This is currently not in use but may be used in future.
|
||||
* `AZURE_OPENAI_API_EMBEDDINGS_DEPLOYMENT_NAME`: The deployment name for embedding. This is currently not in use but may be used in future.
|
||||
|
||||
These two variables are optional but may be used in future updates of this project.
|
||||
|
||||
### Using Plugins with Azure
|
||||
|
||||
Note: To use the Plugins endpoint with Azure OpenAI, you need a deployment supporting **[function calling](https://techcommunity.microsoft.com/t5/azure-ai-services-blog/function-calling-is-now-available-in-azure-openai-service/ba-p/3879241)**. Otherwise, you need to set "Functions" off in the Agent settings. When you are not using "functions" mode, it's recommend to have "skip completion" off as well, which is a review step of what the agent generated.
|
||||
|
||||
To use Azure with the Plugins endpoint, make sure the following environment variables are set:
|
||||
|
||||
* `PLUGINS_USE_AZURE`: If set to "true" or any truthy value, this will enable the program to use Azure with the Plugins endpoint.
|
||||
* `AZURE_API_KEY`: Your Azure API key must be set with an environment variable.
|
||||
|
||||
**Important:**
|
||||
|
||||
- If using `AZURE_OPENAI_BASEURL`, you should not specify instance and deployment names instead of placeholders as the plugin request will fail.
|
||||
This was done to improve upon legacy configuration settings, to allow multiple deployments/model configurations setup with ease: **[#1390](https://github.com/danny-avila/LibreChat/issues/1390)**
|
||||
|
||||
---
|
||||
|
||||
|
|
@ -476,6 +264,10 @@ OpenRouter is integrated to the LibreChat by overriding the OpenAI endpoint.
|
|||
|
||||
#### Setup (legacy):
|
||||
|
||||
**Note:** It is NOT recommended to setup OpenRouter this way with versions 0.6.6 or higher of LibreChat as it may be removed in future versions.
|
||||
|
||||
As noted earlier, [review the Custom Config Guide (click here)](./custom_config.md) to add an `OpenRouter` Endpoint instead.
|
||||
|
||||
- Signup to **[OpenRouter](https://openrouter.ai/)** and create a key. You should name it and set a limit as well.
|
||||
- Set the environment variable `OPENROUTER_API_KEY` in your .env file to the key you just created.
|
||||
- Set something in the `OPENAI_API_KEY`, it can be anyting, but **do not** leave it blank or set to `user_provided`
|
||||
|
|
|
|||
Loading…
Add table
Add a link
Reference in a new issue