mirror of https://github.com/danny-avila/LibreChat.git synced 2025-12-17 17:00:15 +01:00

* Update to use docker-compose.overridge.yml

Add GPU Acceleration links

* Update litellm.md

2024-02-09 07:11:09 -05:00

9.5 KiB

Raw Blame History

title	description	weight
🚅 LiteLLM and Ollama	Using LibreChat with LiteLLM Proxy	-7

Using LibreChat with LiteLLM Proxy

Use LiteLLM Proxy for:

Calling 100+ LLMs Huggingface/Bedrock/TogetherAI/etc. in the OpenAI ChatCompletions & Completions format
Load balancing - between Multiple Models + Deployments of the same model LiteLLM proxy can handle 1k+ requests/second during load tests
Authentication & Spend Tracking Virtual Keys

Start LiteLLM Proxy Server

Pip install litellm

pip install litellm

Create a config.yaml for litellm proxy

More information on LiteLLM configurations here: docs.litellm.ai/docs/simple_proxy

model_list:
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-small-eu
      api_base: https://my-endpoint-europe-berri-992.openai.azure.com/
      api_key: 
      rpm: 6      # Rate limit for this deployment: in requests per minute (rpm)
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-small-ca
      api_base: https://my-endpoint-canada-berri992.openai.azure.com/
      api_key: 
      rpm: 6
  - model_name: gpt-3.5-turbo
    litellm_params:
      model: azure/gpt-turbo-large
      api_base: https://openai-france-1234.openai.azure.com/
      api_key: 
      rpm: 1440

Start the proxy

litellm --config /path/to/config.yaml

#INFO: Proxy running on http://0.0.0.0:8000

Use LiteLLM Proxy Server with LibreChat

1. Clone the repo

git clone https://github.com/danny-avila/LibreChat.git

2. Modify Librechat's `docker-compose.yml`

OPENAI_REVERSE_PROXY=http://host.docker.internal:8000/v1/chat/completions

Important: As of v0.6.6, it's recommend you use the librechat.yaml Configuration file (guide here) to add Reverse Proxies as separate endpoints.

3. Save fake OpenAI key in Librechat's `.env`

Copy Librechat's .env.example to .env and overwrite the default OPENAI_API_KEY (by default it requires the user to pass a key).

OPENAI_API_KEY=sk-1234

4. Run LibreChat:

docker compose up

Why use LiteLLM?

Access to Multiple LLMs: It allows calling over 100 LLMs from platforms like Huggingface, Bedrock, TogetherAI, etc., using OpenAI's ChatCompletions and Completions format.
Load Balancing: Capable of handling over 1,000 requests per second during load tests, it balances load across various models and deployments.
Authentication & Spend Tracking: The server supports virtual keys for authentication and tracks spending.

Key components and features include:

Installation: Easy installation.
Testing: Testing features to route requests to specific models.
Server Endpoints: Offers multiple endpoints for chat completions, completions, embeddings, model lists, and key generation.
Supported LLMs: Supports a wide range of LLMs, including AWS Bedrock, Azure OpenAI, Huggingface, AWS Sagemaker, Anthropic, and more.
Proxy Configurations: Allows setting various parameters like model list, server settings, environment variables, and more.
Multiple Models Management: Configurations can be set up for managing multiple models with fallbacks, cooldowns, retries, and timeouts.
Embedding Models Support: Special configurations for embedding models.
Authentication Management: Features for managing authentication through virtual keys, model upgrades/downgrades, and tracking spend.
Custom Configurations: Supports setting model-specific parameters, caching responses, and custom prompt templates.
Debugging Tools: Options for debugging and logging proxy input/output.
Deployment and Performance: Information on deploying LiteLLM Proxy and its performance metrics.
Proxy CLI Arguments: A wide range of command-line arguments for customization.

Overall, LiteLLM Server offers a comprehensive suite of tools for managing, deploying, and interacting with a variety of LLMs, making it a versatile choice for large-scale AI applications.

Ollama

Use Ollama for

Run large language models on local hardware
Host multiple models
Dynamically load the model upon request

GPU Acceleration

Linux: Requires a Linux distrubution support by official Nvidia drivers. Nvidia CUDA Toolkit
Windows: Requires Windows Subsytem for Linux. Follow Nvidia instructions at Nvidia WSL User Guide
macOS: macOS Ollama Download

docker-compose.override.yml with GPU

version: "3.8"
services:
  litellm:
    image: ghcr.io/berriai/litellm:main-latest
    volumes:
      - ./litellm/litellm-config.yaml:/app/config.yaml
    command: [ "--config", "/app/config.yaml", "--port", "8000", "--num_workers", "8" ]
  ollama:
    image: ollama/ollama
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              capabilities: [compute, utility]
    ports:
      - "11434:11434"
    volumes:
      - ./ollama:/root/.ollama

Loading Models in Ollama

Browse the available models at Ollama Library
Run docker exec -it ollama /bin/bash
Copy the text from the Tags tab from the library website. It should begin with 'ollama run'
Check model size. Models that can run in GPU memory perform the best.
Use /bye to exit the terminal

Litellm Ollama Configuration

Add the below lines to the config to access the Ollama models

  - model_name: mixtral
    litellm_params:
      model: ollama/mixtral:8x7b-instruct-v0.1-q5_K_M
      api_base: http://ollama:11434
      stream: True
  - model_name: mistral
    litellm_params:
      model: ollama/mistral
      api_base: http://ollama:11434
      stream: True

Caching with Redis

Litellm supports in-memory, redis, and s3 caching. Note: Caching currently only works with exact matching.

Update docker-compose.override.yml to enable Redis

Add the below service to your docker-compose.override.yml

  redis:
    image: redis:7-alpine
    command:
    - sh
    - -c # this is to evaluate the $REDIS_PASSWORD from the env
    - redis-server --appendonly yes --requirepass $$REDIS_PASSWORD ## $$ because of docker-compose
    environment:
      REDIS_PASSWORD: RedisChangeMe
    volumes:
    - ./redis:/data

Add the following to the environment variables in the litellm service inside the docker-compose.override.yml

  litellm:
    image: ghcr.io/berriai/litellm:main-latest
    volumes:
      - ./litellm/litellm-config.yaml:/app/config.yaml
    command: [ "--config", "/app/config.yaml", "--port", "8000", "--num_workers", "8" ]
    environment:
      REDIS_HOST: redis
      REDIS_PORT: 6379
      REDIS_PASSWORD: RedisChangeMe

Update Litellm Config File

Add the below options to the litellm config file

litellm_settings: # module level litellm settings - https://github.com/BerriAI/litellm/blob/main/litellm/__init__.py
  cache: True          # set cache responses to True, litellm defaults to using a redis cache
  cache_params:         # cache_params are optional
    type: "redis"  # The type of cache to initialize. Can be "local" or "redis". Defaults to "local".

    # Optional configurations
    supported_call_types: ["acompletion", "completion", "embedding", "aembedding"] # defaults to all litellm call types

Performance Monitoring with Langfuse

Litellm supports various logging and observability options. The settings below will enable Langfuse which will provide a cache_hit tag showing which conversations used cache.

Update docker-compose.override.yml to enable Langfuse

Langfuse requires a postgres database, so add both postgres and langfuse services to the docker-compose.override.yml

  langfuse-server:
    image: ghcr.io/langfuse/langfuse:latest
    depends_on:
      - db
    ports:
      - "3000:3000"
    environment:
      - NODE_ENV=production
      - DATABASE_URL=postgresql://postgres:PostgresChangeMe@db:5432/postgres
      - NEXTAUTH_SECRET=ChangeMe
      - SALT=ChangeMe
      - NEXTAUTH_URL=http://localhost:3000
      - TELEMETRY_ENABLED=${TELEMETRY_ENABLED:-true}
      - NEXT_PUBLIC_SIGN_UP_DISABLED=${NEXT_PUBLIC_SIGN_UP_DISABLED:-false}
      - LANGFUSE_ENABLE_EXPERIMENTAL_FEATURES=${LANGFUSE_ENABLE_EXPERIMENTAL_FEATURES:-false}

  db:
    image: postgres
    restart: always
    environment:
      - POSTGRES_USER=postgres
      - POSTGRES_PASSWORD=PostgresChangeMe
      - POSTGRES_DB=postgres
    volumes:
      - ./postgres:/var/lib/postgresql/data

Once Langfuse is running, create an account by accessing the web interface on port 3000. Create a new project to obtain the needed public and private key used by the litellm config Add environement variable within the litellm service within docker-compose.override.yml

  litellm:
    image: ghcr.io/berriai/litellm:main-latest
    ports:
      - "8000:8000"
    volumes:
      - /srv/litellm/config/litellm-config.yaml:/app/config.yaml
    command: [ "--config", "/app/config.yaml", "--port", "8000", "--num_workers", "8" ]
    environment:
      LANGFUSE_PUBLIC_KEY: pk-lf-RandomStringFromLangfuseWebInterface
      LANGFUSE_SECRET_KEY: sk-lf-RandomStringFromLangfuseWebInterface
      LANGFUSE_HOST: http://langfuse-server:3000

Update litellm config file

litellm_settings:
  success_callback: ["langfuse"]

9.5 KiB Raw Blame History