Files
compose-anything/src/lmdeploy/README.md
2026-01-22 10:48:31 +08:00

1.3 KiB

LMDeploy Docker Compose

LMDeploy is a toolkit for compressing, deploying, and serving LLMs.

Quick Start

  1. (Optional) Configure the model and port in .env.

  2. Start the service:

    docker compose up -d
    
  3. Access the OpenAI compatible API at http://localhost:23333/v1.

Configuration

Environment Variable Default Description
LMDEPLOY_VERSION v0.11.1-cu12.8 LMDeploy image version
LMDEPLOY_PORT_OVERRIDE 23333 Host port for the API server
LMDEPLOY_MODEL internlm/internlm2-chat-1_8b HuggingFace model ID or local path
HF_TOKEN HuggingFace token for private models

Monitoring Health

The service includes a health check that verifies if the OpenAI /v1/models endpoint is responsive.

GPU Support

By default, this configuration reserves 1 NVIDIA GPU. Ensure you have the NVIDIA Container Toolkit installed on your host.