fastllm Docker & Self-Hosted Deployment

GitHub

Fastllm是后端无依赖的高性能大模型推理库。同时支持张量并行推理稠密模型和混合模式推理MOE模型，任意10G以上显卡即可推理满血DeepSeek。双路9004/9005服务器+单显卡部署DeepSeek满血满精度原版模型，单并发20tps；INT4量化模型单并发30tps，多并发可达60+。

5,112 stars C++ C++ 104 files · ~126,163 tokens
RAW Compose
1. Pull Image (DOCKERHUB) docker pull ztxz16/fastllm
2. Docker Run (1-Click Launch) docker run -d --name fastllm -p 11234:11234 --restart always ztxz16/fastllm:latest
3. Podman Run podman run -d --name fastllm -p 11234:11234 --restart always ztxz16/fastllm:latest
4. Skopeo Inspect (Zero-Pull Manifest) skopeo inspect docker://docker.io/ztxz16/fastllm:latest
5. ORAS Pull (OCI Artifact) oras pull ztxz16/fastllm:latest
6. Compose Deployment docker compose up -d
docker-compose.yaml
Download .yml 
version: '3.8'
services:
  fastllm:
    build:
      context: .
      args:
        DOCKER_BUILDKIT: 0
        # privileged: true
      platforms:
        - "linux/amd64"
      tags:
        - "fastllm:v0.9"
    restart: always
    ports:
      - 11234:8081
    volumes:
      - ./models/:/models/
    command: /fastllm/build/webui -p /models/chatglm2-6b-int8.flm -w ./example/webui/web