Docker for ML — মেশিন লার্নিং

Hook — ‘আমার machine এ চলে!’

তোমার laptop এ Python 3.11 + CUDA 12, server এ Python 3.9 + CUDA 11 — model crash। Docker container এ সব dependency পুরে দাও — যেখানেই চালাও, একই behavior।

Core Concepts

Image — blueprint (read-only)।
Container — running instance।
Dockerfile — image build recipe।
Registry — image store (Docker Hub, GHCR, ECR)।
Volume — persistent data।
Network — container যোগাযোগ।

Dockerfile — ML API

Dockerfile

FROM python:3.11-slim

WORKDIR /app

# system dep
RUN apt-get update && apt-get install -y --no-install-recommends \
    build-essential && rm -rf /var/lib/apt/lists/*

# requirements first → cache layer
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt

# app code
COPY app.py model.pkl ./

EXPOSE 8000
HEALTHCHECK CMD curl -f http://localhost:8000/health || exit 1
CMD ["uvicorn", "app:app", "--host", "0.0.0.0", "--port", "8000"]

build & run

docker build -t iris-api:1.0 .
docker run -d -p 8000:8000 --name iris iris-api:1.0
docker logs -f iris

GPU Container (PyTorch)

Dockerfile.gpu

FROM pytorch/pytorch:2.4.0-cuda12.1-cudnn9-runtime
WORKDIR /app
COPY . .
RUN pip install -r requirements.txt
CMD ["python", "serve.py"]

run with GPU

docker run --gpus all -p 8000:8000 my-llm-api

docker-compose — Multi-service

docker-compose.yml

services:
  api:
    build: .
    ports: ["8000:8000"]
    environment:
      - REDIS_URL=redis://cache:6379
    depends_on: [cache, db]
  cache:
    image: redis:7-alpine
  db:
    image: postgres:16
    environment:
      POSTGRES_PASSWORD: secret
    volumes: ["pgdata:/var/lib/postgresql/data"]
volumes:
  pgdata:

Best Practices

Slim base image (python:slim, distroless)।
Multi-stage build — final image ছোট।
.dockerignore দিয়ে অপ্রয়োজনীয় ফাইল বাদ।
Layer caching — কম-পরিবর্তনশীল layer আগে।
Non-root user দিয়ে run।
Tag image semver (1.0.0, not :latest)।
Secret env var দিয়ে, image এ নয়।

multi-stage

FROM python:3.11 AS builder
COPY requirements.txt .
RUN pip wheel --wheel-dir=/wheels -r requirements.txt

FROM python:3.11-slim
COPY --from=builder /wheels /wheels
RUN pip install --no-index --find-links=/wheels /wheels/*
COPY . /app
WORKDIR /app
CMD ["uvicorn", "app:app", "--host", "0.0.0.0"]

Summary

এক নজরে

Docker = Reproducible environment। Dockerfile + compose + GPU runtime + multi-stage = production-ready ML container।