docker-compose.yaml

services:
  convnets:
    build:
      args:
        - FROM_IMAGE_NAME=nvcr.io/nvaie/pytorch-2-2:22.07-nvaie-2.2-py3
      context: https://github.com/NVIDIA/DeepLearningExamples.git#master:PyTorch/Classification/ConvNets
      dockerfile: Dockerfile
    image: quickperf:convnets
    shm_size: 8gb
    ulimits:
      memlock: -1
      stack: 67108864
    runtime: "nvidia"
    entrypoint:
      [
        "python",
        "./multiproc.py",
        "--nproc_per_node=1",
        "./main.py",
        "/imagenet",
        "--arch=resnet50",
        "--workers=16",
        "--epochs=10",
        "--batch-size=256",
        "--optimizer-batch-size=2048",
        "--learning-rate=2.048",
        "--lr-schedule=cosine",
        "--warmup=8",
        "--label-smoothing=0.1",
        "--optimizer=sgd",
        "--momentum=0.875",
        "--weight-decay=3.0517578125e-05",
        "--print-freq=100",
        "--amp",
        "--static-loss-scale=128",
        "--prof=1000",
        "--seed=42",
        "--raport-file=/tmp/training-logs.json",
        "--no-checkpoints",
        "--workspace=/workspace",
        "--training-only",
        "--data-backend=synthetic",
      ]
    volumes:
      - type: bind
        source: logs/
        target: /tmp/

  nvidia-smi:
    build:
      args:
        - FROM_IMAGE_NAME=nvcr.io/nvaie/pytorch-2-2:22.07-nvaie-2.2-py3
      context: .
    image: quickperf:logger
    runtime: "nvidia"
    depends_on:
      - convnets
    volumes:
      - type: bind
        source: logs/
        target: /workspace/