Tutorial: DNN Training#

This example uses SkyPilot to train a Transformer-based language model from HuggingFace.

First, define a task YAML with the resource requirements, the setup commands, and the commands to run:

# dnn.yaml

name: huggingface

resources:
  accelerators: V100:4

# Optional: upload a working directory to remote ~/sky_workdir.
# Commands in "setup" and "run" will be executed under it.
#
# workdir: .

# Optional: upload local files.
# Format:
#   /remote/path: /local/path
#
# file_mounts:
#   ~/.vimrc: ~/.vimrc
#   ~/.netrc: ~/.netrc

setup: |
  set -e  # Exit if any command failed.
  git clone https://github.com/huggingface/transformers/ || true
  cd transformers
  pip install .
  cd examples/pytorch/text-classification
  pip install -r requirements.txt torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113

run: |
  set -e  # Exit if any command failed.
  cd transformers/examples/pytorch/text-classification
  python run_glue.py \
    --model_name_or_path bert-base-cased \
    --dataset_name imdb  \
    --do_train \
    --max_seq_length 128 \
    --per_device_train_batch_size 32 \
    --learning_rate 2e-5 \
    --max_steps 50 \
    --output_dir /tmp/imdb/ --overwrite_output_dir \
    --fp16

Tip

In the YAML, the workdir and file_mounts fields are commented out. To learn about how to use them to mount local dirs/files or object store buckets (S3, GCS, R2) into your cluster, see Syncing Code and Artifacts.

Then, launch training:

$ sky launch -c lm-cluster dnn.yaml

This will provision the cheapest cluster with the required resources, execute the setup commands, then execute the run commands.

After the training job starts running, you can safely Ctrl-C to detach from logging and the job will continue to run remotely on the cluster. To stop the job, use the sky cancel <cluster_name> <job_id> command (refer to CLI reference).

After training, transfer artifacts such as logs and checkpoints using familiar tools.

Tip

Feel free to copy-paste the YAML above and customize it for your own project.