Optimus/code/scripts/scripts_philly/train_clm_wiki103.yaml

description: Train AE on Wiki 103 Dataset

auth:
  # which virtual cluster you belong to (msrlabs, etc.). Everyone has access to "pnrsy".
  vc: msrlabs
  # physical cluster to use (cam, gcr, rr1) or Azure clusters (eu1, eu2, etc.)
  # cluster: rr2, eu2, eu1 et1 
  cluster: cam
  # docker environment (vm) in which your job will run. we provide "generic" dockers
  # with the main deep learning toolkit installed (PyTorch, TF, Chainer, etc.)
  docker:
    # image: philly/jobs/custom/generic-docker:py27
    # registry: phillyregistry.azurecr.io
    image: chunyl/pytorch-transformers:v0
    registry: index.docker.io

storage:
  _default:
    #use_phillyfs: True
    storage_account_name: textae
    container_name: bigtextae
    mount_path: /mnt/_default

code:
  # local directory of the code. this will be uploaded to the server.
  # $CONFIG_DIR is expanded to the directory of this config file
  code_upload: False
  remote_dir: code/
  local_dir: $CONFIG_DIR/code

#data:
  # data upload is not required for this example
  #data_upload: False

search:
  job_template:
    name: vq_{experiment_name:s}_{bs_option:.1f}
    sku: G4 # G4 # G1
    command:
    - pip install --user --editable .
    - python examples/run_lm_finetuning.py --output_dir ../output/philly_clm_wiki103 --model_type gpt2 --model_name_or_path gpt2  --do_train --train_data_file ../data/datasets/wikitext-103/train.txt --do_eval --eval_data_file ../data/datasets/wikitext-103/valid.txt --per_gpu_train_batch_size {bs_option}  --save_steps 500 --overwrite_output_dir
  max_trials: 20
  type: grid
  params:
    - name: bs_option
      spec: discrete
      values: [2] # [top,bottom]