gpt-nemo.yaml

type: basic
format_version: 1
maintainers: [mcore]
loggers: [stdout]
spec:
  name: "{test_case}"
  model: gpt-nemo
  build: mcore-nemo
  nodes: 1
  gpus: 8
  platforms: dgx_a100
  time_limit: 1800
  scope: null
  script: |-
    ls
    cd /opt/NeMo
  
    ARGUMENTS=(
        "DATA_PATH='-'"
        "DATA_CACHE_PATH='-'"
        "OUTPUT_PATH={assets_dir}"
        "TENSORBOARD_PATH={assets_dir}/tensorboard"
        "CHECKPOINT_PATH=/workspace/checkpoints"
        "TRAINING_SCRIPT_PATH=/opt/NeMo/examples/nlp/language_modeling/megatron_gpt_pretraining.py"
        "TRAINING_PARAMS_PATH=/opt/megatron-lm/tests/functional_tests/test_cases/{model}/{test_case}/model_config.yaml"
        "GOLDEN_VALUES_PATH=/opt/megatron-lm/tests/functional_tests/test_cases/{model}/{test_case}/golden_values_{environment}.json"
        "N_REPEAT={n_repeat}"
    )

    bash /opt/megatron-lm/tests/functional_tests/shell_test_utils/run_ci_test.sh ${{ARGUMENTS[@]}}

products:
  - environment: [dev]
    scope: [mr]
    n_repeat: [5]
    test_case:
    - gpt3-nemo_126m_mr_mbs1_gbs8_mcore_te_tp2_pp4_vp3_seq_par_overlap_p2p_dgx_a100_1N8G
    - gpt3-nemo_126m_mr_mbs4_gbs64_mcore_te_tp1_pp1_dgx_a100_1N8G