name: hessian-recipes-managed-cpu-lifecycle
description: Eight-step CPU checkpoint and pause/resume qualification
workspace: JAIF
project: hessian-recipes
resources:
  resource_pool: 42_Compute
  slots_per_trial: 0
max_restarts: 1
entrypoint:
  - /bin/bash
  - -lc
  - >-
    cd /pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes &&
    PYTHONPATH=src exec /usr/bin/timeout --signal=TERM --kill-after=30s 900s
    python3 -m hessian_recipes.managed_trial cpu
    --output runs/managed-cpu --steps 8 --delay 3 --max-wall-seconds 900
searcher:
  name: single
  metric: lifecycle_value
  smaller_is_better: false
  max_length:
    batches: 8
environment:
  image:
    cpu: ctctctct/forty-two-prod:test
    cuda: ctctctct/forty-two-prod:test
    rocm: ctctctct/forty-two-prod:test
  environment_variables:
    - CUDA_VISIBLE_DEVICES=
    - OMP_NUM_THREADS=2
    - OPENBLAS_NUM_THREADS=2
    - MKL_NUM_THREADS=2
    - TOKENIZERS_PARALLELISM=false
    - HF_HUB_OFFLINE=1
    - TRANSFORMERS_OFFLINE=1
bind_mounts:
  - host_path: /pfss/mlde/workspaces/mlde_wsp_JAIF
    container_path: /pfss/mlde/workspaces/mlde_wsp_JAIF
    read_only: false
checkpoint_storage:
  type: directory
  container_path: /pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/runs/determined-managed-storage
  save_trial_latest: 2
  save_trial_best: 0
  save_experiment_best: 0