# Prepare and qualify this exact plan before submitting; model_def is omitted.
name: hessian-recipes-managed-openenv
description: Qualified native GRPO training; DEV selection and final assessment remain separate
workspace: JAIF
project: hessian-recipes
resources:
  resource_pool: 42_Compute
  slots_per_trial: 1
  shm_size: 8589934592
max_restarts: 1
entrypoint:
  - /bin/bash
  - -lc
  - >-
    cd /pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes &&
    exec /usr/bin/timeout --signal=TERM --kill-after=30s 28800s
    .venv-gpu/bin/python -m hessian_recipes.managed_training
    --plan /pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/runs/agent-gyms/jaif-managed-openenv-04/experiment-plan.json
    --backend openenv
    --service-python /pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/.venv-openenv/bin/python
    --max-wall-seconds 28800
searcher:
  name: single
  metric: completed_optimizer_steps
  smaller_is_better: false
  max_length:
    batches: 256
environment:
  image: python:3.11-bookworm@sha256:a54ac40b50750eb18d584a0cccf902330c2708cc4248b1be3d5e2be7d07f2cad
  force_pull_image: false
  environment_variables:
    - OMP_NUM_THREADS=2
    - OPENBLAS_NUM_THREADS=2
    - MKL_NUM_THREADS=2
    - TOKENIZERS_PARALLELISM=false
    - HF_HUB_OFFLINE=1
    - TRANSFORMERS_OFFLINE=1
    - HF_DATASETS_OFFLINE=1
    - USER=ct1002
    - LOGNAME=ct1002
    - PYTHONHASHSEED=0
    - CUBLAS_WORKSPACE_CONFIG=:4096:8
    - HF_HOME=/pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/.cache/huggingface
    - XDG_CACHE_HOME=/pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/.cache
    - TORCHINDUCTOR_CACHE_DIR=/pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/.cache/torchinductor
    - TRITON_CACHE_DIR=/pfss/mlde/workspaces/mlde_wsp_JAIF/hessian_recipes/.cache/triton
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