recipes_collection/config.yaml

# Original Copyright (c), NVIDIA CORPORATION. Modifications © Amazon.com

defaults:
  - _self_
  - cluster: slurm  # set to `slurm`, `k8s` or `sm_jobs`, depending on the desired cluster
  - recipes: training/llama/hf_llama3_8b_seq16k_gpu_p5x16_pretrain # select desired config inside the training directory
  - override hydra/job_logging: stdout

cluster_type: slurm  # bcm, bcp, k8s or sm_jobs. If bcm, k8s or sm_jobs, it must match - cluster above.
# If using sm_jobs cluster_type, set sm_jobs_config. See cluster/sm_jobs.yaml for example.

hydra:
  run:
    dir: .
  output_subdir: null

debug: False

instance_type: p5.48xlarge
base_results_dir: null  # Location to store the results, checkpoints and logs.

container: null

git:
  repo_url_or_path: null
  branch: null
  commit: null
  entry_script: null
  token: null
  update_adapter: false # if true it will re-install the Adapter code but not its dependencies

env_vars:
  NCCL_DEBUG: WARN  # Logging level for NCCL. Set to "INFO" for debug information

# Do not modify below, use the values above instead.
training_config: ${hydra:runtime.choices.recipes}