config.yaml

# @package _global_

# specify here default training configuration
defaults:
  - _self_
  - callbacks: # pytorch-lightning callbacks
    - default

  # - trainer: ddp
  - logger: null # set logger here or use command line (e.g. `python train.py logger=tensorboard`)
  - paths: default
  - hydra: default
  - experiment: lm/dplm_150m # specifies pipeline and model

# default name for the experiment, determines logging folder path
# (you can overwrite this name in experiment configs)
name: ???

train:
  # set False to skip model training
  train: True
  # evaluate on test set, using best model weights achieved during training
  # lightning chooses best weights based on the metric specified in checkpoint callback
  test: True

  debug: false  

  force_restart: false # force to train from scratch

  # simply provide checkpoint path to resume training
  # it can be either an absolute path, 
  # or an relative path which will then be inferred from
  # 1) current workding directory (cwd), or
  # 2) checkpoint directory (${paths.ckpt_dir})
  ckpt_path: last.ckpt

  seed: 42 # seed for random number generators in pytorch, numpy and python.random

  lr: 1e-3 # learning rate
  monitor: ??? # name of the logged metric which determines when model is improving. Used by scheduler (plateau), checkpointer, and early stopping
  mode: ??? # "max" means higher metric value is better, can be also "min". Used by scheduler (plateau), checkpointer, and early stopping
  patience: 30 # how many validation epochs of not improving until training stops 
  val_and_save_every_n_steps: null

print_config: True # pretty print config at the start of the run using Rich library
ignore_warnings: True # disable python warnings if they annoy you
seed: 42 # seed for random number generators in pytorch, numpy and python.random