reference/code/flm/configs/config.yaml

defaults:
  - _self_
  - /callbacks: [checkpoint_every_n_steps, checkpoint_monitor, learning_rate_monitor]
  - /data: openwebtext  # switch to synthetic_alpha8 for debugging: set to /data: synthetic_alpha8
  - /model: small
  - /strategy: ddp
  - /noise: log-linear
  - /lr_scheduler: constant_warmup
  - /prior: none
  - /algo: flm

mode: train  # train / ppl_eval / sample_eval

# di4c configs:
is_di4c: False
is_di4c_deterministic: True

seed: 1

loader:
  global_batch_size: 512
  eval_global_batch_size: ${.global_batch_size}
  # Note: batch_size and eval_batch_size are **per machine**
  batch_size: ${div_up:${.global_batch_size}, ${eval:${trainer.devices} * ${trainer.num_nodes}}}
  eval_batch_size: ${div_up:${.eval_global_batch_size}, ${eval:${trainer.devices} * ${trainer.num_nodes}}}
  num_workers: 4
  pin_memory: True

sampling:
  predictor: ancestral  # ancestral_cache (only for MDLM), ancestral, analytic
  steps: [1024]
  noise_removal: ancestral  # 'ancestral', 'greedy', 'none', 'meanflow'
  use_float64: True
  p_nucleus: 1.0
  num_sample_batches: 1  # Total samples: `num_gpus` * `loader.eval_batch_size` * num_sample_batches
  num_sample_log: 2
  semi_ar: False
  stride_length: 1
  num_strides: 1
  num_reflow_samples: 50000 # for generating reflow dataset
  temperature: 1.0
  solver: euler 
  gamma: 0.0

training:
  loss_type: flow
  pred_type: x0
  ema: 0.9999
  antithetic_sampling: True
  importance_sampling: False
  sampling_eps: 1e-3
  change_of_variables: False
  loss_precision: 'bf16'  # bf16, float32, float64
  finetune_path: ''
  not_sampling_t: False
  load_ema: False  # load EMA parameters when finetuning
  t_curriculum_steps: 100000
  t_curriculum_min: 0.00
  t_curriculum_max: 1.00

eval:
  checkpoint_path: ''  # Used to evaluate a checkpoint after training.
  disable_ema: False
  ema_decay: null  # If set, select matching EMA from checkpoint
  compute_generative_perplexity: True
  perplexity_batch_size: 8
  compute_perplexity_on_sanity: False
  gen_ppl_eval_model_name_or_path: gpt2-large  # gpt2-large, meta-llama/Llama-2-7b-hf, meta-llama/Llama-3.1-8B
  generate_samples: True
  generated_samples_path: ${cwd:}/samples.json

optim:
  weight_decay: 0
  lr: 3e-4
  beta1: 0.9
  beta2: 0.999
  eps: 1e-8
  ln_tune: none

trainer:
  _target_: lightning.Trainer
  accelerator: cuda
  num_nodes: 1
  devices: ${device_count:}
  accumulate_grad_batches: ${div_up:${loader.global_batch_size}, ${eval:${trainer.devices} * ${loader.batch_size} * ${trainer.num_nodes}}}
  gradient_clip_val: 1.0
  precision: 'bf16'
  num_sanity_val_steps: 2
  max_steps: 1_000_000
  log_every_n_steps: 100
  limit_train_batches: 1.0   # train on full dataset, can be used to toggle quick run
  limit_val_batches: 1.0     # validate on full dataset, can be used to toggle quick run
  val_check_interval: 5000

wandb:
  project: flm
  notes: FLM
  group: null
  job_type: null
  name: null
  id: ${.name}_${seed}
  tags:
    - ${data.train}
    - ${algo.name}

hydra:
  run:
    dir: ./outputs/${data.train}/${now:%Y.%m.%d}/${now:%H%M%S}
  job:
    chdir: true

checkpointing:
  # Use custom `save_dir` if, e.g., saving to S3 bucket, otherwise leave this parameter as is
  save_dir: ${cwd:}
  # Note: `checkpoints` path should correspond to `checkpoint_every_n_steps.dirpath`
  resume_from_ckpt: false
  resume_ckpt_path: ${.save_dir}/checkpoints/last.ckpt