20B.yml

# DISCLAIMER: This is the configuration file for the GPT-NeoX-20B model as it was trained on 96x 40GB A100
# GPUs. Depending on your system configuration, you may need to change some parameters in order to fit
# the model in memory.

{
  # Tokenizer /  checkpoint settings - you will need to change these to the location you have them saved in
  "vocab_file": "./20B_checkpoints/20B_tokenizer.json",
  "save": "./20B_checkpoints",
  "load": "./20B_checkpoints",

  # If finetuning, edit the following to the location of your finetuning dataset:
  "data_path": "./data/pile_20B_tokenizer/pile_20B_tokenizer_text_document",

  # parallelism settings ( you will want to change these based on your cluster setup, ideally scheduling pipeline stages
  # across the node boundaries )
  "pipe_parallel_size": 4,
  "model_parallel_size": 2,

  # model settings
  "num_layers": 44,
  "hidden_size": 6144,
  "num_attention_heads": 64,
  "seq_length": 2048,
  "max_position_embeddings": 2048,
  "norm": "layernorm",
  "pos_emb": "rotary",
  "rotary_pct": 0.25,
  "no_weight_tying": true,
  "gpt_j_residual": true,
  "output_layer_parallelism": "column",
  "scaled_upper_triang_masked_softmax_fusion": true,
  "bias_gelu_fusion": true,
  "rope_fusion": false,
  "layernorm_fusion": false,

  # init methods
  "init_method": "small_init",
  "output_layer_init_method": "wang_init",

  # optimizer settings
  "optimizer": {
    "type": "Adam",
    "params": {
      "lr": 0.97e-4,
      "betas": [0.9, 0.95],
      "eps": 1.0e-8,
      }
      },

  "min_lr": 0.97e-5,

  # for all zero_optimization options, see https://www.deepspeed.ai/docs/config-json/#zero-optimizations-for-fp16-training
  "zero_optimization": {
  "stage": 1,
  "allgather_partitions": True,
  "allgather_bucket_size": 1260000000,
  "overlap_comm": True,
  "reduce_scatter": True,
  "reduce_bucket_size": 1260000000,
  "contiguous_gradients": True,
  },

  # batch / data settings (assuming 96 GPUs)
  "train_micro_batch_size_per_gpu": 4,
  "gradient_accumulation_steps": 32,
  "data_impl": "mmap",
  "split": "995,4,1",

  # activation checkpointing
  "checkpoint_activations": true,
  "checkpoint_num_layers": 1,
  "partition_activations": false,
  "synchronize_each_layer": true,

  # regularization
  "gradient_clipping": 1.0,
  "weight_decay": 0.01,
  "hidden_dropout": 0,
  "attention_dropout": 0,

  # precision settings
  "fp16": {
    "fp16": true,
    "enabled": true,
    "loss_scale": 0,
    "loss_scale_window": 1000,
    "initial_scale_power": 12,
    "hysteresis": 2,
    "min_loss_scale": 1
    },

  # misc. training settings
  "train_iters": 150000,
  "lr_decay_iters": 150000,

  "distributed_backend": "nccl",
  "lr_decay_style": "cosine",
  "warmup": 0.01,
  "checkpoint_factor": 500, # this variable previously called `save-interval`
  "eval_interval": 1000,
  "eval_iters": 10,

  # logging
  "log_interval": 2,
  "steps_per_print": 2,
  "wall_clock_breakdown": false,

  ### NEW DATA: ####
  "tokenizer_type": "HFTokenizer",
  "tensorboard-dir": "./tensorboard",
  "log_dir": "./logs",

}