configs/gmlp_small.yml

# GPT-2 pretraining setup
{
   # parallelism settings ( you will want to change these based on your cluster setup, ideally scheduling pipeline stages
   # across the node boundaries )
   "pipe_parallel_size": 1,
   "model_parallel_size": 1,
   "attention_config": [[["gmlp"], "all"]],


   # model settings
   "num_layers": 12,
   "hidden_size": 768, # gmlp d_ff defaults to hidden_size * 4
   "gmlp_attn_dim": 64,
   "num_attention_heads": 12, # this has no effect with gmlp - and amlp defaults to single head attention.
   "seq_length": 2048,
   "max_position_embeddings": 2048,
   "norm": "layernorm",
   "pos_emb": "none",
   "no_weight_tying": true,

   # optimizer settings
   "optimizer": {
     "type": "Adam",
     "params": {
       "lr": 0.0006,
       "betas": [0.9, 0.999],
       "eps": 1.0e_8,
     }
   },

   # batch / data settings
   "train_micro_batch_size_per_gpu": 4,
   "data_impl": "mmap",
   "split": "949,50,1",

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

   # regularization
   "gradient_clipping": 1.0,
   "weight_decay": 0.1,
   "hidden_dropout": 0.0,
   "attention_dropout": 0.0,

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

   # misc. training settings
   "train_iters": 320000,
   "lr_decay_iters": 320000,
   "distributed_backend": "nccl",
   "lr_decay_style": "cosine",
   "warmup": 0.01,
   "checkpoint_factor": 10000,
   "eval_interval": 1000,
   "eval_iters": 10,

   # logging
   "log_interval": 100,
   "steps_per_print": 10,
   "keep_last_n_checkpoints": 4,
   "wall_clock_breakdown": true,
}