pythia-dedup-125M.yml

{
  "pipe-parallel-size": 1,
  "model-parallel-size": 1,

  # model settings
  "num-layers": 12,
  "hidden-size": 768,
  "num-attention-heads": 12,
  "seq-length": 2048,
  "max-position-embeddings": 2048,
  "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,

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

  "optimizer": {
    "type": "Adam",
    "params": {
      "lr": 0.0006,
      "betas": [0.9, 0.95],
      "eps": 1.0e-8,
    }
  },
  "min_lr": 0.00006,

  "zero_optimization": {
    "stage": 1,
    "allgather_partitions": True,
    "allgather_bucket_size": 500000000,
    "overlap_comm": True,
    "reduce_scatter": True,
    "reduce_bucket_size": 500000000,
    "contiguous_gradients": True,
    "cpu_offload": False
  },

  "train_micro_batch_size_per_gpu": 16,
  "gas": 2,
  "data-impl": "mmap",
  "num_workers": 1,

  # activation checkpointing
  "checkpoint-activations": true,
  "checkpoint-num-layers": 1,
  "partition-activations": true,
  "synchronize-each-layer": true,

  # regularization
  "gradient_clipping": 1.0,
  "weight-decay": 0.1,
  "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,
  },

  "train-iters": 71500,
  "lr-decay-iters": 71250,
  "distributed-backend": "nccl",
  "lr-decay-style": "cosine",
  "warmup": 0.01,
  "save-interval": 250,
  "eval-interval": 71500,
  "eval-iters": 10,
  #"eval-tasks": ["logiqa", "arc_challenge", "arc_easy", "lambada", "winogrande", "wsc", "piqa", "sciq"], 

  "log-interval": 10,
  "steps_per_print": 10,
  "wall_clock_breakdown": true,

  "save": "/fsx/hailey/pythia/ckpts/125M_dedup",
  "load": "/fsx/hailey/pythia/ckpts/125M_dedup",

  "train-data-paths": ["/fsx/pile_deduped/pile_0.87_deduped_text_document"],
  "valid-data-paths": ["/fsx/pile_deduped/pile_0.87_deduped_text_document"],
  "test-data-paths": ["/fsx/pile_deduped/pile_0.87_deduped_text_document"],

  "tokenizer-type": "HFTokenizer",
  "vocab-file": "/fsx/pile/20B_tokenizer.json",

  "tensorboard-dir": "/fsx/code-fim/FIMlogs/1.3B-AR-Pile-9-6-22-rotary-1MtokBS",
  "log-dir": "/fsx/code-fim/FIMlogs/1.3B-AR-Pile-9-6-22-rotary-1MtokBS", 

  "use_wandb": true,
  "wandb_group": "125M Dedup",
  "wandb_team": "eleutherai",
  "wandb_project": "pythia",

  "launcher": "openmpi",
  "deepspeed_mpi": true,
}