mcore_gpt: true
micro_batch_size: 1
global_batch_size: 2308
tensor_model_parallel_size: 8
pipeline_model_parallel_size: 2
virtual_pipeline_model_parallel_size: null
encoder_seq_length: 4096
max_position_embeddings: 4096
num_layers: 96
hidden_size: 18432
ffn_hidden_size: 73728
num_attention_heads: 96
init_method_std: 0.0063
use_scaled_init_method: true
hidden_dropout: 0.0
attention_dropout: 0.0
ffn_dropout: 0.0
kv_channels: null
apply_query_key_layer_scaling: true
normalization: layernorm1p
layernorm_epsilon: 1.0e-05
do_layer_norm_weight_decay: false
make_vocab_size_divisible_by: 128
pre_process: true
post_process: true
persist_layer_norm: true
bias: false
activation: squared-relu
headscale: false
transformer_block_type: pre_ln
openai_gelu: false
normalize_attention_scores: true
position_embedding_type: rope
rotary_percentage: 0.5
attention_type: multihead
share_embeddings_and_output_weights: false
num_query_groups: 8
tokenizer:
  library: sentencepiece
  type: null
  model: nemo:d6b0ba93e9734b138f5fc61f5652efbd_nemotron_2_256k.model
  delimiter: null
  vocab_file: null
  merge_file: null
  sentencepiece_legacy: false
  tokenizer_model: nemo:29e0db5f7dd14bcf9f32727ff482502b_nemotron_2_256k.model
native_amp_init_scale: 4294967296
native_amp_growth_interval: 1000
hysteresis: 2
fp32_residual_connection: false
fp16_lm_cross_entropy: false
megatron_amp_O2: true
grad_allreduce_chunk_size_mb: 125
grad_div_ar_fusion: true
gradient_accumulation_fusion: false
bias_activation_fusion: false
bias_dropout_add_fusion: false
masked_softmax_fusion: true
seed: 1234
resume_from_checkpoint: null
use_cpu_initialization: false
onnx_safe: false
apex_transformer_log_level: 30
gradient_as_bucket_view: true
sync_batch_comm: false
activations_checkpoint_granularity: null
activations_checkpoint_method: null
activations_checkpoint_num_layers: null
num_micro_batches_with_partial_activation_checkpoints: null
activations_checkpoint_layers_per_pipeline: null
sequence_parallel: true
transformer_engine: false
fp8: false
fp8_e4m3: false
fp8_hybrid: false
fp8_margin: 0
fp8_interval: 1
fp8_amax_history_len: 1
fp8_amax_compute_algo: most_recent
reduce_amax: true
use_emha: false
optim:
  name: distributed_fused_adam
  lr: '1.0e-4'
  weight_decay: 0.1
  betas:
  - 0.9
  - 0.95
  sched:
    name: CosineAnnealing
    warmup_steps: 500
    constant_steps: 50000
    min_lr: '1.0e-7'
precision: bf16
target: nemo.collections.nlp.models.language_modeling.megatron_gpt_model.MegatronGPTModel
nemo_version: 1.15.0rc0