Hugging Face's logo

nvidia
/
Nemotron-4-340B-Reward

NeMo
English
nvidia
steerlm
reward model
Model card Files Community
7
Nemotron-4-340B-Reward
File size: 3,205 Bytes 
08c78f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
154bbb8
08c78f2
 
 
 
 
 
 
 
154bbb8
08c78f2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
 
mcore_gpt: true
micro_batch_size: 1
global_batch_size: 128
tensor_model_parallel_size: 8
pipeline_model_parallel_size: 24
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: full
activations_checkpoint_method: uniform
activations_checkpoint_num_layers: 1
num_micro_batches_with_partial_activation_checkpoints: null
activations_checkpoint_layers_per_pipeline: null
sequence_parallel: false
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: 7.0e-07
  weight_decay: 0.1
  betas:
  - 0.9
  - 0.98
  sched:
    name: CosineAnnealing
    warmup_steps: 10
    constant_steps: 0
    min_lr: 7.0e-07
  bucket_cap_mb: 200
  overlap_grad_sync: false
  contiguous_grad_buffer: true
precision: bf16
reward_model_type: regression
regression:
  num_attributes: 9
  merge_attributes: false
  attribute_weights: null
  loss_mask_val: -100
output_sequence: false
use_avg_pool: false
force_head_dtype: float32
data:
  data_impl: jsonl
  splits_string: null
  seq_length: 4096
  skip_warmup: true
  num_workers: 2
  dataloader_type: single
  reset_position_ids: false
  reset_attention_mask: false
  eod_mask_loss: false
  index_mapping_dir: null
  data_prefix:
    train:
    - /dataset/train_2_epochs_reg.jsonl
    validation:
    - /dataset/val_2_epochs_reg.jsonl
    test:
    - /dataset/val_2_epochs_reg.jsonl
target: nemo_aligner.models.nlp.gpt.megatron_gpt_regression_reward_model.MegatronGPTRegressionRewardModel
nemo_version: 1.22.0