Delete config.yaml
Browse files- config.yaml +0 -116
config.yaml
DELETED
|
@@ -1,116 +0,0 @@
|
|
| 1 |
-
checkpoints:
|
| 2 |
-
checkpoint_interval: 191
|
| 3 |
-
checkpoints_path: /fsx/marcmarone/upcycle/artifacts/nanotron_checkpoints/upcycle-670m-c4-seed42
|
| 4 |
-
checkpoints_path_is_shared_file_system: true
|
| 5 |
-
load_lr_scheduler: true
|
| 6 |
-
load_optimizer: true
|
| 7 |
-
resume_checkpoint_path: /fsx/marcmarone/upcycle/artifacts/nanotron_checkpoints/upcycle-670m-c4-seed42/4966
|
| 8 |
-
save_final_state: true
|
| 9 |
-
save_initial_state: false
|
| 10 |
-
data_stages:
|
| 11 |
-
- data:
|
| 12 |
-
dataset:
|
| 13 |
-
dataset_overwrite_cache: false
|
| 14 |
-
dataset_processing_batch_size: 1000
|
| 15 |
-
dataset_processing_num_proc_per_process: 16
|
| 16 |
-
dataset_shuffle: false
|
| 17 |
-
hf_dataset_config_name: default
|
| 18 |
-
hf_dataset_or_datasets: /fsx/marcmarone/upcycle/prepared/c4-megadocsrewritten-6.7b-retok
|
| 19 |
-
hf_dataset_splits: train
|
| 20 |
-
text_column_name: text
|
| 21 |
-
num_loading_workers: 2
|
| 22 |
-
seed: 42
|
| 23 |
-
name: FineWeb Stage (epoch 1)
|
| 24 |
-
sequence_length: 4096
|
| 25 |
-
start_training_step: 1
|
| 26 |
-
general:
|
| 27 |
-
benchmark_csv_path: /fsx/marcmarone/upcycle/artifacts/nanotron_benchmarks/upcycle-670m-c4-seed42.csv
|
| 28 |
-
consumed_train_samples: 1626368
|
| 29 |
-
ignore_sanity_checks: true
|
| 30 |
-
project: upcycle
|
| 31 |
-
run: upcycle-670m-c4-seed42
|
| 32 |
-
seed: 42
|
| 33 |
-
step: 6353
|
| 34 |
-
lighteval: null
|
| 35 |
-
logging:
|
| 36 |
-
iteration_step_info_interval: 1
|
| 37 |
-
log_level: info
|
| 38 |
-
log_level_replica: info
|
| 39 |
-
metrics_logging: null
|
| 40 |
-
model:
|
| 41 |
-
ddp_bucket_cap_mb: 25
|
| 42 |
-
dtype: bfloat16
|
| 43 |
-
init_method:
|
| 44 |
-
scaling_method: NUM_LAYERS
|
| 45 |
-
std: 0.02
|
| 46 |
-
make_vocab_size_divisible_by: 1
|
| 47 |
-
model_config:
|
| 48 |
-
_attn_implementation: flash_attention_2
|
| 49 |
-
attention_bias: false
|
| 50 |
-
bos_token_id: 1
|
| 51 |
-
eos_token_id: 2
|
| 52 |
-
hidden_act: silu
|
| 53 |
-
hidden_size: 1024
|
| 54 |
-
initializer_range: 0.02
|
| 55 |
-
intermediate_size: 7168
|
| 56 |
-
is_llama_config: true
|
| 57 |
-
max_position_embeddings: 4096
|
| 58 |
-
num_attention_heads: 16
|
| 59 |
-
num_hidden_layers: 24
|
| 60 |
-
num_key_value_heads: 8
|
| 61 |
-
pad_token_id: 0
|
| 62 |
-
pretraining_tp: 1
|
| 63 |
-
rms_norm_eps: 1.0e-05
|
| 64 |
-
rope_interleaved: false
|
| 65 |
-
rope_scaling: null
|
| 66 |
-
rope_theta: 10000.0
|
| 67 |
-
tie_word_embeddings: false
|
| 68 |
-
use_cache: true
|
| 69 |
-
vocab_size: 32000
|
| 70 |
-
z_loss_coefficient: 0.0001
|
| 71 |
-
z_loss_enabled: false
|
| 72 |
-
optimizer:
|
| 73 |
-
accumulate_grad_in_fp32: true
|
| 74 |
-
clip_grad: 1.0
|
| 75 |
-
learning_rate_scheduler:
|
| 76 |
-
learning_rate: 0.001
|
| 77 |
-
lr_decay_starting_step: null
|
| 78 |
-
lr_decay_steps: 6353
|
| 79 |
-
lr_decay_style: cosine
|
| 80 |
-
lr_warmup_steps: 256
|
| 81 |
-
lr_warmup_style: linear
|
| 82 |
-
min_decay_lr: 0.0001
|
| 83 |
-
optimizer_factory:
|
| 84 |
-
adam_beta1: 0.9
|
| 85 |
-
adam_beta2: 0.95
|
| 86 |
-
adam_eps: 1.0e-08
|
| 87 |
-
name: adamW
|
| 88 |
-
torch_adam_is_fused: true
|
| 89 |
-
weight_decay: 1.6
|
| 90 |
-
weight_decay_exclude_named_params: []
|
| 91 |
-
zero_stage: 0
|
| 92 |
-
parallelism:
|
| 93 |
-
context_parallel_size: 1
|
| 94 |
-
dp: 8
|
| 95 |
-
expert_parallel_size: 1
|
| 96 |
-
pp: 1
|
| 97 |
-
pp_engine: 1f1b
|
| 98 |
-
recompute_layer: false
|
| 99 |
-
tp: 1
|
| 100 |
-
tp_linear_async_communication: true
|
| 101 |
-
tp_mode: REDUCE_SCATTER
|
| 102 |
-
tp_recompute_allgather: true
|
| 103 |
-
profiler: null
|
| 104 |
-
s3_upload: null
|
| 105 |
-
tokenizer:
|
| 106 |
-
tokenizer_max_length: null
|
| 107 |
-
tokenizer_name_or_path: /fsx/marcmarone/upcycle/tokenizer/llama2
|
| 108 |
-
tokenizer_revision: null
|
| 109 |
-
tokens:
|
| 110 |
-
batch_accumulation_per_replica: 8
|
| 111 |
-
limit_test_batches: 0
|
| 112 |
-
limit_val_batches: 0
|
| 113 |
-
micro_batch_size: 4
|
| 114 |
-
sequence_length: 4096
|
| 115 |
-
train_steps: 6353
|
| 116 |
-
val_check_interval: -1
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|