Upload resolved_config.yaml
Browse files- resolved_config.yaml +130 -0
resolved_config.yaml
ADDED
|
@@ -0,0 +1,130 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
experiment:
|
| 2 |
+
seed: 42
|
| 3 |
+
deterministic_algorithms: true
|
| 4 |
+
optimizers:
|
| 5 |
+
- adamw
|
| 6 |
+
- nadamw
|
| 7 |
+
- lion
|
| 8 |
+
- muon
|
| 9 |
+
- sophia
|
| 10 |
+
model:
|
| 11 |
+
d_model: 512
|
| 12 |
+
n_heads: 8
|
| 13 |
+
n_layers: 6
|
| 14 |
+
mlp_hidden_size: 2048
|
| 15 |
+
dropout: 0.1
|
| 16 |
+
tie_embeddings: true
|
| 17 |
+
tokenizer:
|
| 18 |
+
type: byte_level_bpe
|
| 19 |
+
vocab_size: 16000
|
| 20 |
+
min_frequency: 2
|
| 21 |
+
special_tokens:
|
| 22 |
+
- <pad>
|
| 23 |
+
- <bos>
|
| 24 |
+
- <eos>
|
| 25 |
+
- <unk>
|
| 26 |
+
dataset:
|
| 27 |
+
name: browndw/human-ai-parallel-corpus
|
| 28 |
+
revision: b514ff64988d9e322fd81c5d70d69a38e78491f5
|
| 29 |
+
text_field: null
|
| 30 |
+
id_field: null
|
| 31 |
+
cache_directory: outputs/part2/cache/huggingface
|
| 32 |
+
split:
|
| 33 |
+
ratios:
|
| 34 |
+
train: 0.9
|
| 35 |
+
validation: 0.05
|
| 36 |
+
test: 0.05
|
| 37 |
+
training:
|
| 38 |
+
sequence_length: 256
|
| 39 |
+
batch_size: 32
|
| 40 |
+
number_of_workers: 0
|
| 41 |
+
token_budget: null
|
| 42 |
+
evaluation_fractions:
|
| 43 |
+
- 0.1
|
| 44 |
+
- 0.2
|
| 45 |
+
- 0.3
|
| 46 |
+
- 0.4
|
| 47 |
+
- 0.5
|
| 48 |
+
- 0.6
|
| 49 |
+
- 0.7
|
| 50 |
+
- 0.8
|
| 51 |
+
- 0.9
|
| 52 |
+
- 1.0
|
| 53 |
+
warmup_fraction: 0.05
|
| 54 |
+
minimum_lr_ratio: 0.1
|
| 55 |
+
device: cuda
|
| 56 |
+
precision: float16
|
| 57 |
+
save_checkpoints: true
|
| 58 |
+
checkpoint_fractions:
|
| 59 |
+
- 1.0
|
| 60 |
+
evaluation:
|
| 61 |
+
batch_size: 32
|
| 62 |
+
max_validation_examples: null
|
| 63 |
+
prefix_fraction: 0.5
|
| 64 |
+
generation_tokens: 64
|
| 65 |
+
bleu_examples: null
|
| 66 |
+
generation_batch_size: 256
|
| 67 |
+
saved_generations_per_checkpoint: 64
|
| 68 |
+
optimizers:
|
| 69 |
+
adamw:
|
| 70 |
+
lr: 0.0012
|
| 71 |
+
betas:
|
| 72 |
+
- 0.9
|
| 73 |
+
- 0.95
|
| 74 |
+
eps: 1.0e-08
|
| 75 |
+
weight_decay: 0.1
|
| 76 |
+
max_grad_norm: 1.0
|
| 77 |
+
nadamw:
|
| 78 |
+
lr: 0.0024
|
| 79 |
+
betas:
|
| 80 |
+
- 0.9
|
| 81 |
+
- 0.95
|
| 82 |
+
eps: 1.0e-08
|
| 83 |
+
weight_decay: 0.1
|
| 84 |
+
max_grad_norm: 1.0
|
| 85 |
+
lion:
|
| 86 |
+
lr: 0.0004
|
| 87 |
+
betas:
|
| 88 |
+
- 0.9
|
| 89 |
+
- 0.99
|
| 90 |
+
weight_decay: 0.3
|
| 91 |
+
max_grad_norm: 1.0
|
| 92 |
+
muon:
|
| 93 |
+
lr: 0.04
|
| 94 |
+
momentum: 0.95
|
| 95 |
+
weight_decay: 0.1
|
| 96 |
+
ns_steps: 5
|
| 97 |
+
ns_eps: 1.0e-07
|
| 98 |
+
adamw_lr: 0.0006
|
| 99 |
+
adamw_betas:
|
| 100 |
+
- 0.9
|
| 101 |
+
- 0.95
|
| 102 |
+
adamw_eps: 1.0e-08
|
| 103 |
+
adamw_weight_decay: 0.1
|
| 104 |
+
max_grad_norm: 1.0
|
| 105 |
+
sophia:
|
| 106 |
+
lr: 0.0003
|
| 107 |
+
betas:
|
| 108 |
+
- 0.965
|
| 109 |
+
- 0.99
|
| 110 |
+
rho: 0.04
|
| 111 |
+
weight_decay: 0.1
|
| 112 |
+
eps: 1.0e-15
|
| 113 |
+
hessian_interval: 10
|
| 114 |
+
max_grad_norm: 1.0
|
| 115 |
+
sweep:
|
| 116 |
+
budget_fraction: 0.1
|
| 117 |
+
lr_multipliers:
|
| 118 |
+
- 0.25
|
| 119 |
+
- 0.5
|
| 120 |
+
- 1.0
|
| 121 |
+
- 2.0
|
| 122 |
+
- 4.0
|
| 123 |
+
max_extensions: 3
|
| 124 |
+
output:
|
| 125 |
+
directory: outputs/part2
|
| 126 |
+
synthetic:
|
| 127 |
+
vocab_size: 32
|
| 128 |
+
train_blocks: 16
|
| 129 |
+
validation_blocks: 4
|
| 130 |
+
test_blocks: 4
|