File size: 1,488 Bytes
2e2007c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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

train:
  experiment_name: "isen1"
  lr: 2.5e-3
  accum_steps: 6
  warmup_steps: 10000
  max_steps: 100000
  eval_steps: 1000
  max_checkpoints: 10
  precision: "bfloat16" # or float16 with an older GPU
  enable_torch_compile: true 
  checkpoint_strategy: best
  early_stopping_patience: 0
  early_stopping_metric: chrf
  use_ema: true
  ema_decay: 0.9999
  ema_start_step: 10000
  z_loss_coeff: 0.0005
  weight_decay_embeddings: false
  scheduler_type: "cosine"
  
data:
  src_lang: "is"
  tgt_lang: "en"
  src_dev_path: "quickmt-valid.is-en.is"
  tgt_dev_path: "quickmt-valid.is-en.en"
  input_sentence_size: 10000000
  max_tokens_per_batch: 20000
  buffer_size: 40000
  num_workers:  4
  prefetch_factor: 128
  pad_multiple: 1
  corpora: 
    - src_file: "quickmt-train.is-en.is"
      tgt_file: "quickmt-train.is-en.en"
      weight: 10
      start_step: 0 
    - src_file: "finetranslations-sample-is-en.is"
      tgt_file: "finetranslations-sample-is-en.en"
      weight: 4
      start_step: 0
      stop_step: 80000
    - src_file: "newscrawl2024-en-backtranslated-is.is"
      tgt_file: "newscrawl2024-en-backtranslated-is.en"
      start_step: 0
      weight: 5
      stop_step: 80000 

model:
  d_model: 768
  enc_layers: 12
  dec_layers: 2
  n_heads: 12
  ffn_dim: 4096
  max_len: 256
  vocab_size_src: 32000
  vocab_size_tgt: 32000
  norm_type: "rmsnorm"
  mlp_type: "gated"
  activation: "silu"
  ff_bias: false
  layernorm_eps: 1.0e-5
  dropout: 0.1

export:
  max_len: 256