File size: 1,259 Bytes
5df2c77
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# okto_version: "1.2"

# Teste 3: T5 com CONTROL - Decisões Automáticas
# Modelo: google/t5-small
# Objetivo: Testar bloco CONTROL com lógica condicional

PROJECT "test_t5_control"
DESCRIPTION "Teste T5 com bloco CONTROL - decisões automáticas durante treino"

ENV {
  accelerator: "gpu"
  min_memory: "4GB"
  precision: "fp16"
  backend: "oktoseek"
  install_missing: true
}

DATASET {
  train: "dataset/train.jsonl"
  validation: "dataset/val.jsonl"
}

MODEL {
  base: "t5-small"
  device: "auto"
}

TRAIN {
  epochs: 5
  batch_size: 8
  learning_rate: 0.0001
  device: "auto"
}

CONTROL {
  on_step_end {
    LOG loss
  }
  
  on_epoch_end {
    SAVE model
    LOG "Epoch completed"
  }
  
  validate_every: 100
  
  IF loss > 2.0 {
    SET LR = 0.00005
    LOG "High loss detected - reducing learning rate"
  }
  
  IF val_loss > 2.5 {
    STOP_TRAINING
    LOG "Validation loss too high - stopping training"
  }
  
  IF accuracy < 0.4 {
    DECREASE LR BY 0.5
    LOG "Low accuracy - decreasing learning rate by 50%"
  }
  
  WHEN gpu_memory < 8GB {
    SET batch_size = 4
    LOG "Low GPU memory - reducing batch size"
  }
  
  EVERY 500 steps {
    SAVE checkpoint
    LOG "Checkpoint saved"
  }
}

EXPORT {
  format: ["okm"]
  path: "export/"
}