Tonic commited on
Commit
c68717e
·
1 Parent(s): fc29a51

adds parameters to medical config

Browse files
config/train_gpt_oss_medical_o1_sft.py CHANGED
@@ -56,12 +56,13 @@ config = GPTOSSEnhancedCustomConfig(
56
  # TRAINING HYPERPARAMETERS
57
  # ============================================================================
58
  num_train_epochs=1.0,
59
- batch_size=2,
60
- gradient_accumulation_steps=8,
61
  learning_rate=2e-4,
62
  min_lr=2e-5,
63
  weight_decay=0.01,
64
  warmup_ratio=0.03,
 
65
  max_grad_norm=1.0,
66
 
67
  # Sequence length
@@ -113,7 +114,7 @@ config = GPTOSSEnhancedCustomConfig(
113
  # LOGGING & EVAL
114
  # ============================================================================
115
  eval_strategy="steps",
116
- eval_steps=200,
117
  logging_steps=10,
118
  save_strategy="steps",
119
  save_steps=500,
@@ -124,8 +125,8 @@ config = GPTOSSEnhancedCustomConfig(
124
  load_best_model_at_end=False,
125
  eval_accumulation_steps=2,
126
  eval_batch_size=1,
127
- eval_ratio=0.01,
128
- test_ratio=0.01,
129
 
130
  # ============================================================================
131
  # MONITORING & HUB
 
56
  # TRAINING HYPERPARAMETERS
57
  # ============================================================================
58
  num_train_epochs=1.0,
59
+ batch_size=4,
60
+ gradient_accumulation_steps=4,
61
  learning_rate=2e-4,
62
  min_lr=2e-5,
63
  weight_decay=0.01,
64
  warmup_ratio=0.03,
65
+ warmup_steps=50,
66
  max_grad_norm=1.0,
67
 
68
  # Sequence length
 
114
  # LOGGING & EVAL
115
  # ============================================================================
116
  eval_strategy="steps",
117
+ eval_steps=100,
118
  logging_steps=10,
119
  save_strategy="steps",
120
  save_steps=500,
 
125
  load_best_model_at_end=False,
126
  eval_accumulation_steps=2,
127
  eval_batch_size=1,
128
+ eval_ratio=0.001,
129
+ test_ratio=0.0005,
130
 
131
  # ============================================================================
132
  # MONITORING & HUB