Add eval_batch_size for evaluation
Browse files- README.md +1 -0
- src/axolotl/utils/trainer.py +1 -0
README.md
CHANGED
@@ -85,6 +85,7 @@ output_dir: ./completed-model
|
|
85 |
# training hyperparameters
|
86 |
batch_size: 8
|
87 |
micro_batch_size: 2
|
|
|
88 |
num_epochs: 3
|
89 |
warmup_steps: 100
|
90 |
learning_rate: 0.00003
|
|
|
85 |
# training hyperparameters
|
86 |
batch_size: 8
|
87 |
micro_batch_size: 2
|
88 |
+
eval_batch_size: 2
|
89 |
num_epochs: 3
|
90 |
warmup_steps: 100
|
91 |
learning_rate: 0.00003
|
src/axolotl/utils/trainer.py
CHANGED
@@ -47,6 +47,7 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
|
|
47 |
|
48 |
training_args = transformers.TrainingArguments(
|
49 |
per_device_train_batch_size=cfg.micro_batch_size,
|
|
|
50 |
gradient_accumulation_steps=cfg.gradient_accumulation_steps,
|
51 |
num_train_epochs=cfg.num_epochs,
|
52 |
learning_rate=cfg.learning_rate,
|
|
|
47 |
|
48 |
training_args = transformers.TrainingArguments(
|
49 |
per_device_train_batch_size=cfg.micro_batch_size,
|
50 |
+
per_device_eval_batch_size=cfg.eval_batch_size,
|
51 |
gradient_accumulation_steps=cfg.gradient_accumulation_steps,
|
52 |
num_train_epochs=cfg.num_epochs,
|
53 |
learning_rate=cfg.learning_rate,
|