tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Add eval_batch_size for evaluation

Browse Source
This commit is contained in:
NanoCode012
2023-05-06 22:21:24 +09:00
parent c0f50d9c61
commit 0e74b6402e
2 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
README.md
View File

@@ -85,6 +85,7 @@ output_dir: ./completed-model
# training hyperparameters # training hyperparameters
batch_size: 8 batch_size: 8
micro_batch_size: 2 micro_batch_size: 2
eval_batch_size: 2
num_epochs: 3 num_epochs: 3
warmup_steps: 100 warmup_steps: 100
learning_rate: 0.00003 learning_rate: 0.00003

1
src/axolotl/utils/trainer.py
View File

@@ -47,6 +47,7 @@ def setup_trainer(cfg, train_dataset, eval_dataset, model, tokenizer):
training_args = transformers.TrainingArguments( training_args = transformers.TrainingArguments(
per_device_train_batch_size=cfg.micro_batch_size, per_device_train_batch_size=cfg.micro_batch_size,
per_device_eval_batch_size=cfg.eval_batch_size,
gradient_accumulation_steps=cfg.gradient_accumulation_steps, gradient_accumulation_steps=cfg.gradient_accumulation_steps,
num_train_epochs=cfg.num_epochs, num_train_epochs=cfg.num_epochs,
learning_rate=cfg.learning_rate, learning_rate=cfg.learning_rate,