🧠 两小时从 0 训出你的 64M 大模型。Train your own 64M LLM from scratch in 2 hours.
update trl
gongjy02adb7bc0de3
diff --git a/5-dpo_train.py b/5-dpo_train.pyindex 3ae3df7..ef17e2c 100644--- a/5-dpo_train.py+++ b/5-dpo_train.py@@ -3,7 +3,7 @@ import warningsos.environ['CUDA_VISIBLE_DEVICES'] = '0'from transformers import TrainingArguments, AutoModelForCausalLM, AutoTokenizer-from trl import DPOTrainer+from trl import DPOConfig, DPOTrainerfrom datasets import load_datasetwarnings.filterwarnings('ignore')@@ -23,7 +23,7 @@ def init_model():if __name__ == '__main__':model, tokenizer = init_model()- training_args = TrainingArguments(+ training_config = DPOConfig(output_dir="./minimind_dpo",per_device_train_batch_size=1,remove_unused_columns=False,@@ -38,7 +38,7 @@ if __name__ == '__main__':dpo_trainer = DPOTrainer(model,ref_model=None,- args=training_args,+ args=training_config,beta=0.1,train_dataset=train_dataset['train'],tokenizer=tokenizer,diff --git a/requirements.txt b/requirements.txtindex 26c8d29..342675d 100644--- a/requirements.txt+++ b/requirements.txt@@ -23,6 +23,6 @@ torch==2.1.2transformers==4.44.0jinja2==3.1.2jsonlines==4.0.0-trl==0.8.6+trl==0.11.3ujson==5.1.0wandb==0.18.3