Setting up the reward trainer
Your project continues and you now have the model and config objects ready to start training the reward model.
The training and evaluation datasets have been preloaded as train_data and eval_data. The RewardTrainer has been imported from trl.
This exercise is part of the course
Reinforcement Learning from Human Feedback (RLHF)
Exercise instructions
- Initialize the
RewardTrainer()by assigning the model, tokenizer, training dataset, evaluation dataset, and reward configuration to its attributes.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
tokenizer = AutoTokenizer.from_pretrained("openai-gpt")
model = AutoModelForSequenceClassification.from_pretrained('openai-gpt')
config = RewardConfig(output_dir='output_dir', max_length=60)
# Initialize the reward trainer
trainer = ____