trlX is a distributed training framework for fine-tuning large language models with reinforcement...