Provides the three-stage training data for Carbon-X, a DNA language model: large-scale genomic and biology-text data for mid-training, plus supervised fine-tuning and reinforcement-learning examples.