dark-whiteGPLM: Character-level protein language models for marine microalgae proteomes

dark-whiteGPLM: GPT-2 training dynamics on protein sequence datasets Character-level GPT-2 Small models trained from scratch on five protein-sequence corpora (White, Algae, Dark, Random Algae-biased, Random Full). What's included in this release Training code (model.py, train_.py, configurator.py) Training configs for all five datasets (config/train_*.py) Tokenizer vocabulary metadata (data/char_*/meta.pkl) Post-training analysis scripts and figures (analysis/) Random sequence generator (generate_random_sequences.py) Large files (hosted on HuggingFace) Checkpoints (~5 GB): https://huggingface.co/SarahDaakour/dark-whiteGPLM Training data (~1.1 GB): https://huggingface.co/datasets/SarahDaakour/dark-whiteGPLM-data See CHECKPOINTS.md for loading instructions.