340M-Parameter LLM Trained Exclusively on Pre-1900 Texts, Weights Released

A 340M-parameter base model trained solely on historical texts is available on HuggingFace at croqaz/vintage-LLM-340m-v1-base with open-source training code. The project demonstrates full custom pretraining and fine-tuning pipelines built independently, useful as a reference implementation for domain-specific pretraining.