A lean, no-frills text model built for research transparency rather than polish. It handles basic language tasks but lacks the instruction-following refinement of fine-tuned models — it generates continuations, not answers. Its deduped training data makes it a cleaner baseline for studying how language models learn and behave.