FineDNA: Open dataset with 300B+ tokens for DNA-specialized LLM training

·2026.11.01 16:24

Key point

The dataset includes 300B+ tokens of genomes, 200K SFT conversations, and 98K RL prompts, with code and recipes available for other models.

FineDNA is released as the largest dataset for teaching LLMs DNA, containing over 300B tokens of genomes and DNA-rich text for midtraining.
The dataset supports the creation of Carbon-X, described as the strongest open language model on DNA, derived from Qwen3-14B.
It includes approximately 200K conversations for supervised fine-tuning (SFT) and roughly 98K verifiable prompts for reinforcement learning (RL).
Code and training recipes are open-sourced, allowing the application of this data to other LLMs.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.