Skip to content

Dataset · HuggingFaceBio

Carbon-X training data for DNA reading

HuggingFaceBio/FineDNA

Three training stages for Carbon-X: mid-training, SFT, and RL on DNA texts, dialogues, and prompts.

What's inside

FineDNA comprises training data for Carbon-X, the model that taught Qwen3-14B to read DNA, covering all three training stages. Mid-training contains 514.30 billion tokens across 23 sources with 302 billion read during training; SFT contains 203,923 dialogues totaling 230.3 million tokens; RL contains 98,433 prompts totaling 61.1 million tokens. English language, text modality.

What you can do with it

  • Training LLMs for DNA tasks
  • Fine-tuning Qwen3-14B on SFT dialogues
  • Testing RL pipelines on genomics prompts
Tasks
Text generation
Languages
English
Size
100M–1B rows
Downloads this month
13.7K
Likes
70

How to load it

from datasets import load_dataset

ds = load_dataset("HuggingFaceBio/FineDNA")