Oriel — AI & technology worth noticing, every day.

HuggingFaceBio/FineDNA

Dataset · https://huggingface.co/datasets/HuggingFaceBio/FineDNA

FineDNA FineDNA is the training data of Carbon-X, the model that taught Qwen3-14B to read DNA, for all three of its training stages. Each stage is a group of configs: stage configs size starts from Mid-training mid_*, one per source (23) 514.30B tokens, of which the run read 302B Qwen/Qwen3-14B SFT sft_mix, and sft_* for each of its 8 domains 203,923 conversations, 230.3M tokens HuggingFaceBio/Carbon-X-Base RL rl_mix 98,433 prompts, 61.1M tokens… See the full description on the dataset page: https://huggingface.co/datasets/HuggingFaceBio/FineDNA.

First seen by Oriel on .

AI summary

FineDNA releases the complete Carbon-X training data for teaching Qwen3-14B to read DNA, covering weighted mid-training sources, supervised fine-tuning conversations, and verifiable reinforcement-learning prompts.

  • Mid-training contains 514.30B unique tokens, with Carbon-X reading 302B across DNA, biology text, English, and instruction data using source-specific weights.
  • DNA sources include GENERator eukaryote windows, OpenGenome2 mRNA and prokaryote genomes, formatted as 6-bp-aligned sequences wrapped in <dna> tags.
  • SFT provides 203,923 conversations and 230.3M tokens across eight domains, including sequence computation, continuation, classification, and variant-effect prediction.
  • RL provides 98,433 rows covering 80,406 distinct prompts, with code-checked rewards for DNA computation, variant prediction, biology questions, classification, and instruction following.
  • FineDNA is Apache 2.0 licensed, but third-party source datasets retain their own licenses and terms; mid-training configurations total about 950 GB and should be streamed.

#dna language models #genomics #reinforcement learning #supervised fine-tuning

Generated by gpt-5.6-sol. Check the original for details.

Metric trends

Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained

Likes

Total likes reported by the source.

  • Hugging Face Datasets: 68, latest observation ; +59 over observed period since

Downloads (30d)

Rolling 30-day downloads. A change in this value is not new downloads during the observed period.

  • Hugging Face Datasets: 12,393, latest observation ; +12,321 over observed period since

Aliases

SourceExternal IDURL
Hugging Face DatasetsHuggingFaceBio/FineDNAhttps://huggingface.co/datasets/HuggingFaceBio/FineDNA Read discussion