HuggingFaceBio/Carbon-X
AI model · https://huggingface.co/HuggingFaceBio/Carbon-X
First seen by Oriel on .
AI summary
Carbon-X is an open 14.8B-parameter model trained on DNA and natural-language text, achieving a 66.7% average across five DNA benchmarks with thinking disabled.
- A hybrid tokenizer combines Qwen3 BPE text tokens with 5,515 DNA tokens, encoding canonical DNA as non-overlapping 6-mers and preserving sequence tails.
- The model starts from Qwen3-14B and is trained on 302B tokens, followed by supervised fine-tuning and reinforcement learning with verifiable rewards on DNA tasks.
- Carbon-X leads the evaluated open models on coding-variant effects and next-base prediction, and its weights fit on a single 80 GB GPU in bfloat16.
- The authors estimate that analyzing 2.4 million ClinVar variants of uncertain significance would cost about $300; the custom tokenizer requires Transformers v5 or later or vLLM, and GGUF runtimes are unsupported.
#dna #language models #genomics #variant effects
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 Models: 283, latest observation ; +176 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 Models: 3,534, latest observation ; +3,341 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | HuggingFaceBio/Carbon-X | https://huggingface.co/HuggingFaceBio/Carbon-X Read discussion |