Blame for README.md

---
license: apache-2.0
language:
- en
task_categories:
- text-generation
tags:
- biology
- genomics
- dna
- midtraining
- sft
- rl
pretty_name: FineDNA
configs:
- config_name: mid_dna_eukaryote_fungi
  data_files:
  - split: train
    path: mid/dna_eukaryote_fungi/train-*.parquet
- config_name: mid_dna_eukaryote_protozoa
  data_files:
  - split: train
    path: mid/dna_eukaryote_protozoa/train-*.parquet
- config_name: mid_dna_eukaryote_plant
  data_files:
  - split: train
    path: mid/dna_eukaryote_plant/train-*.parquet
- config_name: mid_dna_eukaryote_invertebrate
  data_files:
  - split: train
    path: mid/dna_eukaryote_invertebrate/train-*.parquet
- config_name: mid_dna_eukaryote_vertebrate_mammalian
  data_files:
  - split: train
    path: mid/dna_eukaryote_vertebrate_mammalian/train-*.parquet
- config_name: mid_dna_eukaryote_vertebrate_other
  data_files:
  - split: train
    path: mid/dna_eukaryote_vertebrate_other/train-*.parquet
- config_name: mid_dna_mrna
  data_files:
  - split: train
    path: mid/dna_mrna/train-*.parquet
- config_name: mid_dna_mrna_splice
  data_files:
  - split: train
    path: mid/dna_mrna_splice/train-*.parquet
- config_name: mid_dna_prokaryote
  data_files:
  - split: train
    path: mid/dna_prokaryote/train-*.parquet
- config_name: mid_bio_p90_biorxiv
  data_files:
  - split: train
    path: mid/bio_p90_biorxiv/train-*.parquet
  - split: validation
    path: mid/bio_p90_biorxiv/validation-*.parquet
- config_name: mid_bio_p90_biostars
  data_files:
  - split: train
    path: mid/bio_p90_biostars/train-*.parquet
  - split: validation
    path: mid/bio_p90_biostars/validation-*.parquet
- config_name: mid_bio_p90_finepdfs
  data_files:
  - split: train
    path: mid/bio_p90_finepdfs/train-*.parquet
  - split: validation
    path: mid/bio_p90_finepdfs/validation-*.parquet
- config_name: mid_bio_p90_fineweb
  data_files:
  - split: train
    path: mid/bio_p90_fineweb/train-*.parquet
  - split: validation
    path: mid/bio_p90_fineweb/validation-*.parquet
- config_name: mid_bio_p90_pubmed
  data_files:
  - split: train
    path: mid/bio_p90_pubmed/train-*.parquet
  - split: validation
    path: mid/bio_p90_pubmed/validation-*.parquet
- config_name: mid_bio_rest
  data_files:
  - split: train
    path: mid/bio_rest/train-*.parquet
  - split: validation
    path: mid/bio_rest/validation-*.parquet
- config_name: mid_bio_gemini_fim
  data_files:
  - split: train
    path: mid/bio_gemini_fim/train-*.parquet
- config_name: mid_bio_gemini_nonfim
  data_files:
  - split: train
    path: mid/bio_gemini_nonfim/train-*.parquet
  - split: validation
    path: mid/bio_gemini_nonfim/validation-*.parquet
- config_name: mid_english_fineweb_edu
  data_files:
  - split: train
    path: mid/english_fineweb_edu/train-*.parquet
- config_name: mid_english_dclm
  data_files:
  - split: train
    path: mid/english_dclm/train-*.parquet
- config_name: mid_english_finepdfs_edu
  data_files:
  - split: train
    path: mid/english_finepdfs_edu/train-*.parquet
- config_name: mid_instruct_nemotron_chat_v3
  data_files:
  - split: train
    path: mid/instruct_nemotron_chat_v3/train-*.parquet
- config_name: mid_instruct_nemotron_if_v3
  data_files:
  - split: train
    path: mid/instruct_nemotron_if_v3/train-*.parquet
- config_name: mid_instruct_nemotron_if_v2
  data_files:
  - split: train
    path: mid/instruct_nemotron_if_v2/train-*.parquet
- config_name: sft_mix
  data_files:
  - split: train
    path: sft/mix/train-*.parquet
  - split: test
    path: sft/mix/test-*.parquet
- config_name: sft_sequence_continuation
  data_files:
  - split: train
    path: sft/sequence_continuation/train-*.parquet
- config_name: sft_variant_effect_prediction
  data_files:
  - split: train
    path: sft/variant_effect_prediction/train-*.parquet
- config_name: sft_dna_computation
  data_files:
  - split: train
    path: sft/dna_computation/train-*.parquet
- config_name: sft_sequence_classification
  data_files:
  - split: train
    path: sft/sequence_classification/train-*.parquet
- config_name: sft_chat
  data_files:
  - split: train
    path: sft/chat/train-*.parquet
- config_name: sft_sequence_analysis
  data_files:
  - split: train
    path: sft/sequence_analysis/train-*.parquet
- config_name: sft_instruction_following
  data_files:
  - split: train
    path: sft/instruction_following/train-*.parquet
- config_name: sft_biology_qa
  data_files:
  - split: train
    path: sft/biology_qa/train-*.parquet
- config_name: rl_mix
  data_files:
  - split: train
    path: rl/mix/train-*.parquet
---

![FineDNA](https://huggingface.co/datasets/HuggingFaceBio/FineDNA/resolve/main/assets/finedna-banner.png)

# FineDNA

FineDNA is the training data of [Carbon-X](https://huggingface.co/HuggingFaceBio/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 | rows | starts from |
|---|---|---:|---|
| [Mid-training](#mid-training) | `mid_*`, one per source (23) | 514.30B tokens, of which the run read 302B | `Qwen/Qwen3-14B` |
| [SFT](#sft) | `sft_mix`, and `sft_*` for each of its 8 domains | 203,923 `train`, 100 `test` | `HuggingFaceBio/Carbon-X-Base` |
| [RL](#rl) | `rl_mix` | 98,433 `train` | `HuggingFaceBio/Carbon-X-SFT` |

`sft_mix` and `rl_mix` are the rows the SFT and GRPO trainers used, in the order they used them. Mid-training has no
single mix: Megatron samples the `mid_*` configs with the weights in [the mix](#the-mix) below.

```python
from datasets import load_dataset

# one mid-training source; the mid-training configs total about 950 GB, so stream them
ds = load_dataset("HuggingFaceBio/FineDNA", "mid_dna_mrna", split="train", streaming=True)
# the SFT mixture, or one of its domains
ds = load_dataset("HuggingFaceBio/FineDNA", "sft_mix")
ds = load_dataset("HuggingFaceBio/FineDNA", "sft_chat", split="train")
# the RL blend
ds = load_dataset("HuggingFaceBio/FineDNA", "rl_mix", split="train")
```

The training recipes for every stage are in [`huggingface/carbon`](https://github.com/huggingface/carbon/tree/main/carbon-x).

## Mid-training

The `mid_*` configs hold every source of the 302B-token mid-training run, one config per source: genomes and
transcripts, biology text that contains DNA, English web and PDF text, and instruction data. Their files are under
`mid/<source>/`.

Every config holds all the data Carbon-X sampled from for that source. That is not the whole upstream dataset: the
DNA, for example, is GENERator and OpenGenome2 after the filtering described below. The **share of training** column
below is the weight each config gets, and those weights are the recipe: they decide how much of each config the run
read, so some configs were read several times and some only in part. Tokenize the `text` column with the Carbon-X
tokenizer, train with these weights, and you get the same data diet as Carbon-X.

### Columns

Every mid-training config has the same columns:

| column | |
|---|---|
| `text` | the document, as it was tokenized |
| `id` | the source's document id, where it has one |
| `source` | the config name without `mid_`, or for biology text the original source (`pubmed`, `biorxiv`, …) |
| `metadata` | the source's other fields, as a JSON string |

### The mix

| | unique tokens | share of training in Carbon-X |
|---|---:|---:|
| DNA | 181.96B | 55% |
| Biology text | 15.75B | 15% |
| English | 310.22B | 20% |
| Instruction | 6.36B | 10% |
| **Total** | **514.30B** | **100%** |

Carbon-X trained on 302B of these 514.30B tokens. Long-DNA biology text was read about 7 times and
instruction data about 5; of English, prokaryote and mRNA-splice the run read only a part.

| config | documents | tokens | share of training (weight) | tokens read by Carbon-X | passes |
|---|---:|---:|---:|---:|---:|
| `mid_dna_eukaryote_fungi` | 6,607,490 | 1.88B | 1.12% | 3.37B | 1.79 |
| `mid_dna_eukaryote_protozoa` | 1,257,499 | 366.8M | 0.19% | 581.3M | 1.58 |
| `mid_dna_eukaryote_plant` | 9,429,321 | 5.55B | 3.08% | 9.30B | 1.68 |
| `mid_dna_eukaryote_invertebrate` | 8,716,932 | 13.71B | 7.58% | 22.90B | 1.67 |
| `mid_dna_eukaryote_vertebrate_mammalian` | 7,169,696 | 18.96B | 10.32% | 31.16B | 1.64 |
| `mid_dna_eukaryote_vertebrate_other` | 12,420,543 | 29.56B | 16.21% | 48.95B | 1.66 |
| `mid_dna_mrna` | 52,702,454 | 19.41B | 8.80% | 26.58B | 1.37 |
| `mid_dna_mrna_splice` | 56,877,102 | 32.93B | 2.20% | 6.64B | 0.20 |
| `mid_dna_prokaryote` | 17,408,059 | 59.60B | 5.50% | 16.61B | 0.28 |
| `mid_bio_p90_biorxiv` | 7,262 | 99.1M | 0.24% | 710.3M | 7.17 |
| `mid_bio_p90_biostars` | 2,036 | 3.1M | 0.01% | 25.4M | 8.19 |
| `mid_bio_p90_finepdfs` | 51,241 | 1.96B | 4.58% | 13.84B | 7.07 |
| `mid_bio_p90_fineweb` | 8,205 | 79.7M | 0.19% | 560.2M | 7.03 |
| `mid_bio_p90_pubmed` | 68,330 | 854.2M | 1.99% | 6.01B | 7.03 |
| `mid_bio_rest` | 914,391 | 12.16B | 7.00% | 21.14B | 1.74 |
| `mid_bio_gemini_fim` | 56,866 | 262.8M | 0.50% | 1.51B | 5.74 |
| `mid_bio_gemini_nonfim` | 72,343 | 333.4M | 0.50% | 1.51B | 4.53 |
| `mid_english_fineweb_edu` | 96,541,686 | 99.35B | 8.00% | 24.16B | 0.24 |
| `mid_english_dclm` | 88,340,007 | 111.98B | 8.00% | 24.16B | 0.22 |
| `mid_english_finepdfs_edu` | 17,596,723 | 98.89B | 4.00% | 12.08B | 0.12 |
| `mid_instruct_nemotron_chat_v3` | 562,376 | 2.25B | 3.53% | 10.67B | 4.75 |
| `mid_instruct_nemotron_if_v3` | 249,748 | 632.0M | 0.99% | 3.00B | 4.75 |
| `mid_instruct_nemotron_if_v2` | 1,997,510 | 3.48B | 5.47% | 16.53B | 4.75 |

*Tokens* is the config's size with the Carbon-X tokenizer (estimated from a 2,000-document sample per config).
*Tokens read by Carbon-X* is share x 302B. *Passes* is how many times over the run read the config: below 1, it read
only that fraction of it.

The full training setup we used for Carbon-X are in the
[Carbon-X midtraining recipe](https://github.com/huggingface/carbon/tree/main/carbon-x/training/midtraining).

### What is in it

**DNA.** Eukaryote genomic windows from [GENERator](https://huggingface.co/datasets/GenerTeam/pretrain_data_eukaryote),
up to 100 kbp, in six species groups. mRNA, mRNA with splice and promoter context, and prokaryote genomes (GTDB and
IMG/PR) from [OpenGenome2](https://huggingface.co/datasets/arcinstitute/opengenome2). The raw sequences are also in
[`HuggingFaceBio/carbon-pretraining-corpus`](https://huggingface.co/datasets/HuggingFaceBio/carbon-pretraining-corpus).
Here they are formatted exactly as they were trained on:

- every sequence is cut to a multiple of 6 bp (the tokenizer reads 6-mers) and wrapped in `<dna>…</dna>`;
- eukaryote sequences carry metadata tags half the time, chosen at random: species group and gene type, species group
  only, or gene type only, e.g. `<mammalian_species><protein_coding_region><dna>ATG…</dna>`;
- mRNA and prokaryote sequences are uppercased, splice-junction markers (`@`) are removed, and mRNA longer than
  100,000 bases is dropped.

**Biology text.** Web pages, papers and forum posts that contain DNA, plus synthetic annotations. The web and paper
configs were built in two steps; the code is in
[`mid/pipeline/`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA/tree/main/mid/pipeline).

1. **Find documents that contain DNA:**
   [`1_regex_filter.py`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA/blob/main/mid/pipeline/1_regex_filter.py).
   We ran a regex filter over FineWeb-Edu, FinePDFs, PubMed Central, bioRxiv and Biostars. A document is kept when
   - it has a run of 15 or more nucleotide letters (A, T, C, G, U, N) using at least 3 different ones,
   - that run is not inside curly braces, which in scraped text is usually LaTeX, and
   - it mentions a molecular-biology keyword (dna, gene, primer, pcr, ...).
2. **Split by DNA length:**
   [`2_p90_split.py`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA/blob/main/mid/pipeline/2_p90_split.py),
   which finds DNA with
   [`tag_dna_sequences.py`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA/blob/main/mid/pipeline/tag_dna_sequences.py).
   Documents whose longest DNA sequence is at least 39 bp, the 90th percentile, form `mid_bio_p90_*`; everything else
   forms `mid_bio_rest`. Short runs are mostly primers and motifs, while the p90 documents carry real sequences. In our
   ablations, upweighting them improved DNA benchmarks.

| config | what it is |
|---|---|
| `mid_bio_p90_biorxiv`, `mid_bio_p90_biostars`, `mid_bio_p90_finepdfs`, `mid_bio_p90_fineweb`, `mid_bio_p90_pubmed` | documents with long DNA sequences, one config per source |
| `mid_bio_rest` | the other regex hits from the same five sources, whose longest DNA sequence is under 39 bp |
| `mid_bio_gemini_nonfim` | genomic regions described by Gemini next to their sequence |
| `mid_bio_gemini_fim` | the same kind of documents in Qwen's fill-in-the-middle format: part of each document is moved to the end, and the model learns to fill it in |

**English.** [FineWeb-Edu](https://huggingface.co/datasets/HuggingFaceFW/fineweb-edu) (100BT sample),
[DCLM](https://huggingface.co/datasets/HuggingFaceFW/dclm_100BT-shuffled) and
[FinePDFs-Edu](https://huggingface.co/datasets/HuggingFaceFW/finepdfs_edu_100BT-shuffled), to keep the model's English.

**Instruction.** [Nemotron Instruction-Following-Chat v3](https://huggingface.co/datasets/nvidia/Nemotron-SFT-Instruction-Following-Chat-v3)
(chat and instruction-following splits) and
[v2](https://huggingface.co/datasets/nvidia/Nemotron-SFT-Instruction-Following-Chat-v2) (chat), rendered with the
Qwen3 chat template, reasoning traces kept. Mid-training starts from an instruct model, and this keeps it able to
follow instructions.

Biology configs also have a `validation` split: the documents held out of training to track loss.

## SFT

Supervised fine-tuning conversations for a DNA language model: 204,023 rows (203,923 `train`, 100 `test`).
`sft_mix` holds exactly the rows the SFT trainer built from the Carbon-X SFT mixture, so training on its `train` split
reproduces the mixture without any further weighting.

Each domain also has its own config, `sft_<domain>`, holding the `sft_mix` `train` rows of that domain in the same
order, with no test split, e.g. `load_dataset("HuggingFaceBio/FineDNA", "sft_chat")`.

### Columns

- `messages`: the conversation, as a list of `{role, content}` turns.
- `chat_template_kwargs`: `{"tokenizer_kwargs": {"auto_dna_tags": <bool>}}`, passed per example to
  `apply_chat_template` so the DNA tokenizer wraps sequences in `<dna>...</dna>`. It is `false` only for the sequence
  rewriting rows, whose answers are plain text. Keep it when selecting training columns.
- `domain`: the kind of task, one of the eight below.

### Domains

| config | train | test (in `sft_mix`) | what the rows teach |
| --- | ---: | ---: | --- |
| `sft_sequence_continuation` | 58,727 | 27 | predict the next 1–32 nucleotides after a 4–8 kb eukaryotic context |
| `sft_variant_effect_prediction` | 57,254 | 24 | decide whether a single-nucleotide variant is benign or pathogenic from its sequence window, with templated reasoning |
| `sft_dna_computation` | 44,925 | 25 | answers computed exactly from the sequence: counting, GC content, melting temperature, translation, restriction sites, base lookup, coordinates, edits, rewriting |
| `sft_sequence_classification` | 16,396 | 11 | multiple-choice functional genomics: promoter vs enhancer, species of origin, TF motifs, TF binding sites |
| `sft_chat` | 8,553 | 6 | multi-turn conversations on molecular-biology concepts, the assistant's identity, honesty about its limits, and pushing back on false claims |
| `sft_sequence_analysis` | 7,561 | 3 | free-text, mostly multi-turn discussion of a pasted sequence: primer checks, annotation, coding potential, troubleshooting |
| `sft_instruction_following` | 5,904 | 1 | answering in the format the user asks for, and following format constraints on biology prompts |
| `sft_biology_qa` | 4,603 | 3 | genetics, molecular biology and bioinformatics questions from Biology StackExchange, Biostars, real user chats and Nemotron chat data |

## RL

Verifiable prompts for reinforcement learning of a DNA language model: 98,433 rows in one `train` split, covering
80,406 distinct prompts. Every prompt has a gold answer or constraint set that a reward function checks in code. The
blend was calibrated on rollouts from Carbon-X-SFT and excludes the prompts that model saw during SFT. It is meant for
one epoch. The rows of `rl_mix` are in the order the GRPO trainer's mixture loader produces (seed 0), so training on it
reproduces the blend exactly.

Some prompts appear more than once on purpose, for groups that learn slowly under RL: `basic_dna` repeats half of each
task type (10,153 second copies), `bioreason` repeats 30% of each subgroup and label (4,614), and `dna_skills` repeats
every prompt (3,260).

### Columns

- `prompt`: the conversation to complete, as a list of `{role, content}` turns (2,983 rows open with a system turn).
- `answer_text`: the gold answer that most rewards compare against.
- `task_group`: which reward scores the row. Each task reward returns `None` for rows of other groups.
- `task_type`: the sub-task, used by the `basic_dna` and `seqrecqa_position` rewards (null elsewhere).
- `verifier_metadata`: the constraints an instruction-following reply must meet (`if` rows only).
- `gold_answer_json`: the exact answer the DNA-skills verifier checks (`dna_skills` rows only).
- `domain`: the kind of task. It uses the same labels as the SFT domains. Five of the eight occur here because the
  blend has no chat, sequence-analysis or sequence-continuation prompts.

### Task groups

| domain | task group | rows | what the prompts ask |
| --- | --- | ---: | --- |
| `dna_computation` | `basic_dna` | 30,461 | short-sequence computations with code-checked answers: base counts, melting temperature, restriction fragments, CRISPR cut sites, translation, motif spacing |
| `dna_computation` | `dna_skills` | 6,520 | exact operations on longer sequences: locate a base or snippet, convert coordinates, compare, edit, translate, design |
| `dna_computation` | `seqrecqa_position` | 6,000 | localisation: the position of a base, 6-base block, variant block, 12-base span, single-base difference or snippet |
| `variant_effect_prediction` | `bioreason` | 19,996 | is a coding-sequence or non-SNV variant pathogenic or benign |
| `variant_effect_prediction` | `vepqa` | 7,426 | is a single-nucleotide variant pathogenic or benign, judged from its sequence window |
| `sequence_classification` | `genomeqa` | 14,032 | multiple-choice functional genomics: promoter vs enhancer, species of origin, TF motifs and binding sites, splice sites, histone marks |
| `biology_qa` | `bio_mcq` | 9,998 | four- and ten-option biology knowledge questions |
| `instruction_following` | `if` | 4,000 | prompts with verifiable formatting constraints |

In total, `dna_computation` has 42,981 rows, `variant_effect_prediction` 27,422, `sequence_classification` 14,032,
`biology_qa` 9,998 and `instruction_following` 4,000.

## License

FineDNA is released under the Apache 2.0 license. The mid-training configs are built from third-party sources
(GENERator, OpenGenome2, FineWeb-Edu, DCLM, FinePDFs-Edu, PubMed Central, bioRxiv, Biostars and Nemotron), which keep
their own licenses and terms. Check the source dataset before use.

## Citation

FineDNA was built for Carbon-X, a joint collaboration between the research teams at Hugging Face, Zhongguancun
Academy, and the Telethon Institute of Genetics and Medicine/University of Naples “Federico II”. If you find it
useful, please cite it as:

```bibtex
@misc{benallal2026finedna,
  title        = {FineDNA},
  author       = {Ben Allal, Loubna and Tunstall, Lewis and Aubakirova, Dana and Fiusco, Maurizio and Beeching, Edward and Rasul, Kashif and von Werra, Leandro and Li, Qiuyi and di Bernardo, Diego},
  year         = {2026},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/HuggingFaceBio/FineDNA}}
}
```