Add the FineDNA dataset card with mid, SFT and RL configs

README.md

@@ -0,0 +1,483 @@
+---
+license: other
+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.
+
+### The mix
+
+| | unique tokens | share of training |
+|---|---:|---:|
+| 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.
+
+### How to train on it
+
+1. Tokenize `text` with [`HuggingFaceBio/carbon-x`](https://huggingface.co/HuggingFaceBio/carbon-x). It detects DNA in
+   text and encodes it with its 6-mer DNA vocabulary. DNA configs already wrap sequences in `<dna>…</dna>`. To
+   tokenize a source straight from the Hub, read `hf://datasets/HuggingFaceBio/FineDNA/mid/<source>`.
+2. Sample configs by the *share of training* column. In Megatron, that is the weight of each `--data-path` prefix.
+   Within a config, split its weight across shards in proportion to their size, so every document is equally likely.
+3. Train for 302B tokens: 72,000 iterations at a global batch of 512 x 8,192 tokens, from the frozen-DNA-embedding
+   Qwen3-14B checkpoint (a short first stage that trains only the new DNA embeddings). The full recipe, including the vocabulary extension and that first stage, is in the
+   [Carbon-X midtraining recipe](https://github.com/huggingface/carbon/tree/main/carbon-x/training/midtraining).
+   Training longer with the same weights repeats the same data more often; for more unique DNA or English, add data
+   from the original datasets linked below.
+
+### What is in it
+
+**DNA (55%).** Eukaryote genomic windows from [GENERator](https://huggingface.co/datasets/GenerTeam/pretrain_data_eukaryote),
+up to 100 kbp, in six species groups weighted by the corpus's natural distribution. mRNA, mRNA with splice and promoter
+context, and prokaryote genomes (GTDB and IMG/PR) from [OpenGenome2](https://huggingface.co/datasets/arcinstitute/opengenome2).
+All DNA configs are complete: they hold every sequence the run could sample from. Carbon-X read eukaryote and mRNA
+more than once, and about 28% of prokaryote and
+20% of mRNA-splice.
+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 (15%).** 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; the
+   p90 documents carry real sequences. In our ablations, training on them instead of every regex hit improved DNA
+   benchmarks, so in the mix they get as much weight as `mid_bio_rest` and are read about seven times.
+
+| 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 (20%).** [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 (10%).** [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.
+
+### How the mix matches the run
+
+Every config was rebuilt from its source and checked against the document counts in the training run's log: each
+holds everything the run sampled from for that source. How much of each the run read is set by the weights, as in the table above.
+Where the run read only part of a config, which documents Megatron happened to draw cannot be recovered, but training
+with the same weights draws the same amount from the same pool.
+
+Two pieces of data are included although they never reached the run, so that the pools are complete:
+1,720,988 documents of `mid_dna_eukaryote_invertebrate` and
+1,726,948 of `mid_english_dclm`, from tokenization shards that were missing when Carbon-X
+trained.
+
+Biology configs also have a `validation` split: the documents held out of training to track loss.
+
+### 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 `mid/` files are copied unchanged from [`HuggingFaceBio/FineDNA-Mid`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA-Mid)
+at revision `663538d6`.
+
+## SFT
+
+Supervised fine-tuning conversations for a DNA language model: 204,023 rows (203,923 `train`, 100 `test`) drawn
+from 20 source datasets with the Carbon-X SFT mixture weights. `sft_mix` holds exactly the rows the SFT trainer builds
+from that mixture (seed 0), 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 |
+
+### Sources
+
+Every source is the `_decontaminated_ngram_dna_homology` variant of the config below, `train` split. Each is shuffled
+with seed 0 and truncated to `int(rows × weight)`; the result is concatenated, shuffled with seed 0 and split into
+`train` and a 100-row `test` set with seed 0.
+
+| dataset | config | revision | rows | weight | rows used | domain |
+| --- | --- | --- | ---: | ---: | ---: | --- |
+| `HuggingFaceBio/sequence-recovery-qa` | `varlen` | `e09cae72` | 58,754 | 1.0 | 58,754 | `sequence_continuation` |
+| `HuggingFaceBio/sequence-recovery-qa` | `rewrite` | `e09cae72` | 24,730 | 0.1266728 | 3,132 | `dna_computation` |
+| `HuggingFaceBio/carbon-format-if` | `format_if_v2` | `390592d0` | 5,196 | 1.0 | 5,196 | `instruction_following` |
+| `HuggingFaceBio/basic-dna-sft-data-v3` | `all_diversify_30pct` | `72dc7e73` | 63,848 | 0.151 | 9,641 | `dna_computation` |
+| `HuggingFaceBio/carbon-sequence-contextfree-v8-regex-10k` | `final` | `5f17e938` | 10,000 | 0.3 | 3,000 | `sequence_analysis` |
+| `HuggingFaceBio/carbon-gemini-repair-sft` | `default` | `63cccfc7` | 11,907 | 0.3 | 3,572 | `sequence_analysis` |
+| `HuggingFaceBio/carbon-sequence-contextfree-v6-1000` | `final` | `e2792417` | 992 | 1.0 | 992 | `sequence_analysis` |
+| `HuggingFaceBio/carbon-bioSE-sft` | `final` | `31dc6d39` | 2,072 | 1.0 | 2,072 | `biology_qa` |
+| `HuggingFaceBio/carbon-biostars-sft-v2` | `final` | `5014a6b6` | 1,327 | 1.0 | 1,327 | `biology_qa` |
+| `HuggingFaceBio/wild-bio-prompts-luna-generated-verified-v1` | `final` | `35440c49` | 842 | 1.0 | 842 | `biology_qa` |
+| `HuggingFaceBio/seqqa2-luna-sft-structurednotags-dna8-clean-v3` | `final` | `38644f1f` | 853 | 1.0 | 853 | `dna_computation` |
+| `HuggingFaceBio/genomeqa-sft` | `enhancer_promoter_diversify_10pct` | `398703a5` | 10,559 | 0.4 | 4,223 | `sequence_classification` |
+| `HuggingFaceBio/genomeqa-sft` | `eukaryote_cls_diversify_10pct` | `398703a5` | 9,143 | 0.4 | 3,657 | `sequence_classification` |
+| `HuggingFaceBio/genomeqa-sft` | `tf_motif_diversify_10pct` | `398703a5` | 13,058 | 0.4 | 5,223 | `sequence_classification` |
+| `HuggingFaceBio/genomeqa-sft` | `tfbs_diversify_10pct` | `398703a5` | 8,262 | 0.4 | 3,304 | `sequence_classification` |
+| `HuggingFaceBio/vepqa-sft-draft-loubna` | `templated_reasoning_multiview_v2_all` | `76e39513` | 74,484 | 0.769 | 57,278 | `variant_effect_prediction` |
+| `HuggingFaceBio/nemotron-data-filtered-carbon` | `biology_en_max8704` | `50df980c` | 3,656 | 0.1 | 365 | `biology_qa` |
+| `HuggingFaceBio/nemotron-data-filtered-carbon` | `instruction_following_bio_en_max8704` | `50df980c` | 709 | 1.0 | 709 | `instruction_following` |
+| `HuggingFaceBio/easy-dna-chat-v2` | `all` | `53923d9a` | 11,502 | 0.7442 | 8,559 | `chat` |
+| `HuggingFaceBio/carbon-dna-skills` | `all` | `d703e0cd` | 39,156 | 0.8 | 31,324 | `dna_computation` |
+
+`carbon-format-if` rows reuse examples from basic-dna, seqqa2 and sequence-recovery `varlen` with a format instruction
+added, so they are labelled by purpose (`instruction_following`) rather than content.
+
+### Reproducing
+
+The `sft/` files are copied unchanged from [`HuggingFaceBio/FineDNA-SFT`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA-SFT)
+at revision `400d761b`, which was built with:
+
+```bash
+python sft/scripts/create_dataset.py --output_dataset HuggingFaceBio/FineDNA-SFT
+```
+
+[`sft/scripts/create_dataset.py`](sft/scripts/create_dataset.py) pins the full revision of every source above and needs
+read access to them (all are private). Its output matches the trainer's mixture loader row for row on both splits.
+
+The domain configs are then split from the mixture's `train` split at revision `60b6e0d5`:
+
+```bash
+python sft/scripts/create_subset.py --dataset HuggingFaceBio/FineDNA-SFT --revision 60b6e0d5c9af34bd97de521cc2cf5fb3e977771c
+```
+
+## 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.
+
+### Source
+
+| dataset | config | revision | rows |
+| --- | --- | --- | ---: |
+| `HuggingFaceBio/Rollouts-Carbon-14B-SFT-v36.52-step-000019119` | `blend_all_v0` | `205d8b22` | 98,433 |
+
+Only the six columns the trainer reads are kept, and `domain` is added from `task_group`. The rows are shuffled twice
+with seed 0, matching the trainer's mixture loader (once per source, once after concatenation).
+
+### Reproducing
+
+The `rl/` files are copied unchanged from [`HuggingFaceBio/FineDNA-RL`](https://huggingface.co/datasets/HuggingFaceBio/FineDNA-RL)
+at revision `2315ddc7`, which was built with:
+
+```bash
+python rl/scripts/create_dataset.py --output_dataset HuggingFaceBio/FineDNA-RL
+```
+
+[`rl/scripts/create_dataset.py`](rl/scripts/create_dataset.py) pins the full source revision and needs read access to
+it (the source is private). Its output matches the trainer's mixture loader row for row.
+
+## License
+
+Each mid-training config keeps the license and terms of its source: GENERator, OpenGenome2, FineWeb-Edu, DCLM,
+FinePDFs-Edu, PubMed Central, bioRxiv, Biostars, and Nemotron. The SFT and RL rows are built from the sources listed in
+their sections. Check the source dataset before use.