Add README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-4.0
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
- zh
|
| 6 |
+
size_categories:
|
| 7 |
+
- 100K<n<1M
|
| 8 |
+
task_categories:
|
| 9 |
+
- text-generation
|
| 10 |
+
- question-answering
|
| 11 |
+
tags:
|
| 12 |
+
- biology
|
| 13 |
+
- protein
|
| 14 |
+
- DNA
|
| 15 |
+
- bioinformatics
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# OmniGene-4 SFT corpus
|
| 19 |
+
|
| 20 |
+
Supervised fine-tuning data for the OmniGene-4 / OmniGene-4-MM family.
|
| 21 |
+
See https://github.com/maris205/omnigene4 for the training scripts that
|
| 22 |
+
consume these files.
|
| 23 |
+
|
| 24 |
+
## Files
|
| 25 |
+
|
| 26 |
+
| File | Rows | Used by | Description |
|
| 27 |
+
|---|---|---|---|
|
| 28 |
+
| `bio_sft_v2_train.jsonl` | ~179K | Bio-SFT v2 | Eight task families: protein homology (BioPAWS), DNA, structure (3Di/DSSP), cell biology, molecules, mutation, structure prediction, general bio QA |
|
| 29 |
+
| `distill_seed.jsonl` | ~6K | seed-only | Initial distillation seed used to bootstrap the v2 corpus |
|
| 30 |
+
| `train/omnigene_sft_v1_train.jsonl` | ~179K | SFT v3 base | Same as v2 with cleaner schema and Alpaca template |
|
| 31 |
+
| `train/omnigene_sft_v1_train_with_remote.jsonl` | **~199K** | **SFT v3-v5 (final)** | Above + 20K BioPAWS `protein_pair_remote` rows; this is the file used by Bio-SFT v3, v4, v5, and the OmniGene-4-MM Stage 2/3 LoRA training |
|
| 32 |
+
| `eval/omnigene_sft_v1_eval.jsonl` | ~1.5K | held-out eval | Used by `40-eval_omnigene4mm.py`, `60-eval_stage2.py`, `92-eval_stage3v3.py`, etc. |
|
| 33 |
+
| `master/omnigene_sft_v1_master.jsonl` | ~285K | (intermediate) | Pre-split master corpus before train/eval split |
|
| 34 |
+
| `master/cell_sft_master.jsonl` | ~37K | task subset | Cell-biology SFT subset |
|
| 35 |
+
| `master/mol_sft_master.jsonl` | ~50K | task subset | Molecule SFT subset |
|
| 36 |
+
| `stats/data_mix_report.json` | — | metadata | Per-category counts and ratios |
|
| 37 |
+
|
| 38 |
+
## Schema
|
| 39 |
+
|
| 40 |
+
Each line is a JSON object with at least `instruction`, `input`, `output`,
|
| 41 |
+
and `category`. Some rows additionally carry `task_name` and `subtask` for
|
| 42 |
+
fine-grained accounting.
|
| 43 |
+
|
| 44 |
+
## Citation
|
| 45 |
+
|
| 46 |
+
```bibtex
|
| 47 |
+
@article{wang2026omnigene4,
|
| 48 |
+
author = {Wang, Liang},
|
| 49 |
+
title = {{OmniGene-4}: A Unified Bio-Language MoE Model with Router-Level
|
| 50 |
+
Interpretability and Modality-Invariant Transfer},
|
| 51 |
+
year = {2026},
|
| 52 |
+
journal = {bioRxiv},
|
| 53 |
+
doi = {10.64898/2026.05.12.724542}
|
| 54 |
+
}
|
| 55 |
+
```
|