dnagpt/OmniGene-4-bio
gemma4-bio Code and per-example results for the paper: Scientific Data Composition as a Capability-Shaping Mechanism for Foundation Models: Evidence from Biological Continued Pretraining — Liang Wang (HUST). The study has two parts on a 26B-parameter Mixture-of-Experts model (Gemma-4-26B-A4B): Part I — training-free re-analysis of one checkpoint lineage (instruction-tuned base → biological CPT → SFT) across four capability axes. Part II — a controlled seven-model experiment:… See the full description on the dataset page: https://huggingface.co/datasets/dnagpt/OmniGene-4-bio.
Replace placeholder Gemma-4 citation with real arXiv report (2607.02770)
README: update title to match paper's new framing
Retitle + reframe paper: data composition as capability-shaping mechanism
README: add Part IV robustness + mechanism results
Add Part IV result artifacts
Update paper: add Part IV (robustness + mechanism)
README: add Part III small-model probe results
Add Part III results (GPT-2, 3 seeds x 2 arms x 8 tasks)
Update paper: add Part III (small-model matched-compute probe)
README: add Part II mixture experiment results
Add Part II results (7 models x 6 tasks) + figures
Update paper: add Part II (controlled mixture experiment)
Add OmniGene-4 self-citation; recompile paper
Phase-1 re-analysis: per-example results, summary tables, paper
initial commit