datasets
Training and evaluation data, with the modality, task and licence stated up front. Listed live from the Hugging Face Hub.
mgnify-evo2-l26-small-qualmgnify-evo2-l26-fullevo2-sae-handson
Evo 2 sparse autoencoder activations for the Escherichia coli K-12 MG1655 genome
SAE activations for every base of the chromosome (RefSeq GCF_000005845.2, NC_000913.3,
4,641,652 bp), with the embeddings, annotation and structures used by the
hands-on notebook.
How the activations were made
Evo 2 7B (arcinstitute/evo2_7b) was run over the circular chromosome in 16,384 bp
windows at a stride of 15,360 bp, discarding the first 1,024 bp of each window as context.
Each… See the full description on the dataset page: https://huggingface.co/datasets/suzuki-2001/evo2-sae-handson.evo2_datsetevo2-spaceflight-vep
Evo2 Zero-Shot VEP Scores for Spaceflight Radiation-Response Genes
Pre-computed zero-shot variant effect prediction scores from the Evo2 genomic foundation model (7B parameters) across 10 spaceflight radiation-response genes (215,001 scored variants).
Code: github.com/jang1563/evo2-spaceflight-vep
Dataset Description
Each row is a single variant (SNV or indel) scored by Evo2 using an 8,192 bp context window with reverse-complement averaging.
Columns… See the full description on the dataset page: https://huggingface.co/datasets/jang1563/evo2-spaceflight-vep.evo2_dataset
Evo2 Dataset
Dataset Description
Evo2 Dataset is an Evo2 mini genome dataset adapted for OneScience/evo2/. It contains FASTA, compressed FASTA, and merged FASTA files for human chr20, chr21, and chr22, as well as train, validation, and test .bin/.idx splits preprocessed with the Byte-Level tokenizer.
Supported Tasks
This dataset is not the complete OpenGenome2 dataset and is not intended to reproduce full-scale pretraining. It is intended for… See the full description on the dataset page: https://huggingface.co/datasets/OneScience-Group/evo2_dataset.evo2-clinvar
ClinVar Variant Effect Prediction Benchmark
Dataset Description
A curated subset of the NCBI ClinVar database (release: February 28, 2024). Each variant includes precomputed scores from Evo 2 and a set of baseline models used in the paper. Variants were filtered to retain only those with a ClinVar final review status of two gold stars or higher, ensuring higher-confidence clinical annotations supported by multiple submitters or expert panels.
Column… See the full description on the dataset page: https://huggingface.co/datasets/goodarzilab/evo2-clinvar.mgnify-evo2-amr-shortread-eval-v0
MGnify × Evo 2 short-read AMR detection eval (v0)
A small reproducibility benchmark for AMR detection on simulated short reads
using the Evo 2 DNA foundation model.
What's here
reads.jsonl — 100 records (50 AMR-positive, 50 matched-negative)
embeddings/<read_id>.npz — Evo 2 layer-26 activations per read
Source
All reads were simulated (MiSeq-like, 301 bp, paired-end, ART simulator) from
two CDSs in MGnify MAG MGYG000307615 (chicken-gut catalogue):
Positives:… See the full description on the dataset page: https://huggingface.co/datasets/JG1310/mgnify-evo2-amr-shortread-eval-v0.TDiG-evo2-hidden-states
TDiG Evo 2 Hidden States Dataset
Raw + RMSnormed hidden states from Evo 2 7B (arcinstitute/evo2_7b_base,
8K context, no FP8) forward pass on chr22, chr17 (human reference hg38),
and 10,910 ClinVar variants in 15 cancer genes.
Companion to the TDiG (Think Deep in Genome) project:
Code + analyses + per-token settling cells: https://github.com/YAICON-8th-Think-Deep-in-Genome/TDiG
Detailed metric guide: METRICS_GUIDE.md — formulas + storage locations + recipes for every cell… See the full description on the dataset page: https://huggingface.co/datasets/darejinn/TDiG-evo2-hidden-states.evo2_logoEvo2_Predictor_GAMEModule Version: Evo2_7b_Predictor_20260619-145824_EDT
GAME Schema Version: v 1.0
Github Link: https://github.com/de-Boer-Lab/GAME-Evo2-predictor
Additional information can be found on GitHub: Genomic API for Model Evaluation
Original publication can be found here: Brixi et al., 2026
H100 GPU required. Evo2 7B does not run on CPU. The Predictor requires at least one CUDA-capable GPU (one H100 is sufficient for the 7B model).
Evo2Evo2_7b_scoresEvo2-1Evo2evo2evo2-resultsmgnify-evo2-l26-amr-pilot