README.md

---
title: DNA Benchmark
emoji: 🥇
colorFrom: red
colorTo: indigo
sdk: gradio
app_file: app.py
pinned: true
license: apache-2.0
short_description: DNA Foundational Models leaderboard
sdk_version: 5.19.0
---

# DNA Benchmark Leaderboard

Interactive leaderboard comparing DNA foundation models across genomics classification tasks.
Data is loaded from [`lokahq/genomic-benchmark-metrics`](https://huggingface.co/datasets/lokahq/genomic-benchmark-metrics) on startup.

## Development

```bash
# Install uv (if not already installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies
uv sync

# Run locally
make app

# Format code
make style
```

## Context length bubble chart

The bubble chart requires mean sequence lengths per task to be pre-computed once:

```bash
make seq-lengths        # computes lengths and writes tools/task_seq_lengths.json
git add tools/task_seq_lengths.json && git commit -m "populate task seq lengths"
```

The chart is a no-op until `tools/task_seq_lengths.json` is populated and committed.

## Configuration

| File | Purpose |
|---|---|
| `src/constants.py` | Task categories, metric names, seq length lookup |
| `src/data.py` | HF Hub loading, deduplication, filtering |
| `src/plots.py` | All Plotly figure factories |
| `tools/task_seq_lengths.json` | Mean sequence lengths per task (committed) |
| `.env` | `HF_TOKEN` and `LEADERBOARD_DATASET` (use HF Spaces secrets in deployment) |