HNet DNA 1mer Model (Config: T1m3 → T15 → m4)

This repository contains a pretrained Hierarchical Neural Network (HNet) trained on the hg38 human genome using 1‑mer tokenization.
The model is designed for DNA sequence modeling and compression research.

---

Model Architecture

High-level layout

• Architecture Layout: ["T1m3", ["T15"], "m4"]
  • Stage 1: Transformer + Mamba (3 layers)
  • Stage 2: Transformer (15 layers)
  • Stage 3: Mamba (4 layers)

Vocabulary

• Vocab size: 6 (A, C, G, T, N, padding)

---

Intended Uses & Limitations

Intended Uses

• Genomic sequence modeling
• DNA sequence analysis
• Research on sequence compression and representation learning

Limitations

• Not intended for clinical or medical diagnostic use.
• Trained only on the hg38 human genome. Performance on other genomes may vary.

---