BatatasLM-SP5-28L_pretrain

Model Description

BatatasLM pretrained DNA foundation model: BatatasLM-SP5-28L_pretrain.

Model Architecture

The configuration declares model_type caduceus and CaduceusForMaskedLM. Caduceus, Mamba, RCPS, architectures, auto_map, Python module names, and class names are technical identifiers retained for loading.

Training Data

This model uses BatatasLM-SP5-Pretraining-Data, five sweetpotato genomes or accessions. Sequence length is 4096 bp.

Intended Use

Research involving DNA sequence representation and masked-language-modeling workflows.

Model Files

Configuration, pytorch_model.bin, tokenizer files, and custom Python modules referenced by auto_map.

Loading the Model

from transformers import AutoModelForMaskedLM
model = AutoModelForMaskedLM.from_pretrained("BatatasLM-SP5-28L_pretrain", trust_remote_code=True)

Limitations

This card does not assert publication status, benchmark metrics, or a performance conclusion.

Citation

BatatasLM: A Sweetpotato-Centric DNA Foundation Model.

Project Information

• Project: BatatasLM
• GitHub: https://github.com/Weiyiliang/BatatasLM
• Institution: Jiangsu Normal University

Contact

Yiliang Wei - weiyiliang.2013@tsinghua.org.cn