Nucleotide Transformer v3
Collection
21 items • Updated • 4
# Use a pipeline as a high-level helper
from transformers import pipeline
pipe = pipeline("fill-mask", model="InstaDeepAI/ntv3_base_model")# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("InstaDeepAI/ntv3_base_model", dtype="auto")Quick Links
InstaDeepAI/ntv3_base_model
Unified base model repository for NTv3 models.
This repository contains shared modeling code used by both:
- Pre-trained models (masked language models)
- Post-trained models (conditioned multi-species models with functional genomics heads)
Note: This repo should not be used standalone. It provides modeling code that is referenced by individual model checkpoints via trust_remote_code=True.
Contents
| File | Purpose |
|---|---|
configuration_ntv3_pretrained.py |
Config class: Ntv3PreTrainedConfig |
configuration_ntv3_posttrained.py |
Config classes: DiscreteConditionedNTv3Config, NTv3PostTrainedConfig |
modeling_ntv3_pretrained.py |
Pre-trained model: NTv3PreTrained |
modeling_ntv3_posttrained.py |
Post-trained model: NTv3PostTrained with conditioned towers and heads |
tokenization_ntv3.py |
Tokenizer: NTv3Tokenizer (DNA) |
Architecture
- U-Net style conv tower → Transformer stack → deconv tower → LM head
- Post-trained models add adaptive layer norms and multi-species prediction heads
- Tokenizer: character-level over A T C G N + specials (
<unk><pad><mask><cls><eos><bos>)
# Gated model: Login with a HF token with gated access permission hf auth login