NTv3 — Foundation Models for Long-Range Genomics

This Space is the companion hub for NTv3 models: runnable notebooks for inference, fine-tuning, interpretation, and sequence generation.

Long-context genomics Torch notebooks Inference • Fine-tune • Interpret • Generate

Notebooks

Browse notebooks folder
00 — Quickstart inference
01 — Tracks prediction
02 — Genome annotation / segmentation
03 — Fine-tune a head
04 — Model interpretation
05 — Sequence generation

Models

Pretrained checkpoints:

InstaDeepAI/ntv3_8M_7downsample_pretrained_le_1mb

InstaDeepAI/ntv3_106M_7downsample_pretrained_le_1mb

InstaDeepAI/ntv3_650M_7downsample_pretrained_le_1mb
Post-trained checkpoints:

InstaDeepAI/ntv3_650M_7downsample_post_trained_1mb

InstaDeepAI/ntv3_106M_7downsample_post_trained_1mb

Model usage (to update)

Here is a quick example of how to use NTv3 models.

from transformers import pipeline

pipe = pipeline(
    task="ntv3-tracks",
    model="InstaDeepAI/ntv3_106M_7downsample_post_trained_1mb",
    trust_remote_code=True,
    device="cuda",
    torch_dtype=torch.bfloat16,
)

Links

Paper: (add link)
JAX training code