Skip to content
 
 

Repository files navigation

maxtoki-perturb

In-silico dynamic ANY GENE perturbation modules for MaxToki, NVIDIA + Christina's temporal model for predicting cell aging trajectories.

The upstream release is CUDA-only and requires NVIDIA BioNeMo.

Run from Apptainer/Docker using the .def (->.sif) image provided in upstream

Status

  • [Done] 217M (float32), zero-shot perturbation prediction results with viz
  • [Done] 217M (float32), rank re-ordered prediction
  • [TO-DO] 217M (float32), attention matrix sparsity implementation for multi-modality using Peak-to-Gene mass transport

Model details

MaxToki is a decoder-only Llama trained on 175M cells. Two variants published on HuggingFace:

217M 1B
Layers 11 20
Hidden 1232 2304
Heads 8 16
KV heads 8 8 (GQA)
Head dim 154 144
Vocab 20275 20275
RoPE standard llama3 scaled

Input format: rank-value encoding of gene_name, where its expression median-normalized across entire 175M corpus gives its rank. Each cell becomes [<bos>, gene_rank_1, gene_rank_2, ..., <eos>]

Credits

  • MaxToki: J Gomez Ortega et al., Theodoris Lab @ Gladstone / NVIDIA BioNeMo, bioRxiv 10.64898/2026.03.30.715396
  • Geneformer (tokenizer architecture): Theodoris et al.
  • MLX-port: Apple ML team & @srijitiyer

License

Apache 2.0 (matches upstream MaxToki license).

About

in-silico perturbation implementation of MaxToki - temporal cell aging model by Gladstone + NVIDIA

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages