generated: '2026-07-20'
method: searched
source: https://github.com/outpost-bio/waypoint
name: waypoint
summary: >-
  Command-line tool shipped with the waypoint-bio Python package for pretraining,
  benchmarking, fine-tuning, and embedding microbiome taxonomic-abundance foundation
  models. Data and gated model checkpoints load from the Hugging Face Hub.
package: waypoint-bio
install:
- method: pip
  command: pip install waypoint-bio
commands:
- name: pretrain
  description: Pretrain a GPT-2 style transformer on microbiome abundance data (Atlas dataset).
- name: benchmark
  description: Evaluate a model against the 8 downstream Compass tasks.
- name: finetune
  description: Adapt a pretrained checkpoint to a custom labeled dataset.
- name: embed
  description: Produce fixed-size embedding vectors from microbiome samples.
- name: prepare-dataset
  description: Convert abundance matrices (parquet/CSV taxa + relative-abundance) into Waypoint format.
notes: >-
  Atlas, Compass, and the published Waypoint checkpoints (e.g. outpost-bio/Waypoint-6m)
  are gated on the Hugging Face Hub and require an access request.