---
license: mit
library_name: pytorch
tags:
- biology
- molecule-generation
- antimicrobial-peptides
- apexoracle
---

# ApexOracle guided-generation quickstart assets

This repository contains inference-only checkpoints and one BAA-3170
genome/text condition bundle for the ApexOracle 256-step MIC+peptide guided
generation smoke test.

The three checkpoints preserve the deployed tensor state while omitting
optimizer state, training loops, callbacks, and the unused MIC classification
head. The condition bundle contains only the target used by the public smoke;
it does not redistribute the complete paper condition bank.

Use these files with the fixed ApexOracle super-repository and
`modules/generation/scripts/reproduce/run_paper_mic_peptide.py --asset-root`.
The smoke validates runtime and output contracts. A generated sample is not an
experimental activity result, and sampling is not bitwise deterministic on
CUDA.

See `manifest.json` for sizes, SHA-256 values, source-checkpoint provenance,
and the validated protocol.