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.