Phase 2 paper support checkpoints v1

Frozen paper-only assets needed to reproduce the Phase 2 canid and Arabis analyses after Betty/Sesame retirement. These are not general-purpose user models; use the six separately versioned model releases for new inference.

The checkpoint files retain the exact trained state_dict and loading metadata while omitting optimizer and trainer state. The fastrho reproduce/checkpoints.json registry records both each downloadable SHA-256 and the original full training-checkpoint SHA-256. SHA256SUMS identifies every release asset.

fastrho selfing-v1 — Arabidopsis specialist

Selfing-aware checkpoint used for the paper Arabidopsis analysis. This is the empirically retained self2 model, not the later 15k retrain that transferred worse to real Arabidopsis. It requires phased, ancestrally polarized haplotypes and input_mode="phased". The embedded model card records the selection history and limitations.

fastrho high-ne-v1 — mosquito and dipteran specialist

Phased high-effective-population-size specialist used for the Anopheles atlas and dipteran analyses. The training regime extends to approximately Ne=2e6. Use input_mode="phased" and validate absolute-rate calibration on cohort-matched simulations. The release includes the exact paper checkpoint, matching statistics, and a deterministic checksummed bundle.

fastrho domain-randomized-v1

Primary domain-randomized checkpoint used by the fastrho paper. The release contains the exact checkpoint, its matching feature statistics, and a deterministic archive with the frozen manifest, model card, license, and SHA256SUMS. Checkpoint SHA-256: d1ef4caae0c16b8c0cb1e5cfc79d1982950bff2e3509ea08a9f244fc9bb3f79f. Statistics SHA-256: 6198b021f1df6bba48cb672a71386d188714f3e87c5cb166dbdee25c497e55ba. Archive SHA-256: b4c711f7031372e11a85229ac726675bbd16ad6875030d581e5cab2eeb1e97d7.

fastrho dog-bottleneck-v1 — canine specialist

Folded composite-LD specialist for village-dog and breed-like bottleneck histories. Use input_mode="unpolarized". The released 15k checkpoint uses the default 5/25/50 kb LD radii; its model card distinguishes it from the earlier long-range prototype and explains why breed-only fine-scale recovery can be unidentifiable.

fastrho composite-ld-v1

Folded composite-LD specialist for unphased, unpolarized diploid genotypes. Use input_mode="unpolarized". The release contains the exact retained 15k checkpoint, metadata-enriched matching statistics, and a deterministic checksummed bundle. See the embedded MODEL_CARD.md for qualification limits.

fastrho base-v1

Phased, ancestrally polarized ordinary-demography specialist. The release contains the exact retained 15k checkpoint, its metadata-enriched matching feature statistics, and a deterministic checksummed bundle. Use input_mode="phased". See the embedded MODEL_CARD.md for qualification limits.