lalala404/healformer-nside256-mixed

HealFormer nside256 mixed-mask model

Inference-ready HealFormer v0.2.0 checkpoint for NESTED HEALPix maps at Nside=256. One checkpoint is used unchanged for the fixed KiDS, DES, DECaLS, and Planck masks; evaluation does not rotate the maps.

The public repository deliberately uses standard short filenames: config.json and model.safetensors.

100-sample diagnostics

Statistics are computed on 100 independent skies. Power error is the RMSE of the predicted-to-true power-spectrum ratio around one. Cross correlation is the mean harmonic cross-correlation coefficient.

Fixed mask	Power-ratio RMSE	Mean cross correlation
KiDS	0.1748 ± 0.2656	0.9501 ± 0.0058
DES	0.0892 ± 0.0143	0.9569 ± 0.0023
DECaLS	0.0542 ± 0.0163	0.9762 ± 0.0013
Planck	0.0544 ± 0.0044	0.9746 ± 0.0011

The uncertainty is one sample standard deviation, not standard error.

Usage

from healformers import MassMappingPipeline

pipeline = MassMappingPipeline.from_pretrained(
    "lalala404/healformer-nside256-mixed"
)
kappa = pipeline(gamma1, gamma2, mask_npix)

Inputs are physical gamma1, gamma2, and integer mask_npix arrays in NESTED ordering. Mask values are 0 visible, 1 reconstruction edge, and 2 unseen. The returned convergence map is in physical units.

Integrity and limitations

release-manifest.json records byte sizes and SHA-256 checksums. This model is for simulated weak-lensing mass mapping at the stated resolution and should be validated before scientific use on a new survey pipeline.

Citation

Yihe Wang and Yu Yu, Advancing weak lensing mass mapping with a mask-aware HEALPix transformer, arXiv:2603.25471.