Model
A ViT-Base vision transformer with 87 million parameters. It learns from images first, without labels, and is then tuned for redshift.
1 · Any filters in
2 · A distance out
3 · Learning without labels
4 · Speed
| Method | Per galaxy | 595k COSMOS-Web galaxies | Hardware |
|---|---|---|---|
| This work, images, 8 rotations | 2.3 ms | 23 min | one H100 GPU |
| This work, images, 1 pass | 0.3 ms | 3 min | one H100 GPU |
| NNPZ (Euclid DR1 PHZ) | ~6 ms | ~1 h | 24-core Xeon workstation |
| EAZY | ~40 ms | ~6 h | 2008 laptop CPU |
| LePhare | ~1 ms – 1 s | 10 min – 1 week | one CPU core, depends on the template grid |
| Phosphoros (Euclid Q1 PHZ) | ~0.8 s | ~5.5 days | 24-core Xeon workstation |
| Bayesian SED fit (Prospector-type) | 10–25 CPU-h | ~10 million CPU-h | CPU |
Sources: NNPZ and Phosphoros from Euclid Collaboration (2024, arXiv:2407.07940), on a 24-core Xeon Silver 4116. EAZY from Brammer et al. (2008, arXiv:0807.1533): 6,300 galaxies in about 4 minutes. LePhare has no standard benchmark: the Rubin LePhare plan targets about 1 ms per object on a coarse grid, and comparisons often assume about 1 s for a full fit. Bayesian SED fits from Leja et al. and Tacchella et al. (2022), as summarised in arXiv:2506.19919. We found no published runtimes for DEmP or Mizuki. Our 595k times scale the measured 20,449-galaxy runs linearly.