Reproducible scaling laws for contrastive language-image learning (https://arxiv.org/abs/2212.07143)
Terminal score: 0–1000 raw, weighted across 4 dimensions. Public score: 0–10 normalized (shown in the 30-day stars chart above).
