OpenCLIP Adds MaMMUT Support and MaMMUT2 Validation Experiments

OpenCLIP now supports Google's MaMMUT architecture (image encoder plus text decoder) with fixed pooling and a new MaMMUT2, validated via small-scale CC12M experiments, alongside a collection of OpenMaMMUT weights on DataComp, DFN and Re-LAION.

2026-08-20 ~ 2026-08-20 · 4 related posts

• OpenCLIP Adds MaMMUT Support: Fixes Pooling, Introduces MaMMUT2 — wightmanr · 2026-08-20
• MaMMUT2 Architecture Validation: Multiple CC12M Training Runs Compared — wightmanr · 2026-08-20
• OpenMaMMUT Collection: Weights on DataComp and Re-LAION Available — wightmanr · 2026-08-20
• Google Research: MaMMUT for Joint Multimodal Learning — wightmanr · 2026-08-20