OpenCLIP Adds Layer-Wise LR Decay to Ease CLIP Fine-Tuning
OpenCLIP contributors added layer-wise learning rate decay for its text and image encoders to mitigate forgetting during fine-tuning under distribution shift, addressing the common failure of contrastive fine-tuning CLIP with insufficient or out-of-distribution data.
2026-11-13 ~ 2026-11-13 · 2 related posts
- Further fine-tuning CLIP with contrastive losses often fails on out-of-distribution data — wightmanr · 2026-11-13
- OpenCLIP adds layer-wise LR decay for text and image encoders to curb forgetting under distribution shift — wightmanr · 2026-11-13