Are there any vision encoder to replace the " CLIP-Large-336 and CLIP-ConvNext-320-d " ?

[RonaldYuren]

Are there any vision encoder to replace the " CLIP-Large-336 and CLIP-ConvNext-320-d " because when I load these models' weight ( message like: " Loading pretrained weights (convnext_large_d_320) " ) it came across with CUDA out of memory.
So maybe there are other choices even with lower efficiency?