siglip2¶
Model type |
|
Class |
|
Task |
|
Source |
|
Description¶
SigLIP vision model for standalone image feature extraction.
SigLIP is similar to CLIP but without a CLS token and without pre-layernorm. The position embedding size matches num_patches exactly (no +1 for CLS).
Usage¶
mobius build --model <MODEL_ID> output_dir/
from mobius import build
model = build("<MODEL_ID>")