lazyslide_models.vision.GigaPathFlash

lazyslide_models.vision.GigaPathFlash#

class GigaPathFlash(model_path=None, token=None)#

Bases: TimmViTModel

gigapath-flash 🤗Hugging Face GitHub Paper Params: 21.7M Apache 2.0 [Usuyama et al., 2026] An efficient whole-slide foundation model for digital pathology GigaPath-Flash tile encoder: a DINOv2-small ViT-S/16 with SwiGLU FFN.

transform_kws = {'crop_pct': 0.875}#

Upstream eval recipe, forwarded to get_default_transform (interpolation, crop_pct, mean, std). Registered models pin it because hub pretrained_cfg entries are often stale; None (a plain timm name) reads the model’s own pretrained_cfg.