From 7234f5c6c55ba4e538ad339dcb804ebd72707511 Mon Sep 17 00:00:00 2001
From: Ross Wightman <rwightman@gmail.com>
Date: Mon, 17 Feb 2025 12:59:10 -0800
Subject: [PATCH] Add 448 so150m2 weight/model, add updated internvit 300m
 weight

---
 timm/models/_hub.py               |  2 +-
 timm/models/vision_transformer.py | 20 ++++++++++++++++++++
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/timm/models/_hub.py b/timm/models/_hub.py
index 809705ed..1da0942b 100644
--- a/timm/models/_hub.py
+++ b/timm/models/_hub.py
@@ -395,7 +395,7 @@ def push_to_hf_hub(
 
 
 def generate_readme(model_card: dict, model_name: str):
-    tags = model_card.get('tags', None) or ['image-classification', 'timm']
+    tags = model_card.get('tags', None) or ['image-classification', 'timm', 'transformers']
     readme_text = "---\n"
     if tags:
         readme_text += "tags:\n"
diff --git a/timm/models/vision_transformer.py b/timm/models/vision_transformer.py
index 6dbd758e..bba7afac 100644
--- a/timm/models/vision_transformer.py
+++ b/timm/models/vision_transformer.py
@@ -2174,12 +2174,20 @@ default_cfgs = {
     'vit_so150m2_patch16_reg1_gap_384.sbb_e200_in12k_ft_in1k': _cfg(
         hf_hub_id='timm/',
         input_size=(3, 384, 384), crop_pct=1.0),
+    'vit_so150m2_patch16_reg1_gap_448.sbb_e200_in12k_ft_in1k': _cfg(
+        hf_hub_id='timm/',
+        input_size=(3, 448, 448), crop_pct=1.0, crop_mode='squash'),
 
     'vit_intern300m_patch14_448.ogvl_dist': _cfg(
         hf_hub_id='timm/',
         mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD,
         input_size=(3, 448, 448), crop_pct=1.0, num_classes=0,
     ),
+    'vit_intern300m_patch14_448.ogvl_2pt5': _cfg(
+        hf_hub_id='timm/',
+        mean=IMAGENET_DEFAULT_MEAN, std=IMAGENET_DEFAULT_STD,
+        input_size=(3, 448, 448), crop_pct=1.0, num_classes=0,
+    ),
 
     'aimv2_large_patch14_224.apple_pt': _cfg(
         hf_hub_id='timm/',
@@ -3538,6 +3546,18 @@ def vit_so150m2_patch16_reg1_gap_384(pretrained: bool = False, **kwargs) -> Visi
     return model
 
 
+@register_model
+def vit_so150m2_patch16_reg1_gap_448(pretrained: bool = False, **kwargs) -> VisionTransformer:
+    """ SO150M v2 (shape optimized, but diff than paper def, optimized for GPU) """
+    model_args = dict(
+        patch_size=16, embed_dim=832, depth=21, num_heads=13, mlp_ratio=34/13, init_values=1e-5,
+        qkv_bias=False, class_token=False, reg_tokens=1, global_pool='avg',
+    )
+    model = _create_vision_transformer(
+        'vit_so150m2_patch16_reg1_gap_448', pretrained=pretrained, **dict(model_args, **kwargs))
+    return model
+
+
 @register_model
 def vit_intern300m_patch14_448(pretrained: bool = False, **kwargs) -> VisionTransformer:
     model_args = dict(