baseline

timm/models/vision_transformer.py

@@ -48,7 +48,6 @@ from ._builder import build_model_with_cfg
 from ._manipulate import named_apply, checkpoint_seq, adapt_input_conv
 from ._registry import generate_default_cfgs, register_model, register_model_deprecations
 
-from ttt import TTTForCausalLM, TTTConfig, TTT_STANDARD_CONFIGS
 
 __all__ = ['VisionTransformer']  # model_registry will add each entrypoint fn to this
 
@@ -153,17 +152,13 @@ class Block(nn.Module):
         self.drop_path1 = DropPath(drop_path) if drop_path > 0. else nn.Identity()
 
         self.norm2 = norm_layer(dim)
-        # self.mlp = mlp_layer(
-        #     in_features=dim,
-        #     hidden_features=int(dim * mlp_ratio),
-        #     act_layer=act_layer,
-        #     drop=proj_drop,
-        # )
-        
-        ############## TTT-MLP ###################
+        self.mlp = mlp_layer(
+            in_features=dim,
+            hidden_features=int(dim * mlp_ratio),
+            act_layer=act_layer,
+            drop=proj_drop,
+        )
         
-        tttmlp_config = TTTConfig()
-        self.mlp = TTTMLP(tttmlp_config)
         self.ls2 = LayerScale(dim, init_values=init_values) if init_values else nn.Identity()
         self.drop_path2 = DropPath(drop_path) if drop_path > 0. else nn.Identity()