fix(models): Forward timm model kwargs to timm.create_model for OmDet-Turbo

src/transformers/models/omdet_turbo/configuration_omdet_turbo.py

@@ -161,6 +161,13 @@ def __init__(
             **kwargs,
         )
 
+        # Extract timm.create_model kwargs; TimmBackbone doesn't forward arbitrary config attrs to timm
+        timm_kwargs = {}
+        if getattr(backbone_config, "model_type", None) == "timm_backbone":
+            for attr in ("img_size", "always_partition"):
+                if hasattr(backbone_config, attr):
+                    timm_kwargs[attr] = getattr(backbone_config, attr)
+
         if text_config is None:
             logger.info("`text_config` is `None`. Initializing the config with the default `clip_text_model`")
             text_config = CONFIG_MAPPING["clip_text_model"]()
@@ -212,8 +219,14 @@ def __init__(
         self.eval_size = eval_size
         self.learn_initial_query = learn_initial_query
         self.cache_size = cache_size
+        self.timm_kwargs = timm_kwargs
 
         super().__init__(is_encoder_decoder=is_encoder_decoder, **kwargs)
 
+    def to_dict(self):
+        output = super().to_dict()
+        output.pop("timm_kwargs", None)
+        return output
+
 
 __all__ = ["OmDetTurboConfig"]

src/transformers/models/omdet_turbo/modeling_omdet_turbo.py

@@ -26,7 +26,6 @@
 
 from ... import initialization as init
 from ...activations import ACT2CLS, ACT2FN
-from ...backbone_utils import load_backbone
 from ...integrations import use_kernel_forward_from_hub
 from ...masking_utils import create_bidirectional_mask
 from ...modeling_layers import GradientCheckpointingLayer
@@ -39,7 +38,7 @@
     logging,
     torch_compilable_check,
 )
-from ..auto import AutoModel
+from ..auto import AutoBackbone, AutoModel
 from .configuration_omdet_turbo import OmDetTurboConfig
 
 
@@ -279,7 +278,7 @@ class OmDetTurboVisionBackbone(nn.Module):
     def __init__(self, config: OmDetTurboConfig):
         super().__init__()
         self.apply_layernorm_after_vision_backbone = config.apply_layernorm_after_vision_backbone
-        self.vision_backbone = load_backbone(config)
+        self.vision_backbone = AutoBackbone.from_config(config.backbone_config, **getattr(config, "timm_kwargs", {}))
         self.layer_norms = nn.ModuleList(
             [nn.LayerNorm(in_channel_dim, eps=config.layer_norm_eps) for in_channel_dim in config.encoder_in_channels]
         )