Fix checkpoint weight loading for Qwen2.5-VL vision and embedding models (microsoft#380)

Author: apsonawane

Qwen-Qwen2.5-VL-3B-Instruct/builtin/codes/modeling_qwen2_5_vl.py

@@ -1039,8 +1039,8 @@ def forward(
 
 @auto_docstring
 class Qwen2_5_VLModel(Qwen2_5_VLPreTrainedModel):
-    base_model_prefix = ""
-    _checkpoint_conversion_mapping = {"^model": "language_model"}
+    base_model_prefix = "model"
+    _checkpoint_conversion_mapping = {}
     # Reference: fix gemma3 grad acc #37208
     accepts_loss_kwargs = False
     config: Qwen2_5_VLConfig