[Core] Initialize LoRA support for tower and connector in multi-modal models (#26674)

Signed-off-by: bk-201 <joy25810@foxmail.com>
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
Signed-off-by: prashanth058 <prashanth.dannamaneni@uipath.com>
Co-authored-by: bk-201 <joy25810@foxmail.com>
Co-authored-by: prashanth058 <prashanth.dannamaneni@uipath.com>
Co-authored-by: Anexdeus <5142168@mail.ru>

vllm/model_executor/models/qwen2_5_vl.py

@@ -1026,6 +1026,7 @@ class Qwen2_5_VLForConditionalGeneration(
     packed_modules_mapping = {
         "qkv_proj": ["q_proj", "k_proj", "v_proj"],
         "gate_up_proj": ["gate_proj", "up_proj"],
+        "qkv": ["qkv"],  # For vision tower's already-packed QKV
     }
 
     # To ensure correct weight loading and mapping.
@@ -1568,6 +1569,25 @@ class Qwen2_5_VLForConditionalGeneration(
             tower_model="visual.",
         )
 
+    def get_num_mm_encoder_tokens(
+        self,
+        num_image_tokens: int,
+    ) -> int:
+        hf_config = self.config
+        vision_config = hf_config.vision_config
+        merge_size = vision_config.spatial_merge_size
+
+        return num_image_tokens * merge_size**2
+
+    def get_num_mm_connector_tokens(
+        self,
+        num_vision_tokens: int,
+    ) -> int:
+        hf_config = self.config
+        vision_config = hf_config.vision_config
+        merge_size = vision_config.spatial_merge_size
+        return num_vision_tokens // merge_size**2
+
     @classmethod
     def get_language_model_spec(cls) -> tuple[nn.Module | None, str | None]:
         """