[Bugfix] Fix prefix strings for quantized VLMs (#9772)

vllm/model_executor/models/llava_next_video.py

@@ -257,14 +257,20 @@ class LlavaNextVideoForConditionalGeneration(nn.Module, SupportsMultiModal,
 
         # Initialize the vision tower only up to the required feature layer
         self.vision_tower = init_vision_tower_for_llava(
-            config, quant_config, require_post_norm=False)
+            config,
+            quant_config,
+            require_post_norm=False,
+            prefix="vision_tower")
         self.vision_resampler = LlavaNextVideoPooler(config)
         self.multi_modal_projector = LlavaNextMultiModalProjector(
             vision_hidden_size=config.vision_config.hidden_size,
             text_hidden_size=config.text_config.hidden_size,
             projector_hidden_act=config.projector_hidden_act)
         self.language_model = init_vllm_registered_model(
-            config.text_config, cache_config, quant_config)
+            config.text_config,
+            cache_config,
+            quant_config,
+            prefix="language_model")
 
         self.make_empty_intermediate_tensors = (
             self.language_model.model.make_empty_intermediate_tensors)