[None][chore]: Add note in Gemma3VL ctor

Signed-off-by: Balaram Buddharaju <[email protected]>

Author: brb-nv

tensorrt_llm/_torch/models/modeling_gemma3vl.py

@@ -147,6 +147,17 @@ def __init__(self, model_config: ModelConfig[Gemma3Config]):
                 f"the {_MULTIMODAL_ENV_NAME} environment variable, or set it to '0'."
             )
 
+        print(
+            "\n"
+            "#####################################################################################\n"
+            "NOTE: Gemma3VL decoder requires a custom mask while processing images.\n"
+            "To ensure expected behavior, please:\n"
+            "  - Use the FlashInfer attention backend\n"
+            "  - Disable chunked prefill\n"
+            "  - Disable KV cache reuse\n"
+            "#####################################################################################\n"
+            "\n")
+
         config = model_config.pretrained_config
         super().__init__(config)
 
@@ -276,17 +287,3 @@ def _get_image_features(self, pixel_values):
                                                attn_metadata=attn_metadata)[-1]
             image_features = self.mm_projector(image_features)
         return image_features
-
-
-def _load_weights_into_hf_module(
-    model: torch.nn.Module,
-    weights: dict,
-    prefix: str,
-    model_name: str,
-) -> None:
-    filtered_weights = filter_weights(prefix, weights)
-    missing_keys, _ = model.load_state_dict(filtered_weights)
-    if len(missing_keys) > 0:
-        raise KeyError(
-            f"Missing the following keys for the {model_name} in the checkpoint: "
-            f"[{', '.join(missing_keys)}].")