diff --git a/tensorrt_llm/_torch/models/modeling_qwen3vl.py b/tensorrt_llm/_torch/models/modeling_qwen3vl.py
index 4bdcfc767844..e08a11e5749a 100644
--- a/tensorrt_llm/_torch/models/modeling_qwen3vl.py
+++ b/tensorrt_llm/_torch/models/modeling_qwen3vl.py
@@ -739,9 +739,11 @@ def forward(
 
         # Getting positional embedding
         rotary_pos_emb = self.rot_pos_emb(grid_thw)
+        pos_embeds = self.fast_pos_embed_interpolate(grid_thw)
 
         # From this point, pure GPU operation
         hidden_states = self.patch_embed(pixel_values)
+        hidden_states = hidden_states + pos_embeds
         seq_len, _ = hidden_states.size()
         hidden_states = hidden_states.reshape(seq_len, -1)
 
@@ -1139,6 +1141,7 @@ def _get_requests_with_mm_data(self, multimodal_params):
             "video": "<|vision_start|><|video_pad|><|vision_end|>",
         },
         placeholder_placement=MultimodalPlaceholderPlacement.BEFORE_TEXT,
+        placeholders_separator="",
     ),
 )
 class Qwen3VLModel(Qwen3VLModelBase):
diff --git a/tensorrt_llm/_torch/models/modeling_qwen3vl_moe.py b/tensorrt_llm/_torch/models/modeling_qwen3vl_moe.py
index 74cd16ec69c1..66da3c153e2a 100644
--- a/tensorrt_llm/_torch/models/modeling_qwen3vl_moe.py
+++ b/tensorrt_llm/_torch/models/modeling_qwen3vl_moe.py
@@ -41,6 +41,7 @@
             "video": "<|vision_start|><|video_pad|><|vision_end|>",
         },
         placeholder_placement=MultimodalPlaceholderPlacement.BEFORE_TEXT,
+        placeholders_separator="",
     ),
 )
 class Qwen3MoeVLModel(Qwen3VLModelBase):