diff --git a/vllm/model_executor/models/deepseek_vl2.py b/vllm/model_executor/models/deepseek_vl2.py
index 0faf895964bb..4554a997755f 100644
--- a/vllm/model_executor/models/deepseek_vl2.py
+++ b/vllm/model_executor/models/deepseek_vl2.py
@@ -509,7 +509,7 @@ def _pixel_values_to_embedding(
         _, hw, n_dim = images_embeds.shape
         h = w = int(hw**0.5)
 
-        # 根据self.tile_tag & self.global_view_pos填充image token sequence
+        # fill image token based on self.tile_tag & self.global_view_pos
         tile_index = 0
         vision_embeddings = []
         for jdx in range(images_spatial_crop.size(0)):
diff --git a/vllm/transformers_utils/processors/deepseek_vl2.py b/vllm/transformers_utils/processors/deepseek_vl2.py
index d37381ea9925..316281f2af4e 100644
--- a/vllm/transformers_utils/processors/deepseek_vl2.py
+++ b/vllm/transformers_utils/processors/deepseek_vl2.py
@@ -226,7 +226,7 @@ def process_one(
         input_ids[input_ids < 0] = self.pad_id
 
         if inference_mode:
-            # 去掉结尾的eos token
+            # Remove the ending eos token
             assert input_ids[-1] == self.eos_id
             input_ids = input_ids[:-1]
             target_ids = target_ids[:-1]