[Docs] Fix griffe warnings in vllm/multimodal (#25216)

Signed-off-by: windsonsea <haifeng.yao@daocloud.io>

vllm/multimodal/inputs.py

@@ -569,8 +569,8 @@ class MultiModalFieldConfig:
         Args:
             modality: The modality of the multi-modal item that uses this
                 keyword argument.
-            slices: For each multi-modal item, the size of the slice that
-                is used to extract the data corresponding to it.
+            size_per_item: For each multi-modal item, the size of the slice
+                that is used to extract the data corresponding to it.
             dim: The dimension to slice, default to 0.
 
         Example:
@@ -590,7 +590,7 @@ class MultiModalFieldConfig:
 
         ```
         Given:
-            slices: [3, 4, 2]
+            size_per_item: [3, 4, 2]
             dim: 1
 
         Input:

vllm/multimodal/utils.py

@@ -395,7 +395,9 @@ def group_mm_kwargs_by_modality(
     modality together into the same `MultiModalKwargs` instance.
 
     Args:
-        mm_inputs: List of `MultiModalKwargsItem`.
+        mm_kwargs: List of `MultiModalKwargsItem`.
+        device: The device to place the grouped tensors on.
+        pin_memory: Whether to pin memory for faster host-to-device transfer.
 
     Yields:
         A tuple `(modality, num_items, grouped_kwargs)`.