A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Docs] Fix griffe warnings in vllm/multimodal (#25216)
Signed-off-by: windsonsea <haifeng.yao@daocloud.io>
Michael Yao committed
Sep 19, 2025 at 13:10 UTC
6c8a3c099bc12bb00fa3899753e41183f05fe9fc
2 files changed
+6
-4
vllm/multimodal/inputs.py
+3
-3
@@ -569,8 +569,8 @@ class MultiModalFieldConfig:
569
Args:
570
modality: The modality of the multi-modal item that uses this
571
keyword argument.
572
- slices: For each multi-modal item, the size of the slice that
573
- is used to extract the data corresponding to it.
572
+ size_per_item: For each multi-modal item, the size of the slice
573
+ that is used to extract the data corresponding to it.
574
dim: The dimension to slice, default to 0.
575
576
Example:
@@ -590,7 +590,7 @@ class MultiModalFieldConfig:
590
591
```
592
Given:
593
- slices: [3, 4, 2]
593
+ size_per_item: [3, 4, 2]
594
dim: 1
595
596
Input:
vllm/multimodal/utils.py
+3
-1
@@ -395,7 +395,9 @@ def group_mm_kwargs_by_modality(
395
modality together into the same `MultiModalKwargs` instance.
396
397
Args:
398
- mm_inputs: List of `MultiModalKwargsItem`.
398
+ mm_kwargs: List of `MultiModalKwargsItem`.
399
+ device: The device to place the grouped tensors on.
400
+ pin_memory: Whether to pin memory for faster host-to-device transfer.
401
402
Yields:
403
A tuple `(modality, num_items, grouped_kwargs)`.