biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Multi Modal] Configurable MM Profiling (#25631)

Browse Source 
Signed-off-by: wwl2755 <wangwenlong2755@gmail.com>
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Wenlong Wang
2025-10-03 03:59:10 -07:00
committed by GitHub
parent 2ed3f20dba
commit 79aa244678
60 changed files with 654 additions and 99 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
vllm/model_executor/models/llava_onevision.py
View File

@@ -13,6 +13,7 @@ from transformers.models.llava_onevision.modeling_llava_onevision import (
get_anyres_image_grid_shape, unpad_image)
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.model_executor.layers.activation import get_act_fn
from vllm.multimodal import MULTIMODAL_REGISTRY
from vllm.multimodal.inputs import (MultiModalDataDict, MultiModalFieldConfig,
@@ -254,6 +255,7 @@ class LlavaOnevisionDummyInputsBuilder(
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
) -> MultiModalDataDict:
num_images = mm_counts.get("image", 0)
num_videos = mm_counts.get("video", 0)
@@ -264,17 +266,22 @@ class LlavaOnevisionDummyInputsBuilder(
self.info.get_num_frames_with_most_features(seq_len,
mm_counts)
image_overrides = mm_options.get("image") if mm_options else None
video_overrides = mm_options.get("video") if mm_options else None
return {
"image":
self._get_dummy_images(width=target_width,
height=target_height,
num_images=num_images),
num_images=num_images,
overrides=image_overrides),
"video":
self._get_dummy_videos(
width=target_width,
height=target_height,
num_frames=target_num_frames,
num_videos=num_videos,
overrides=video_overrides,
)
}