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

11
vllm/model_executor/models/phi4mm.py
View File

@@ -11,6 +11,7 @@ from transformers import (BatchFeature, PretrainedConfig, ProcessorMixin,
SequenceFeatureExtractor, SiglipVisionConfig)
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.distributed import get_pp_group
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization import QuantizationConfig
@@ -749,6 +750,7 @@ class Phi4MMDummyInputsBuilder(BaseDummyInputsBuilder[Phi4MMProcessingInfo]):
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
) -> MultiModalDataDict:
num_audios = mm_counts.get("audio", 0)
num_images = mm_counts.get("image", 0)
@@ -756,14 +758,19 @@ class Phi4MMDummyInputsBuilder(BaseDummyInputsBuilder[Phi4MMProcessingInfo]):
target_width, target_height = \
self.info.get_image_size_with_most_features()
image_overrides = mm_options.get("image") if mm_options else None
audio_overrides = mm_options.get("audio") if mm_options else None
mm_data = {
"image":
self._get_dummy_images(width=target_width,
height=target_height,
num_images=num_images),
num_images=num_images,
overrides=image_overrides),
"audio":
self._get_dummy_audios(length=_AUDIO_MAX_SOUNDFILE_SIZE,
num_audios=num_audios),
num_audios=num_audios,
overrides=audio_overrides),
}
return mm_data