[Multimodal] Type dummy options per modality (#57576)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
Harry Mellor
2026-09-18 17:29:11 +01:00
committed by GitHub
co-authored by Claude Opus 5
parent 56cac80b14
commit 8a5cf54387
93 changed files with 461 additions and 720 deletions
+6 -13
View File
@@ -17,7 +17,7 @@ from transformers import (
)
from vllm.config import VllmConfig
from vllm.config.multimodal import AudioDummyOptions, BaseDummyOptions
from vllm.config.multimodal import MultiModalDummyOptions
from vllm.distributed import get_pp_group
from vllm.inputs import MultiModalDataDict
from vllm.model_executor.layers.logits_processor import LogitsProcessor
@@ -839,28 +839,21 @@ class Phi4MMDummyInputsBuilder(BaseDummyInputsBuilder[Phi4MMProcessingInfo]):
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Mapping[str, BaseDummyOptions],
mm_options: MultiModalDummyOptions,
) -> MultiModalDataDict:
num_audios = mm_counts.get("audio", 0)
num_images = mm_counts.get("image", 0)
target_width, target_height = self.info.get_image_size_with_most_features()
image_overrides = mm_options.get("image")
audio_overrides = mm_options.get("audio")
assert audio_overrides is None or isinstance(audio_overrides, AudioDummyOptions)
mm_data = {
"image": self._get_dummy_images(
width=target_width,
height=target_height,
num_images=num_images,
overrides=image_overrides,
num_images=mm_counts.get("image", 0),
overrides=mm_options.get("image"),
),
"audio": self._get_dummy_audios(
length=_AUDIO_MAX_SOUNDFILE_SIZE,
num_audios=num_audios,
overrides=audio_overrides,
num_audios=mm_counts.get("audio", 0),
overrides=mm_options.get("audio"),
),
}