[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
@@ -27,7 +27,6 @@ from packaging.version import Version
from transformers.utils.generic import ModelOutput
from vllm.compilation.decorators import should_torch_compile_mm_encoder
from vllm.config.multimodal import AudioDummyOptions
from vllm.config.utils import getattr_iter
from vllm.inputs import (
MultiModalDataBuiltins,
@@ -75,7 +74,7 @@ if TYPE_CHECKING:
from transformers import BatchFeature, PreTrainedModel
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.config.multimodal import MultiModalDummyOptions
logger = init_logger(__name__)
@@ -195,7 +194,7 @@ class MultiModalDummyInputsBuilder(BaseDummyInputsBuilder[MultiModalProcessingIn
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Mapping[str, "BaseDummyOptions"],
mm_options: "MultiModalDummyOptions",
) -> MultiModalDataDict:
data = MultiModalDataBuiltins()
if self.info._is_audio_model() and (num_audios := mm_counts.get("audio", 0)):
@@ -204,21 +203,16 @@ class MultiModalDummyInputsBuilder(BaseDummyInputsBuilder[MultiModalProcessingIn
chunk_length = getattr(sub, "chunk_length", None) if sub else None
if chunk_length is None:
chunk_length = 30
audio_len = int(chunk_length * sampling_rate)
audio_overrides = mm_options.get("audio")
assert audio_overrides is None or isinstance(
audio_overrides, AudioDummyOptions
)
data["audio"] = self._get_dummy_audios(
length=audio_len,
length=int(chunk_length * sampling_rate),
num_audios=num_audios,
overrides=audio_overrides,
overrides=mm_options.get("audio"),
)
if self.info._is_image_model() and (num_images := mm_counts.get("image", 0)):
target_width, target_height = self.info.get_image_size_with_most_features()
width, height = self.info.get_image_size_with_most_features()
data["image"] = self._get_dummy_images(
width=target_width,
height=target_height,
width=width,
height=height,
num_images=num_images,
overrides=mm_options.get("image"),
)