[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
@@ -18,7 +18,7 @@ import torch.nn as nn
from transformers import BatchFeature
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.config.multimodal import MultiModalDummyOptions
from vllm.forward_context import set_forward_context
from vllm.inputs import MultiModalDataDict
from vllm.multimodal import MULTIMODAL_REGISTRY
@@ -193,7 +193,7 @@ class HCXVisionV2DummyInputsBuilder(BaseDummyInputsBuilder[HCXVisionV2Processing
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Mapping[str, BaseDummyOptions] | None = None,
mm_options: MultiModalDummyOptions | None = None,
mm_processor_kwargs: Mapping[str, object] | None = None,
) -> ProcessorInputs:
"""Build dummy processor inputs for memory profiling."""
@@ -225,31 +225,25 @@ class HCXVisionV2DummyInputsBuilder(BaseDummyInputsBuilder[HCXVisionV2Processing
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Mapping[str, BaseDummyOptions] | None = None,
mm_options: MultiModalDummyOptions | None = None,
mm_processor_kwargs: Mapping[str, object] | None = None,
) -> MultiModalDataDict:
num_images = mm_counts.get("image", 0)
num_videos = mm_counts.get("video", 0)
target_width, target_height = self.info.get_image_size_with_most_features()
target_num_frames = 16 # Default for video
image_overrides = mm_options.get("image") if mm_options else None
video_overrides = mm_options.get("video") if mm_options else None
result: MultiModalDataDict = {
"image": self._get_dummy_images(
width=target_width,
height=target_height,
num_images=num_images,
overrides=image_overrides, # type: ignore
num_images=mm_counts.get("image", 0),
overrides=mm_options.get("image") if mm_options else None,
),
"video": self._get_dummy_videos(
width=target_width,
height=target_height,
num_frames=target_num_frames,
num_videos=num_videos,
overrides=video_overrides, # type: ignore
num_videos=mm_counts.get("video", 0),
overrides=mm_options.get("video") if mm_options else None,
),
}