[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
+5 -9
View File
@@ -21,7 +21,7 @@ from transformers.models.pixtral.image_processing_pixtral import (
from transformers.models.pixtral.modeling_pixtral import apply_rotary_pos_emb
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.config.multimodal import MultiModalDummyOptions
from vllm.distributed import divide, get_tensor_model_parallel_world_size
from vllm.inputs import MultiModalDataDict
from vllm.model_executor.layers.activation import SiluAndMul, get_act_and_mul_fn
@@ -208,20 +208,16 @@ class PixtralDummyInputsBuilder(BaseDummyInputsBuilder[PixtralProcessingInfo]):
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Mapping[str, BaseDummyOptions],
mm_options: MultiModalDummyOptions,
) -> MultiModalDataDict:
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")
return {
"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"),
)
}
@@ -229,7 +225,7 @@ class PixtralDummyInputsBuilder(BaseDummyInputsBuilder[PixtralProcessingInfo]):
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Mapping[str, BaseDummyOptions],
mm_options: MultiModalDummyOptions,
mm_data: MultiModalDataDict | None = None,
) -> ProcessorInputs:
tokenizer = self.info.get_tokenizer()