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

10
vllm/model_executor/models/pixtral.py
View File

@@ -24,6 +24,7 @@ from transformers.models.pixtral.modeling_pixtral import (
from transformers.tokenization_utils_base import TextInput
from vllm.config import VllmConfig
from vllm.config.multimodal import BaseDummyOptions
from vllm.distributed import divide, get_tensor_model_parallel_world_size
from vllm.model_executor.layers.activation import get_act_and_mul_fn
from vllm.model_executor.layers.layernorm import RMSNorm
@@ -228,28 +229,33 @@ class PixtralDummyInputsBuilder(BaseDummyInputsBuilder[PixtralProcessingInfo]):
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
) -> 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") if mm_options else None
return {
"image":
self._get_dummy_images(width=target_width,
height=target_height,
num_images=num_images)
num_images=num_images,
overrides=image_overrides)
}
def get_dummy_processor_inputs(
self,
seq_len: int,
mm_counts: Mapping[str, int],
mm_options: Optional[Mapping[str, BaseDummyOptions]] = None,
) -> ProcessorInputs:
tokenizer = self.info.get_tokenizer()
dummy_text = self.get_dummy_text(mm_counts)
dummy_mm_data = self.get_dummy_mm_data(seq_len, mm_counts)
dummy_mm_data = self.get_dummy_mm_data(seq_len, mm_counts, mm_options)
dummy_images = dummy_mm_data.get("image", [])
tokenization_kwargs = {"truncation": False}