xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-21 11:05:42 +08:00
Code Issues Packages Projects Releases Wiki Activity

[V0 Deprecation] Remove V0-only methods in multi-modal registry (#25362)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Cyrus Leung 2025-09-22 13:58:26 +08:00 committed by yewentao256
parent 1b3aa0f297
commit 243c358fa8
2 changed files with 1 additions and 32 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
tests/models/multimodal/generation/test_qwen2_vl.py
View File

@ -209,7 +209,6 @@ def batch_make_video_embeddings(
return visual(pixel_values_on_device, return visual(pixel_values_on_device,
grid_thw=video_grid_thw_on_device).cpu() grid_thw=video_grid_thw_on_device).cpu()
# V1 Test: this calls a V0 internal.
video_embeds = torch.concat(llm.apply_model(get_image_embeds)) video_embeds = torch.concat(llm.apply_model(get_image_embeds))
# split into original batches # split into original batches

32
vllm/multimodal/registry.py
View File

@ -12,8 +12,7 @@ from vllm.transformers_utils.tokenizer import (AnyTokenizer,
cached_tokenizer_from_config) cached_tokenizer_from_config)
from vllm.utils import ClassRegistry from vllm.utils import ClassRegistry
from .cache import (BaseMultiModalProcessorCache, from .cache import BaseMultiModalProcessorCache
processor_only_cache_from_config)
from .processing import BaseMultiModalProcessor, BaseProcessingInfo from .processing import BaseMultiModalProcessor, BaseProcessingInfo
from .profiling import (BaseDummyInputsBuilder, DummyDecoderData, from .profiling import (BaseDummyInputsBuilder, DummyDecoderData,
DummyEncoderData, MultiModalProfiler) DummyEncoderData, MultiModalProfiler)
@ -176,35 +175,6 @@ class MultiModalRegistry:
if mm_limits[key] > 0 if mm_limits[key] > 0
} }
# TODO: Remove once V0 is gone
def get_max_tokens_by_modality(
self,
model_config: "ModelConfig",
) -> Mapping[str, int]:
"""
Get the maximum number of tokens from each modality
for profiling the memory usage of a model.
"""
cache = processor_only_cache_from_config(model_config, self)
mm_limits = self.get_mm_limits_per_prompt(model_config, cache=cache)
max_tokens_per_item = self.get_max_tokens_per_item_by_modality(
model_config,
cache=cache,
)
return {
key: mm_limits[key] * max_tokens_per_mm_item
for key, max_tokens_per_mm_item in max_tokens_per_item.items()
}
# TODO: Remove once V0 is gone
def get_max_multimodal_tokens(self, model_config: "ModelConfig") -> int:
"""
Get the maximum number of multi-modal tokens
for profiling the memory usage of a model.
"""
return sum(self.get_max_tokens_by_modality(model_config).values())
def get_mm_limits_per_prompt( def get_mm_limits_per_prompt(
self, self,
model_config: "ModelConfig", model_config: "ModelConfig",