[MM] Cache common token sequences (#53560)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Signed-off-by: Cyrus Leung <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-08-24 07:48:10 -07:00
committed by GitHub
parent 4ca856b0b5
commit ecfa7bb373
14 changed files with 40 additions and 23 deletions
+3 -1
View File
@@ -57,6 +57,7 @@ from vllm.multimodal.processing import (
PromptReplacement,
PromptUpdate,
PromptUpdateDetails,
cached_encode,
)
from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.configs.moondream3 import (
@@ -945,7 +946,8 @@ class Moondream3MultiModalProcessor(BaseMultiModalProcessor[Moondream3Processing
@cached_property
def bos_image_placeholder_tokens(self) -> list[int]:
tokenizer = self.info.get_tokenizer()
token_ids = tokenizer.encode(
token_ids = cached_encode(
tokenizer,
self.bos_image_placeholder,
add_special_tokens=False,
)