[perf]Add fused Kimi image preprocessing (#47416)

Signed-off-by: Kevin-XiongC <kevin_xiong1997@outlook.com>
Signed-off-by: Kevin_Xiong <kevin_xiong1997@outlook.com>
Signed-off-by: Isotr0py <Isotr0py@outlook.com>
Co-authored-by: Codex <codex@openai.com>
Co-authored-by: Isotr0py <2037008807@qq.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
Co-authored-by: Isotr0py <mozf@mail2.sysu.edu.cn>
Co-authored-by: Isotr0py <Isotr0py@outlook.com>
This commit is contained in:
Kevin_Xiong
2026-07-06 08:46:32 -07:00
committed by GitHub
co-authored by Codex Isotr0py mergify[bot] <37929162+mergify[bot]@users.noreply.github.com> Isotr0py Isotr0py
parent f70caef48b
commit 5ad11172b7
5 changed files with 402 additions and 2 deletions
+10
View File
@@ -56,6 +56,10 @@ from vllm.sequence import IntermediateTensors
from vllm.transformers_utils.configs.kimi_k25 import KimiK25Config
from vllm.transformers_utils.processor import cached_get_image_processor
from vllm.transformers_utils.processors.kimi_k25 import KimiK25Processor
from vllm.transformers_utils.processors.kimi_k25_vision_fused import (
KimiK25FusedVisionProcessor,
)
from vllm.utils.import_utils import is_numba_available
from vllm.utils.tensor_schema import TensorSchema, TensorShape
from .utils import (
@@ -108,10 +112,16 @@ class KimiK25ProcessingInfo(BaseProcessingInfo):
self.hf_config = hf_config = self.get_hf_config()
tokenizer = self.get_tokenizer()
processor_cls = KimiK25FusedVisionProcessor if is_numba_available() else None
logger.info_once(
"Using %s image preprocessing for Kimi-K2.5/K2.6 vision chunks.",
"fused CPU" if processor_cls is not None else "remote HF",
)
image_processor = cached_get_image_processor(
self.ctx.model_config.model,
revision=self.ctx.model_config.revision,
trust_remote_code=self.ctx.model_config.trust_remote_code,
processor_cls_overrides=processor_cls,
)
# Resolve token ID from the tokenizer because transformers v5