[Core] Use individual MM items in P0/P1 cache and model runner (#22570)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-13 07:18:07 -07:00
committed by GitHub
parent 20d65aa755
commit 19b927e52d
24 changed files with 549 additions and 486 deletions
+2 -4
View File
@@ -3,15 +3,13 @@
import enum
import time
from collections.abc import Sequence
from typing import Any, Optional, Union
import msgspec
import torch
from vllm.lora.request import LoRARequest
from vllm.multimodal import MultiModalKwargs
from vllm.multimodal.inputs import PlaceholderRange
from vllm.multimodal.inputs import MultiModalKwargsItem, PlaceholderRange
from vllm.pooling_params import PoolingParams
from vllm.sampling_params import SamplingParams
from vllm.v1.metrics.stats import SchedulerStats
@@ -49,7 +47,7 @@ class EngineCoreRequest(
request_id: str
prompt_token_ids: list[int]
mm_inputs: Optional[Sequence[Optional[MultiModalKwargs]]]
mm_kwargs: Optional[list[MultiModalKwargsItem]]
mm_hashes: Optional[list[str]]
mm_placeholders: Optional[list[PlaceholderRange]]
sampling_params: Optional[SamplingParams]