[MM][CG] Enable encoder CUDA Graph for MiniCPM-V (#42785)
Signed-off-by: YunzhuLu <lucia.yunzhu@gmail.com> Signed-off-by: Isotr0py <Isotr0py@outlook.com> Co-authored-by: Isotr0py <Isotr0py@outlook.com>
This commit is contained in:
@@ -6,7 +6,7 @@ Kimi-K2.5 Model Implementation for vLLM.
|
||||
Kimi-K2.5 extends Kimi-K2 with vision support.
|
||||
"""
|
||||
|
||||
from collections.abc import Iterable, Mapping, Sequence
|
||||
from collections.abc import Hashable, Iterable, Mapping, Sequence
|
||||
from dataclasses import dataclass
|
||||
from typing import TYPE_CHECKING, Annotated, Any, ClassVar, Literal
|
||||
|
||||
@@ -594,6 +594,7 @@ class KimiK25ForConditionalGeneration(
|
||||
device: torch.device,
|
||||
dtype: torch.dtype,
|
||||
path: str = "default",
|
||||
axis_keys: tuple[Hashable, ...] | None = None,
|
||||
) -> "EncoderCudaGraphCaptureInputs":
|
||||
from vllm.v1.worker.encoder_cudagraph_defs import EncoderCudaGraphCaptureInputs
|
||||
|
||||
|
||||
Reference in New Issue
Block a user