A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 23:15 UTC
fix incorrect type annotation in KimiMLP (#27885)
Signed-off-by: zitian.zhao <zitian.zhao@tencentmusic.com>
ZiTian Zhao committed
Nov 1, 2025 at 01:38 UTC
bc306fe5e97823e2a2e989725bd5e39a897a43a6
1 file changed
+1
-2
vllm/model_executor/models/kimi_linear.py
+1
-2
@@ -22,7 +22,6 @@ from vllm.model_executor.layers.layernorm import RMSNorm
22
from vllm.model_executor.layers.linear import (
23
ColumnParallelLinear,
24
MergedColumnParallelLinear,
25
- QKVParallelLinear,
25
ReplicatedLinear,
26
RowParallelLinear,
27
)
@@ -61,7 +60,7 @@ class KimiMLP(nn.Module):
60
hidden_size: int,
61
intermediate_size: int,
62
hidden_act: str,
64
- quant_config: QKVParallelLinear | None = None,
63
+ quant_config: QuantizationConfig | None = None,
64
reduce_results: bool = True,
65
prefix: str = "",
66
) -> None: