K3: Move LatentMoERunner (#50678)
Signed-off-by: Jee Jee Li <jeejeelee@inferact.ai>
Jee Jee Li committed
Aug 3, 2026 at 17:36 UTC
b9d1e2437e1ad6d98d301939949072b5c56dfef1
2 files changed
+4
-5
vllm/models/kimi_k3/nvidia/latent_moe_runner.py
renamed
+1
-2
@@ -11,13 +11,12 @@ from vllm.distributed import (
11
tensor_model_parallel_all_reduce,
12
)
13
from vllm.logger import init_logger
14
+from vllm.model_executor.layers.fused_moe.runner.moe_runner import MoERunner, _unpack
15
from vllm.model_executor.layers.layernorm import RMSNorm
16
from vllm.platforms import current_platform
17
from vllm.utils.multi_stream_utils import maybe_execute_in_parallel
18
from vllm.utils.torch_utils import aux_stream
19
19
-from .moe_runner import MoERunner, _unpack
20
-
20
logger = init_logger(__name__)
21
22
vllm/models/kimi_k3/nvidia/model.py
+3
-3
@@ -30,9 +30,6 @@ from vllm.model_executor.layers.fused_moe.router.gate_linear import GateLinear
30
from vllm.model_executor.layers.fused_moe.router.grouped_topk_router import (
31
fused_grouped_topk,
32
)
33
-from vllm.model_executor.layers.fused_moe.runner.latent_moe_runner import (
34
- LatentMoERunner,
35
-)
33
from vllm.model_executor.layers.layernorm import RMSNorm
34
from vllm.model_executor.layers.linear import (
35
MergedColumnParallelLinear,
@@ -97,6 +94,9 @@ from vllm.models.common.ops.sequence_parallel import (
94
from vllm.models.deepseek_v4.nvidia.model import DeepseekV4MegaMoEExperts
95
from vllm.models.deepseek_v4.nvidia.ops.prepare_megamoe import prepare_megamoe_inputs
96
from vllm.models.kimi_k3.nvidia.kda import KimiK3DeltaAttention
97
+from vllm.models.kimi_k3.nvidia.latent_moe_runner import (
98
+ LatentMoERunner,
99
+)
100
from vllm.models.kimi_k3.nvidia.low_latency_gemm import (
101
enable_kimi_k3_low_latency_gemm,
102
)