ZIBEE/bi_150-vllm
1
0
Fork 0
Code Issues Pull Requests Actions Projects Releases Wiki Activity

Upgrade to vllm 0.17.0 corex v4.1 overlay

Browse Source
This commit is contained in:
LZiBee
2026-04-29 19:38:22 +08:00
parent 8fac6062e4
commit 938d0854a5
430 changed files with 35969 additions and 14511 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/model_executor/layers/fused_moe/fallback.py
View File

@@ -11,13 +11,13 @@ from vllm.model_executor.layers.fused_moe.config import FusedMoEParallelConfig
from vllm.model_executor.layers.quantization.utils.quant_utils import QuantKey
class FallbackExperts(mk.FusedMoEPermuteExpertsUnpermute, ABC):
class FallbackExperts(mk.FusedMoEExpertsModular, ABC):
"""Base class for runtime dispatching of expert implementations."""
def __init__(
self,
experts: mk.FusedMoEPermuteExpertsUnpermute,
fallback_experts: mk.FusedMoEPermuteExpertsUnpermute,
experts: mk.FusedMoEExpertsModular,
fallback_experts: mk.FusedMoEExpertsModular,
):
super().__init__(
moe_config=experts.moe_config, quant_config=experts.quant_config
@@ -27,8 +27,8 @@ class FallbackExperts(mk.FusedMoEPermuteExpertsUnpermute, ABC):
@staticmethod
def get_clses() -> tuple[
type[mk.FusedMoEPermuteExpertsUnpermute],
type[mk.FusedMoEPermuteExpertsUnpermute],
type[mk.FusedMoEExpertsModular],
type[mk.FusedMoEExpertsModular],
]:
"""
Get the cls for the experts and fallback experts.
@@ -149,7 +149,7 @@ class FallbackExperts(mk.FusedMoEPermuteExpertsUnpermute, ABC):
hidden_states: torch.Tensor,
w1: torch.Tensor,
w2: torch.Tensor,
) -> mk.FusedMoEPermuteExpertsUnpermute:
) -> mk.FusedMoEExpertsModular:
raise NotImplementedError
def apply(