[Bugfix][MoE] Fix fused MoE expert mapping helper call sites (#44897)

Signed-off-by: Mohammad Miadh Angkad <176301910+mmangkad@users.noreply.github.com>
This commit is contained in:
Mohammad Miadh Angkad
2026-06-08 13:35:04 -07:00
committed by GitHub
parent 2c27c294c0
commit 823a0ab754
6 changed files with 26 additions and 11 deletions
+5 -2
View File
@@ -16,7 +16,10 @@ from vllm.distributed import (
)
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.attention import Attention
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.fused_moe import (
FusedMoE,
fused_moe_make_expert_params_mapping,
)
from vllm.model_executor.layers.linear import (
MergedColumnParallelLinear,
QKVParallelLinear,
@@ -450,7 +453,7 @@ class Cohere2MoeModel(nn.Module):
("gate_up_proj", "up_proj", 1),
]
expert_params_mapping = FusedMoE.make_expert_params_mapping(
expert_params_mapping = fused_moe_make_expert_params_mapping(
self,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",
+6 -2
View File
@@ -43,7 +43,11 @@ from vllm.distributed import (
from vllm.logger import init_logger
from vllm.model_executor.layers.activation import SiluAndMul
from vllm.model_executor.layers.attention import Attention
from vllm.model_executor.layers.fused_moe import FusedMoE, GateLinear
from vllm.model_executor.layers.fused_moe import (
FusedMoE,
GateLinear,
fused_moe_make_expert_params_mapping,
)
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.linear import (
MergedColumnParallelLinear,
@@ -474,7 +478,7 @@ class HYV3Model(nn.Module):
def get_expert_mapping(self) -> list[tuple[str, str, int, str]]:
# Params for weights, fp8 weight scales, fp8 activation scales
# (param_name, weight_name, expert_id, shard_id)
return FusedMoE.make_expert_params_mapping(
return fused_moe_make_expert_params_mapping(
self,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",
+4 -2
View File
@@ -32,7 +32,9 @@ from torch import nn
from transformers import PretrainedConfig
from vllm.config import CacheConfig, ModelConfig, VllmConfig
from vllm.model_executor.layers.fused_moe import FusedMoE
from vllm.model_executor.layers.fused_moe import (
fused_moe_make_expert_params_mapping,
)
from vllm.model_executor.layers.layernorm import RMSNorm
from vllm.model_executor.layers.logits_processor import LogitsProcessor
from vllm.model_executor.layers.quantization import QuantizationConfig
@@ -294,7 +296,7 @@ class HYV3MTP(nn.Module):
]
if _is_moe(self.config):
expert_params_mapping = FusedMoE.make_expert_params_mapping(
expert_params_mapping = fused_moe_make_expert_params_mapping(
self,
ckpt_gate_proj_name="gate_proj",
ckpt_down_proj_name="down_proj",