[Bugfix] Fix HYV3 shared_mlp prefix for compressed-tensors ignore matching (#48682)
Signed-off-by: Xiaoyi Xu <xu_xiaoyi@hotmail.com>
DCo committed
Aug 22, 2026 at 00:07 UTC
c1d7a3808d28dd6e378785852acdccaab3553c39
1 file changed
+1
-1
vllm/model_executor/models/hy_v3.py
+1
-1
@@ -167,7 +167,7 @@ class HYV3MoEFused(nn.Module):
167
intermediate_size=config.expert_hidden_dim * config.num_shared_experts,
168
hidden_act=config.hidden_act,
169
quant_config=quant_config,
170
- prefix=f"{prefix}",
170
+ prefix=f"{prefix}.shared_mlp",
171
reduce_results=False,
172
)
173
else: