diff --git a/vllm/models/kimi_k3/amd/linear.py b/vllm/models/kimi_k3/amd/linear.py index 4e0088e9e1..181704ebf9 100644 --- a/vllm/models/kimi_k3/amd/linear.py +++ b/vllm/models/kimi_k3/amd/linear.py @@ -210,7 +210,10 @@ class KimiMoE(nn.Module): prefix=f"{prefix}.gate", ) - self.gate.e_score_correction_bias = nn.Parameter(torch.empty(num_experts)) + # Preserve FP32 checkpoint values and match FP32 router logits. + self.gate.e_score_correction_bias = nn.Parameter( + torch.empty(num_experts, dtype=torch.float32) + ) if self.num_shared_experts is not None: shared_intermediate_size = moe_intermediate_size * self.num_shared_experts