diff --git a/vllm/model_executor/models/deepseek_v2.py b/vllm/model_executor/models/deepseek_v2.py index 208379403a..4290609cbd 100644 --- a/vllm/model_executor/models/deepseek_v2.py +++ b/vllm/model_executor/models/deepseek_v2.py @@ -307,9 +307,7 @@ class DeepseekV2MoE(nn.Module): self.gate = GateLinear( config.hidden_size, config.n_routed_experts, - params_dtype=self.router_dtype, out_dtype=self.router_dtype, - force_fp32_compute=self.router_dtype == torch.float32, prefix=f"{prefix}.gate", ) if getattr(config, "topk_method", None) == "noaux_tc":