diff --git a/python/sglang/srt/server_args.py b/python/sglang/srt/server_args.py index e81e6c53b..38d5d3480 100644 --- a/python/sglang/srt/server_args.py +++ b/python/sglang/srt/server_args.py @@ -804,7 +804,7 @@ class ServerArgs: hf_config = self.get_hf_config() model_arch = hf_config.architectures[0] - if model_arch in ["DeepseekV3ForCausalLM"]: + if model_arch in ["DeepseekV3ForCausalLM"] and not is_deepseek_nsa(hf_config): if is_cuda() and is_sm100_supported(): if ( self.attention_backend is None