[2/2] Deepseek deterministic: support deepseek v3 deterministic inference on 8 x H200 (#12095)

python/sglang/srt/models/deepseek_v2.py

@@ -515,6 +515,9 @@ class MoEGate(nn.Module):
                 True,  # is_vnni
             )
 
+        if get_global_server_args().enable_deterministic_inference:
+            return F.linear(hidden_states, self.weight, None)
+
         # NOTE: For some unknown reason, router_gemm seems degrade accept length.
         if (
             _is_cuda