Update internode_ll.cu (#374)

csrc/kernels/internode_ll.cu

@@ -932,7 +932,7 @@ void combine(void* combined_x,
     const int num_warp_groups = ceil_div(num_experts, num_device_sms);
     const int num_warps_per_group = 32 / num_warp_groups;
     const int num_recv_per_sm = ceil_div(num_combined_tokens, num_device_sms);
-    EP_HOST_ASSERT(num_warp_groups > 0 and num_warps_per_group > 0 and num_recv_per_sm > 0);
+    EP_HOST_ASSERT(num_warp_groups > 0 and num_warps_per_group > 0 and ((num_combined_tokens == 0) or (num_recv_per_sm > 0)));
 
     const auto num_warps = num_warp_groups * num_warps_per_group;
     const auto num_sms = max(ceil_div(num_experts, num_warp_groups), ceil_div(num_combined_tokens, num_recv_per_sm));