fix: fix a potential synchronize bug when update state

Author: windreamer

src/turbomind/layers/sampling_layers/GuidedDecodeUpdateLayer.cc

@@ -15,6 +15,7 @@
  */
 
 #include "src/turbomind/layers/sampling_layers/GuidedDecodeUpdateLayer.h"
+#include "src/turbomind/core/context.h"
 
 namespace turbomind {
 
@@ -44,6 +45,7 @@ void GuidedDecodeUpdateLayer<T>::Forward(TensorMap& args)
     Tensor_<int>  output_ids_buf{{bsz}, kCPU};
 
     FT_CHECK(bsz == matchers_.size());
+    core::Context::stream().Sync();
     Copy(output_ids.slice(step * bsz, bsz), output_ids_buf);
 
     for (size_t i = 0; i < bsz; ++i) {