diff --git a/paddle/phi/kernels/funcs/unique_functor.h b/paddle/phi/kernels/funcs/unique_functor.h
index 758b9160096d09..fafb1b284c60a8 100644
--- a/paddle/phi/kernels/funcs/unique_functor.h
+++ b/paddle/phi/kernels/funcs/unique_functor.h
@@ -23,6 +23,35 @@
 namespace phi {
 namespace funcs {
 
+template <typename T>
+static bool NaNSafeEqual(const T& a, const T& b) {
+  if constexpr (std::is_floating_point_v<T>) {
+    if (std::isnan(a) && std::isnan(b)) {
+      return &a == &b;
+    }
+    if (std::isnan(a) || std::isnan(b)) {
+      return false;
+    }
+  }
+  return a == b;
+}
+
+template <typename T>
+static bool NaNSafeLess(const T& a, const T& b) {
+  if constexpr (std::is_floating_point_v<T>) {
+    if (std::isnan(a) && !std::isnan(b)) {
+      return false;
+    }
+    if (!std::isnan(a) && std::isnan(b)) {
+      return true;
+    }
+    if (std::isnan(a) && std::isnan(b)) {
+      return &a < &b;
+    }
+  }
+  return a < b;
+}
+
 template <typename Context, typename InT>
 struct UniqueOpFunctor {
   const Context& dev_ctx_;
@@ -122,7 +151,7 @@ static bool Equal(const DenseTensor& a, const DenseTensor& b) {
     return false;
   }
   for (int64_t i = 0; i < a.numel(); ++i) {
-    if (a.data<T>()[i] != b.data<T>()[i]) {
+    if (!NaNSafeEqual(a.data<T>()[i], b.data<T>()[i])) {
       return false;
     }
   }
@@ -140,7 +169,15 @@ static void UniqueFlattenedTensor(const Context& dev_ctx,
                                   bool return_inverse,
                                   bool return_counts) {
   const InT* in_data = in.data<InT>();
-  std::set<InT> unique(in_data, in_data + in.numel());
+
+  auto nan_safe_comp = [](const InT& a, const InT& b) {
+    return NaNSafeLess(a, b);
+  };
+  std::set<InT, decltype(nan_safe_comp)> unique(nan_safe_comp);
+  for (int64_t i = 0; i < in.numel(); ++i) {
+    unique.insert(in_data[i]);
+  }
+
   out->Resize(common::make_ddim({static_cast<int64_t>(unique.size())}));
   auto* out_data = dev_ctx.template Alloc<InT>(out);
   std::copy(unique.begin(), unique.end(), out_data);
@@ -162,29 +199,27 @@ static void UniqueFlattenedTensor(const Context& dev_ctx,
   if (return_inverse) {
     index->Resize(common::make_ddim({in.numel()}));
     auto inverse_data = dev_ctx.template Alloc<IndexT>(index);
-    std::unordered_map<InT, IndexT> inverse_map;
-    inverse_map.reserve(out->numel());
-    for (int64_t i = 0; i < out->numel(); ++i) {
-      inverse_map[out_data[i]] = i;
-    }
     for (int64_t i = 0; i < in.numel(); ++i) {
-      inverse_data[i] = inverse_map[in_data[i]];
+      for (int64_t j = 0; j < out->numel(); ++j) {
+        if (NaNSafeEqual(in_data[i], out_data[j])) {
+          inverse_data[i] = j;
+          break;
+        }
+      }
     }
   }
 
   if (return_counts) {
     count->Resize(common::make_ddim({out->numel()}));
     auto count_data = dev_ctx.template Alloc<IndexT>(count);
-    std::unordered_map<InT, IndexT> counts_map;
-    counts_map.reserve(out->numel());
     for (int64_t i = 0; i < out->numel(); ++i) {
-      counts_map[out_data[i]] = 0;
-    }
-    for (int64_t i = 0; i < in.numel(); i++) {
-      counts_map[in_data[i]] += 1;
-    }
-    for (int64_t i = 0; i < out->numel(); i++) {
-      count_data[i] = counts_map[out_data[i]];
+      IndexT cnt = 0;
+      for (int64_t j = 0; j < in.numel(); ++j) {
+        if (NaNSafeEqual(out_data[i], in_data[j])) {
+          cnt++;
+        }
+      }
+      count_data[i] = cnt;
     }
   }
 }