diff --git a/src/native/cambricon/ops/reshape_and_cache_flash/kernel.h b/src/native/cambricon/ops/reshape_and_cache_flash/kernel.h
new file mode 100644
index 000000000..44c34b2b7
--- /dev/null
+++ b/src/native/cambricon/ops/reshape_and_cache_flash/kernel.h
@@ -0,0 +1,81 @@
+#ifndef INFINI_OPS_CAMBRICON_RESHAPE_AND_CACHE_FLASH_KERNEL_H_
+#define INFINI_OPS_CAMBRICON_RESHAPE_AND_CACHE_FLASH_KERNEL_H_
+
+#include <string>
+
+#include "base/reshape_and_cache_flash.h"
+#include "dispatcher.h"
+#include "native/cambricon/common.h"
+#include "native/cambricon/data_type_.h"
+
+namespace infini::ops {
+
+template <typename T>
+void ReshapeAndCacheFlashUnion(
+    int core_per_cluster, int cluster_count, cnrtQueue_t queue, const void* key,
+    const void* value, const void* slot_mapping, void* key_cache,
+    void* value_cache, std::size_t num_tokens, std::size_t num_heads,
+    std::size_t head_size, std::size_t block_size,
+    std::ptrdiff_t key_token_stride, std::ptrdiff_t value_token_stride,
+    std::ptrdiff_t key_head_stride, std::ptrdiff_t value_head_stride,
+    std::ptrdiff_t key_cache_block_stride,
+    std::ptrdiff_t value_cache_block_stride,
+    std::ptrdiff_t key_cache_page_stride,
+    std::ptrdiff_t value_cache_page_stride,
+    std::ptrdiff_t key_cache_head_stride,
+    std::ptrdiff_t value_cache_head_stride);
+
+template <>
+class Operator<ReshapeAndCacheFlash, Device::Type::kCambricon>
+    : public ReshapeAndCacheFlash {
+ public:
+  using ReshapeAndCacheFlash::ReshapeAndCacheFlash;
+
+  Operator(const Tensor key, const Tensor value, const Tensor slot_mapping,
+           const Tensor k_scale, const Tensor v_scale,
+           const std::string kv_cache_dtype, Tensor key_cache,
+           Tensor value_cache)
+      : ReshapeAndCacheFlash{key,       value,      slot_mapping,
+                             k_scale,   v_scale,    kv_cache_dtype,
+                             key_cache, value_cache} {
+    cnrt_utils::GetLaunchConfig(key.device(), &core_per_cluster_,
+                                &cluster_count_);
+  }
+
+  void operator()(const Tensor key, const Tensor value,
+                  const Tensor slot_mapping, const Tensor /*k_scale*/,
+                  const Tensor /*v_scale*/,
+                  const std::string /*kv_cache_dtype*/, Tensor key_cache,
+                  Tensor value_cache) const override {
+    if (num_tokens_ == 0) {
+      return;
+    }
+
+    auto queue = static_cast<cnrtQueue_t>(stream_ ? stream_ : 0);
+    DispatchFunc<
+        Device::Type::kCambricon,
+        List<DataType::kFloat16, DataType::kBFloat16, DataType::kFloat32>>(
+        {dtype_},
+        [&](auto dtype_tag) {
+          using T = typename decltype(dtype_tag)::type;
+          ReshapeAndCacheFlashUnion<T>(
+              core_per_cluster_, cluster_count_, queue, key.data(),
+              value.data(), slot_mapping.data(), key_cache.data(),
+              value_cache.data(), num_tokens_, num_heads_, head_size_,
+              block_size_, key_token_stride_, value_token_stride_,
+              key_head_stride_, value_head_stride_, key_cache_block_stride_,
+              value_cache_block_stride_, key_cache_page_stride_,
+              value_cache_page_stride_, key_cache_head_stride_,
+              value_cache_head_stride_);
+        },
+        "CambriconReshapeAndCacheFlash::operator()");
+  }
+
+ private:
+  int core_per_cluster_{0};
+  int cluster_count_{0};
+};
+
+}  // namespace infini::ops
+
+#endif
diff --git a/src/native/cambricon/ops/reshape_and_cache_flash/kernel.mlu b/src/native/cambricon/ops/reshape_and_cache_flash/kernel.mlu
new file mode 100644
index 000000000..5def4e0b2
--- /dev/null
+++ b/src/native/cambricon/ops/reshape_and_cache_flash/kernel.mlu
@@ -0,0 +1,97 @@
+#include "kernel.h"
+
+namespace infini::ops {
+namespace reshape_and_cache_flash_detail {
+
+template <typename T>
+__mlu_global__ void ReshapeAndCacheFlashKernel(
+    const T* key, const T* value, const int64_t* slot_mapping, T* key_cache,
+    T* value_cache, std::size_t num_tokens, std::size_t num_heads,
+    std::size_t head_size, std::size_t block_size,
+    std::ptrdiff_t key_token_stride, std::ptrdiff_t value_token_stride,
+    std::ptrdiff_t key_head_stride, std::ptrdiff_t value_head_stride,
+    std::ptrdiff_t key_cache_block_stride,
+    std::ptrdiff_t value_cache_block_stride,
+    std::ptrdiff_t key_cache_page_stride,
+    std::ptrdiff_t value_cache_page_stride,
+    std::ptrdiff_t key_cache_head_stride,
+    std::ptrdiff_t value_cache_head_stride) {
+  const std::size_t task_count = num_tokens * num_heads;
+  for (std::size_t task = taskId; task < task_count; task += taskDim) {
+    const std::size_t token_idx = task / num_heads;
+    const std::size_t head_idx = task % num_heads;
+    const int64_t slot = slot_mapping[token_idx];
+    if (slot < 0) {
+      continue;
+    }
+
+    const auto physical_block = static_cast<std::size_t>(slot) / block_size;
+    const auto block_offset = static_cast<std::size_t>(slot) % block_size;
+    const T* key_src =
+        key + token_idx * key_token_stride + head_idx * key_head_stride;
+    const T* value_src =
+        value + token_idx * value_token_stride + head_idx * value_head_stride;
+    T* key_dst = key_cache + physical_block * key_cache_block_stride +
+                 block_offset * key_cache_page_stride +
+                 head_idx * key_cache_head_stride;
+    T* value_dst = value_cache + physical_block * value_cache_block_stride +
+                   block_offset * value_cache_page_stride +
+                   head_idx * value_cache_head_stride;
+
+    for (std::size_t i = 0; i < head_size; ++i) {
+      key_dst[i] = key_src[i];
+      value_dst[i] = value_src[i];
+    }
+  }
+}
+
+}  // namespace reshape_and_cache_flash_detail
+
+template <typename T>
+void ReshapeAndCacheFlashUnion(
+    int core_per_cluster, int cluster_count, cnrtQueue_t queue, const void* key,
+    const void* value, const void* slot_mapping, void* key_cache,
+    void* value_cache, std::size_t num_tokens, std::size_t num_heads,
+    std::size_t head_size, std::size_t block_size,
+    std::ptrdiff_t key_token_stride, std::ptrdiff_t value_token_stride,
+    std::ptrdiff_t key_head_stride, std::ptrdiff_t value_head_stride,
+    std::ptrdiff_t key_cache_block_stride,
+    std::ptrdiff_t value_cache_block_stride,
+    std::ptrdiff_t key_cache_page_stride,
+    std::ptrdiff_t value_cache_page_stride,
+    std::ptrdiff_t key_cache_head_stride,
+    std::ptrdiff_t value_cache_head_stride) {
+  cnrtDim3_t kernel_dim;
+  kernel_dim.x = core_per_cluster;
+  kernel_dim.y = cluster_count;
+  kernel_dim.z = 1;
+
+  (void)cnrtGetLastError();
+  reshape_and_cache_flash_detail::ReshapeAndCacheFlashKernel<T>
+      <<<kernel_dim, cnrtFuncTypeUnion1, queue>>>(
+          reinterpret_cast<const T*>(key), reinterpret_cast<const T*>(value),
+          reinterpret_cast<const int64_t*>(slot_mapping),
+          reinterpret_cast<T*>(key_cache), reinterpret_cast<T*>(value_cache),
+          num_tokens, num_heads, head_size, block_size, key_token_stride,
+          value_token_stride, key_head_stride, value_head_stride,
+          key_cache_block_stride, value_cache_block_stride,
+          key_cache_page_stride, value_cache_page_stride, key_cache_head_stride,
+          value_cache_head_stride);
+  CNRT_CHECK(cnrtGetLastError());
+}
+
+#define INFINI_INSTANTIATE_RESHAPE_AND_CACHE_FLASH(T)                      \
+  template void ReshapeAndCacheFlashUnion<T>(                              \
+      int, int, cnrtQueue_t, const void*, const void*, const void*, void*, \
+      void*, std::size_t, std::size_t, std::size_t, std::size_t,           \
+      std::ptrdiff_t, std::ptrdiff_t, std::ptrdiff_t, std::ptrdiff_t,      \
+      std::ptrdiff_t, std::ptrdiff_t, std::ptrdiff_t, std::ptrdiff_t,      \
+      std::ptrdiff_t, std::ptrdiff_t)
+
+INFINI_INSTANTIATE_RESHAPE_AND_CACHE_FLASH(__half);
+INFINI_INSTANTIATE_RESHAPE_AND_CACHE_FLASH(__bang_bfloat16);
+INFINI_INSTANTIATE_RESHAPE_AND_CACHE_FLASH(float);
+
+#undef INFINI_INSTANTIATE_RESHAPE_AND_CACHE_FLASH
+
+}  // namespace infini::ops