diff --git a/neuralnetworks/1.0/vts/functional/GeneratedTestHarness.cpp b/neuralnetworks/1.0/vts/functional/GeneratedTestHarness.cpp
index f5cb0d7..106f332 100644
--- a/neuralnetworks/1.0/vts/functional/GeneratedTestHarness.cpp
+++ b/neuralnetworks/1.0/vts/functional/GeneratedTestHarness.cpp
@@ -52,6 +52,7 @@
 using ::test_helper::MixedTyped;
 using ::test_helper::MixedTypedExample;
 using ::test_helper::resize_accordingly;
+using HidlToken = hidl_array<uint8_t, static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN)>;
 
 template <typename T>
 void copy_back_(std::map<int, std::vector<T>>* dst, const std::vector<RequestArgument>& ra,
@@ -540,7 +541,8 @@
     sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
     ASSERT_NE(nullptr, preparedModelCallback.get());
     Return<ErrorStatus> prepareLaunchStatus = device->prepareModel_1_2(
-        model, ExecutionPreference::FAST_SINGLE_ANSWER, preparedModelCallback);
+            model, ExecutionPreference::FAST_SINGLE_ANSWER, hidl_vec<hidl_handle>(),
+            hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
     ASSERT_TRUE(prepareLaunchStatus.isOk());
     ASSERT_EQ(ErrorStatus::NONE, static_cast<ErrorStatus>(prepareLaunchStatus));
 
diff --git a/neuralnetworks/1.2/IDevice.hal b/neuralnetworks/1.2/IDevice.hal
index 43e2c56..d83f9e6 100644
--- a/neuralnetworks/1.2/IDevice.hal
+++ b/neuralnetworks/1.2/IDevice.hal
@@ -124,44 +124,83 @@
             generates (ErrorStatus status, vec<bool> supportedOperations);
 
     /**
-     * Gets whether the driver supports compilation caching.
+     * Gets the caching requirements of the driver implementation.
      *
-     * isCachingSupported indicates whether the driver supports compilation caching.
-     * Even if so, the driver may still choose not to cache certain compiled models.
+     * There are two types of cache file descriptors provided to the driver: model cache
+     * and data cache.
      *
-     * If the device reports the caching is not supported, the user may avoid calling
-     * IDevice::prepareModelFromCache and IPreparedModel::saveToCache.
+     * The data cache is for caching constant data, possibly including preprocessed
+     * and transformed tensor buffers. Any modification to the data cache should
+     * have no worse effect than generating bad output values at execution time.
+     *
+     * The model cache is for caching security-sensitive data such as compiled
+     * executable machine code in the device's native binary format. A modification
+     * to the model cache may affect the driver's execution behavior, and a malicious
+     * client could make use of this to execute beyond the granted permission. Thus,
+     * the driver must always check whether the model cache is corrupted before
+     * preparing the model from cache.
+     *
+     * getNumberOfCacheFilesNeeded returns how many of each type of cache files the driver
+     * implementation needs to cache a single prepared model. Returning 0 for both types
+     * indicates compilation caching is not supported by this driver. The driver may
+     * still choose not to cache certain compiled models even if it reports that caching
+     * is supported.
+     *
+     * If the device reports that caching is not supported, the user may avoid calling
+     * IDevice::prepareModelFromCache or providing cache file descriptors to
+     * IDevice::prepareModel_1_2.
      *
      * @return status Error status of the call, must be:
      *     - NONE if successful
      *     - DEVICE_UNAVAILABLE if driver is offline or busy
      *     - GENERAL_FAILURE if there is an unspecified error
-     * @return supported A boolean indicating whether the driver supports compilation
-     *                   caching. Even on returning true, the driver may still choose
-     *                   not to cache certain compiled models.
+     * @return numModelCache An unsigned integer indicating how many files for model cache
+     *                       the driver needs to cache a single prepared model. It must
+     *                       be less than or equal to Constant::MAX_NUMBER_OF_CACHE_FILES.
+     * @return numDataCache An unsigned integer indicating how many files for data cache
+     *                      the driver needs to cache a single prepared model. It must
+     *                      be less than or equal to Constant::MAX_NUMBER_OF_CACHE_FILES.
      */
-    isCachingSupported() generates (ErrorStatus status, bool supported);
+    getNumberOfCacheFilesNeeded()
+            generates (ErrorStatus status, uint32_t numModelCache, uint32_t numDataCache);
 
     /**
-     * Creates a prepared model for execution.
+     * Asynchronously creates a prepared model for execution and optionally saves it
+     * into cache files.
      *
-     * prepareModel is used to make any necessary transformations or alternative
+     * prepareModel is used to make any necessary transformations to or alternative
      * representations to a model for execution, possibly including
      * transformations on the constant data, optimization on the model's graph,
      * or compilation into the device's native binary format. The model itself
      * is not changed.
      *
+     * Optionally, caching information may be provided for the driver to save
+     * the prepared model to cache files for faster model compilation time
+     * when the same model preparation is requested in the future. There are
+     * two types of cache file handles provided to the driver: model cache
+     * and data cache. For more information on the two types of cache handles,
+     * refer to getNumberOfCacheFilesNeeded.
+     *
+     * The file descriptors must be opened with read and write permission. A file may
+     * have any size, and the corresponding file descriptor may have any offset. The
+     * driver must truncate a file to zero size before writing to that file. The file
+     * descriptors may be closed by the client once the asynchronous preparation has
+     * finished. The driver must dup a file descriptor if it wants to get access to
+     * the cache file later.
+     *
      * The model is prepared asynchronously with respect to the caller. The
-     * prepareModel function must verify the inputs to the prepareModel function
-     * are correct. If there is an error, prepareModel must immediately invoke
+     * prepareModel function must verify the inputs to the preparedModel function
+     * related to preparing the model (as opposed to saving the prepared model to
+     * cache) are correct. If there is an error, prepareModel must immediately invoke
      * the callback with the appropriate ErrorStatus value and nullptr for the
-     * IPreparedModel, then return with the same ErrorStatus. If the inputs to
-     * the prepareModel function are valid and there is no error, prepareModel
-     * must launch an asynchronous task to prepare the model in the background,
-     * and immediately return from prepareModel with ErrorStatus::NONE. If the
-     * asynchronous task fails to launch, prepareModel must immediately invoke
-     * the callback with ErrorStatus::GENERAL_FAILURE and nullptr for the
-     * IPreparedModel, then return with ErrorStatus::GENERAL_FAILURE.
+     * IPreparedModel, then return with the same ErrorStatus. If the inputs to the
+     * prepareModel function that are related to preparing the model are valid and
+     * there is no error, prepareModel must launch an asynchronous task
+     * to prepare the model in the background, and immediately return from
+     * prepareModel with ErrorStatus::NONE. If the asynchronous task fails to launch,
+     * prepareModel must immediately invoke the callback with
+     * ErrorStatus::GENERAL_FAILURE and nullptr for the IPreparedModel, then return
+     * with ErrorStatus::GENERAL_FAILURE.
      *
      * When the asynchronous task has finished preparing the model, it must
      * immediately invoke the callback function provided as an input to
@@ -171,6 +210,14 @@
      * the callback object must be invoked with the appropriate ErrorStatus
      * value and nullptr for the IPreparedModel.
      *
+     * Optionally, the driver may save the prepared model to cache during the
+     * asynchronous preparation. Any error that occurs when saving to cache must
+     * not affect the status of preparing the model. Even if the input arguments
+     * related to the cache may be invalid, or the driver may fail to save to cache,
+     * the prepareModel function must finish preparing the model. The driver
+     * may choose not to save to cache even if the caching information is
+     * provided and valid.
+     *
      * The only information that may be unknown to the model at this stage is
      * the shape of the tensors, which may only be known at execution time. As
      * such, some driver services may return partially prepared models, where
@@ -184,6 +231,26 @@
      * @param model The model to be prepared for execution.
      * @param preference Indicates the intended execution behavior of a prepared
      *     model.
+     * @param modelCache A vector of handles with each entry holding exactly one
+     *     cache file descriptor for the security-sensitive cache. The length of
+     *     the vector must either be 0 indicating that caching information is not provided,
+     *     or match the numModelCache returned from getNumberOfCacheFilesNeeded. The cache
+     *     handles will be provided in the same order when retrieving the
+     *     preparedModel from cache files with prepareModelFromCache.
+     * @param dataCache A vector of handles with each entry holding exactly one
+     *     cache file descriptor for the constants' cache. The length of
+     *     the vector must either be 0 indicating that caching information is not provided,
+     *     or match the numDataCache returned from getNumberOfCacheFilesNeeded. The cache
+     *     handles will be provided in the same order when retrieving the
+     *     preparedModel from cache files with prepareModelFromCache.
+     * @param token A caching token of length Constant::BYTE_SIZE_OF_CACHE_TOKEN
+     *     identifying the prepared model. The same token will be provided when retrieving
+     *     the prepared model from the cache files with prepareModelFromCache.
+     *     Tokens should be chosen to have a low rate of collision for a particular
+     *     application. The driver cannot detect a collision; a collision will result
+     *     in a failed execution or in a successful execution that produces incorrect
+     *     output values. If both modelCache and dataCache are empty indicating that
+     *     caching information is not provided, this token must be ignored.
      * @param callback A callback object used to return the error status of
      *     preparing the model for execution and the prepared model if
      *     successful, nullptr otherwise. The callback object's notify function
@@ -193,9 +260,12 @@
      *     - NONE if preparation task is successfully launched
      *     - DEVICE_UNAVAILABLE if driver is offline or busy
      *     - GENERAL_FAILURE if there is an unspecified error
-     *     - INVALID_ARGUMENT if one of the input arguments is invalid
+     *     - INVALID_ARGUMENT if one of the input arguments related to preparing the
+     *       model is invalid
      */
     prepareModel_1_2(Model model, ExecutionPreference preference,
+                     vec<handle> modelCache, vec<handle> dataCache,
+                     uint8_t[Constant:BYTE_SIZE_OF_CACHE_TOKEN] token,
                      IPreparedModelCallback callback)
           generates (ErrorStatus status);
 
@@ -203,22 +273,17 @@
      * Creates a prepared model from cache files for execution.
      *
      * prepareModelFromCache is used to retrieve a prepared model directly from
-     * cache files to avoid slow model compilation time. There are exactly two
-     * cache file descriptors provided to the driver: modelCache and dataCache.
+     * cache files to avoid slow model compilation time. There are
+     * two types of cache file handles provided to the driver: model cache
+     * and data cache. For more information on the two types of cache handles,
+     * refer to getNumberOfCacheFilesNeeded.
      *
-     * The dataCache is for caching constant data, possibly including preprocessed
-     * and transformed tensor buffers. Any modification to the dataCache should
-     * have no worse effect than generating bad output values at execution time.
-     *
-     * The modelCache is for caching security-sensitive data such as compiled
-     * executable machine code in the device's native binary format. A modification
-     * to the modelCache may affect the driver's execution behavior, and a malicious
-     * client could make use of this to execute beyond the granted permission. Thus,
-     * the driver must always check whether the modelCache is corrupted before preparing
-     * the model from cache.
-     *
-     * The two file descriptors may be closed by the client once the asynchronous
-     * preparation has finished. The driver has to copy all the data it needs.
+     * The file descriptors must be opened with read and write permission. A file may
+     * have any size, and the corresponding file descriptor may have any offset. The
+     * driver must truncate a file to zero size before writing to that file. The file
+     * descriptors may be closed by the client once the asynchronous preparation has
+     * finished. The driver must dup a file descriptor if it wants to get access to
+     * the cache file later.
      *
      * The model is prepared asynchronously with respect to the caller. The
      * prepareModelFromCache function must verify the inputs to the
@@ -252,13 +317,17 @@
      * used with different shapes of inputs on different (possibly concurrent)
      * executions.
      *
-     * @param modelCache A handle holding exactly one cache file descriptor for the
-     *     security-sensitive cache.
-     * @param dataCache A handle holding exactly one cache file descriptor for the
-     *     constants' cache.
+     * @param modelCache A vector of handles with each entry holding exactly one
+     *     cache file descriptor for the security-sensitive cache. The length of
+     *     the vector must match the numModelCache returned from getNumberOfCacheFilesNeeded.
+     *     The cache handles will be provided in the same order as with prepareModel_1_2.
+     * @param dataCache A vector of handles with each entry holding exactly one
+     *     cache file descriptor for the constants' cache. The length of the vector
+     *     must match the numDataCache returned from getNumberOfCacheFilesNeeded.
+     *     The cache handles will be provided in the same order as with prepareModel_1_2.
      * @param token A caching token of length Constant::BYTE_SIZE_OF_CACHE_TOKEN
      *     identifying the prepared model. It is the same token provided when saving
-     *     the cache files with IPreparedModel::saveToCache. Tokens should be chosen
+     *     the cache files with prepareModel_1_2. Tokens should be chosen
      *     to have a low rate of collision for a particular application. The driver
      *     cannot detect a collision; a collision will result in a failed execution
      *     or in a successful execution that produces incorrect output values.
@@ -274,7 +343,7 @@
      *       unspecified error
      *     - INVALID_ARGUMENT if one of the input arguments is invalid
      */
-    prepareModelFromCache(handle modelCache, handle dataCache,
+    prepareModelFromCache(vec<handle> modelCache, vec<handle> dataCache,
                           uint8_t[Constant:BYTE_SIZE_OF_CACHE_TOKEN] token,
                           IPreparedModelCallback callback)
             generates (ErrorStatus status);
diff --git a/neuralnetworks/1.2/IPreparedModel.hal b/neuralnetworks/1.2/IPreparedModel.hal
index 757d5f1..5d2d80f 100644
--- a/neuralnetworks/1.2/IPreparedModel.hal
+++ b/neuralnetworks/1.2/IPreparedModel.hal
@@ -157,62 +157,4 @@
                             fmq_sync<FmqRequestDatum> requestChannel,
                             fmq_sync<FmqResultDatum> resultChannel)
                  generates (ErrorStatus status, IBurstContext context);
-
-    /*
-     * Saves the prepared model to cache files.
-     *
-     * saveToCache is used to save a prepared model to cache files for faster
-     * model compilation time when the same model preparation is requested in
-     * the future. There are exactly two cache file descriptors provided to the
-     * driver: modelCache and dataCache.
-     *
-     * The dataCache is for caching constant data, possibly including preprocessed
-     * and transformed tensor buffers. Any modification to the dataCache should
-     * have no worse effect than generating bad output values at execution time.
-     *
-     * The modelCache is for caching security-sensitive data such as compiled
-     * executable machine code in the device's native binary format. A modification
-     * to the modelCache may affect the driver's execution behavior, and a malicious
-     * client could make use of this to execute beyond the granted permission. Thus,
-     * the driver must always check whether the modelCache is corrupted before preparing
-     * the model from cache.
-     *
-     * The two file descriptors must point to two zero-length files with offset
-     * positioned at the beginning of the file. The file descriptors may be closed
-     * by the client once the method has returned.
-     *
-     * If the driver decides not to save the prepared model without looking at the
-     * input arguments to the saveToCache function, saveToCache must return with
-     * ErrorStatus::GENERAL_FAILURE. Otherwise, the saveToCache function must verify
-     * the input arguments to the saveToCache function are valid, and return with
-     * ErrorStatus::INVALID_ARGUMENT if not. If the inputs are valid but the driver
-     * could not save the prepared model, saveToCache must return with the appropriate
-     * ErrorStatus. Otherwise, it must write the cache files and return
-     * ErrorStatus::NONE. Unless saveToCache returns ErrorStatus::NONE, the contents
-     * of the cache files are undefined.
-     *
-     * @param modelCache A handle holding exactly one cache file descriptor for the
-     *                   security-sensitive cache.
-     * @param dataCache A handle holding exactly one cache file descriptor for the
-     *                  constants' cache.
-     * @param token A caching token of length Constant::BYTE_SIZE_OF_CACHE_TOKEN
-     *              identifying the prepared model. The same token will be provided
-     *              when retrieving the prepared model from cache files with
-     *              IDevice::prepareModelFromCache. Tokens should be chosen to have
-     *              a low rate of collision for a particular application. The driver
-     *              cannot detect a collision; a collision will result in a failed
-     *              execution or in a successful execution that produces incorrect
-     *              output values.
-     * @return status Error status of saveToCache, must be:
-     *                - NONE if saveToCache is performed successfully
-     *                - DEVICE_UNAVAILABLE if driver is offline or busy
-     *                - GENERAL_FAILURE if the driver could not save the
-     *                  prepared model or if there is an unspecified error
-     *                - INVALID_ARGUMENT if one of the input arguments is invalid,
-     *                  unless the driver decides not to save the prepared model
-     *                  without looking at the input arguments
-     */
-    saveToCache(handle modelCache, handle dataCache,
-                uint8_t[Constant:BYTE_SIZE_OF_CACHE_TOKEN] token)
-        generates (ErrorStatus status);
 };
diff --git a/neuralnetworks/1.2/types.hal b/neuralnetworks/1.2/types.hal
index 5f69f91..7691642 100644
--- a/neuralnetworks/1.2/types.hal
+++ b/neuralnetworks/1.2/types.hal
@@ -30,6 +30,11 @@
      * The byte size of the cache token.
      */
     BYTE_SIZE_OF_CACHE_TOKEN = 32,
+
+    /**
+     * The maximum number of files for each type of cache in compilation caching.
+     */
+    MAX_NUMBER_OF_CACHE_FILES = 32,
 };
 
 enum OperandType : @1.0::OperandType {
diff --git a/neuralnetworks/1.2/vts/functional/BasicTests.cpp b/neuralnetworks/1.2/vts/functional/BasicTests.cpp
index 61aa571..5c269df 100644
--- a/neuralnetworks/1.2/vts/functional/BasicTests.cpp
+++ b/neuralnetworks/1.2/vts/functional/BasicTests.cpp
@@ -102,10 +102,15 @@
     EXPECT_TRUE(ret.isOk());
 }
 
-// isCachingSupported test
-TEST_F(NeuralnetworksHidlTest, IsCachingSupported) {
-    Return<void> ret = device->isCachingSupported(
-            [](ErrorStatus status, bool) { EXPECT_EQ(ErrorStatus::NONE, status); });
+// getNumberOfCacheFilesNeeded test
+TEST_F(NeuralnetworksHidlTest, getNumberOfCacheFilesNeeded) {
+    Return<void> ret = device->getNumberOfCacheFilesNeeded(
+            [](ErrorStatus status, uint32_t numModelCache, uint32_t numDataCache) {
+                EXPECT_EQ(ErrorStatus::NONE, status);
+                EXPECT_LE(numModelCache,
+                          static_cast<uint32_t>(Constant::MAX_NUMBER_OF_CACHE_FILES));
+                EXPECT_LE(numDataCache, static_cast<uint32_t>(Constant::MAX_NUMBER_OF_CACHE_FILES));
+            });
     EXPECT_TRUE(ret.isOk());
 }
 }  // namespace functional
diff --git a/neuralnetworks/1.2/vts/functional/CompilationCachingTests.cpp b/neuralnetworks/1.2/vts/functional/CompilationCachingTests.cpp
index 00989e5..167fc09 100644
--- a/neuralnetworks/1.2/vts/functional/CompilationCachingTests.cpp
+++ b/neuralnetworks/1.2/vts/functional/CompilationCachingTests.cpp
@@ -54,29 +54,39 @@
 [[maybe_unused]] auto dummy_createTestModel = createTestModel_dynamic_output_shape;
 [[maybe_unused]] auto dummy_get_examples = get_examples_dynamic_output_shape;
 
-enum class AccessMode { READ_ONLY, WRITE_ONLY };
+enum class AccessMode { READ_WRITE, READ_ONLY, WRITE_ONLY };
 
-void createCacheHandle(const std::vector<std::string>& files, AccessMode mode,
-                       hidl_handle* handle) {
-    std::vector<int> fds;
-    for (const auto& file : files) {
-        int fd;
-        if (mode == AccessMode::READ_ONLY) {
-            fd = open(file.c_str(), O_RDONLY);
-        } else if (mode == AccessMode::WRITE_ONLY) {
-            fd = open(file.c_str(), O_WRONLY | O_TRUNC | O_CREAT, S_IRUSR | S_IWUSR);
-        } else {
-            FAIL();
+// Creates cache handles based on provided file groups.
+// The outer vector corresponds to handles and the inner vector is for fds held by each handle.
+void createCacheHandles(const std::vector<std::vector<std::string>>& fileGroups,
+                        const std::vector<AccessMode>& mode, hidl_vec<hidl_handle>* handles) {
+    handles->resize(fileGroups.size());
+    for (uint32_t i = 0; i < fileGroups.size(); i++) {
+        std::vector<int> fds;
+        for (const auto& file : fileGroups[i]) {
+            int fd;
+            if (mode[i] == AccessMode::READ_ONLY) {
+                fd = open(file.c_str(), O_RDONLY);
+            } else if (mode[i] == AccessMode::WRITE_ONLY) {
+                fd = open(file.c_str(), O_WRONLY | O_CREAT, S_IRUSR | S_IWUSR);
+            } else if (mode[i] == AccessMode::READ_WRITE) {
+                fd = open(file.c_str(), O_RDWR | O_CREAT, S_IRUSR | S_IWUSR);
+            } else {
+                FAIL();
+            }
+            ASSERT_GE(fd, 0);
+            fds.push_back(fd);
         }
-        ASSERT_GE(fd, 0);
-        fds.push_back(fd);
+        native_handle_t* cacheNativeHandle = native_handle_create(fds.size(), 0);
+        ASSERT_NE(cacheNativeHandle, nullptr);
+        std::copy(fds.begin(), fds.end(), &cacheNativeHandle->data[0]);
+        (*handles)[i].setTo(cacheNativeHandle, /*shouldOwn=*/true);
     }
-    native_handle_t* cacheNativeHandle = native_handle_create(fds.size(), 0);
-    ASSERT_NE(cacheNativeHandle, nullptr);
-    for (uint32_t i = 0; i < fds.size(); i++) {
-        cacheNativeHandle->data[i] = fds[i];
-    }
-    handle->setTo(cacheNativeHandle, /*shouldOwn=*/true);
+}
+
+void createCacheHandles(const std::vector<std::vector<std::string>>& fileGroups, AccessMode mode,
+                        hidl_vec<hidl_handle>* handles) {
+    createCacheHandles(fileGroups, std::vector<AccessMode>(fileGroups.size(), mode), handles);
 }
 
 }  // namespace
@@ -88,38 +98,43 @@
         NeuralnetworksHidlTest::SetUp();
         ASSERT_NE(device.get(), nullptr);
 
-        // Create cache directory. The cache directory and cache files are always created to test
-        // the behavior of prepareModelFromCache, even when caching is not supported.
+        // Create cache directory. The cache directory and a temporary cache file is always created
+        // to test the behavior of prepareModelFromCache, even when caching is not supported.
         char cacheDirTemp[] = "/data/local/tmp/TestCompilationCachingXXXXXX";
         char* cacheDir = mkdtemp(cacheDirTemp);
         ASSERT_NE(cacheDir, nullptr);
         mCacheDir = cacheDir;
+        mCacheDir.push_back('/');
 
-        // Create empty cache files.
-        mCache1 = mCacheDir + "/cache1";
-        mCache2 = mCacheDir + "/cache2";
-        mCache3 = mCacheDir + "/cache3";
-        // A dummy handle, use AccessMode::WRITE_ONLY for createCacheHandle to create files.
-        hidl_handle handle;
-        createCacheHandle({mCache1, mCache2, mCache3}, AccessMode::WRITE_ONLY, &handle);
-
-        // Check if caching is supported.
-        bool isCachingSupported;
-        Return<void> ret = device->isCachingSupported(
-                [&isCachingSupported](ErrorStatus status, bool supported) {
+        Return<void> ret = device->getNumberOfCacheFilesNeeded(
+                [this](ErrorStatus status, uint32_t numModelCache, uint32_t numDataCache) {
                     EXPECT_EQ(ErrorStatus::NONE, status);
-                    isCachingSupported = supported;
+                    mNumModelCache = numModelCache;
+                    mNumDataCache = numDataCache;
                 });
         EXPECT_TRUE(ret.isOk());
-        if (isCachingSupported) {
-            mIsCachingSupported = true;
-        } else {
+        mIsCachingSupported = mNumModelCache > 0 || mNumDataCache > 0;
+
+        // Create empty cache files.
+        mTmpCache = mCacheDir + "tmp";
+        for (uint32_t i = 0; i < mNumModelCache; i++) {
+            mModelCache.push_back({mCacheDir + "model" + std::to_string(i)});
+        }
+        for (uint32_t i = 0; i < mNumDataCache; i++) {
+            mDataCache.push_back({mCacheDir + "data" + std::to_string(i)});
+        }
+        // Dummy handles, use AccessMode::WRITE_ONLY for createCacheHandles to create files.
+        hidl_vec<hidl_handle> modelHandle, dataHandle, tmpHandle;
+        createCacheHandles(mModelCache, AccessMode::WRITE_ONLY, &modelHandle);
+        createCacheHandles(mDataCache, AccessMode::WRITE_ONLY, &dataHandle);
+        createCacheHandles({{mTmpCache}}, AccessMode::WRITE_ONLY, &tmpHandle);
+
+        if (!mIsCachingSupported) {
             LOG(INFO) << "NN VTS: Early termination of test because vendor service does not "
                          "support compilation caching.";
             std::cout << "[          ]   Early termination of test because vendor service does not "
                          "support compilation caching."
                       << std::endl;
-            mIsCachingSupported = false;
         }
     }
 
@@ -127,22 +142,49 @@
         // The tmp directory is only removed when the driver reports caching not supported,
         // otherwise it is kept for debugging purpose.
         if (!mIsCachingSupported) {
-            remove(mCache1.c_str());
-            remove(mCache2.c_str());
-            remove(mCache3.c_str());
+            remove(mTmpCache.c_str());
             rmdir(mCacheDir.c_str());
         }
         NeuralnetworksHidlTest::TearDown();
     }
 
-    void saveModelToCache(sp<IPreparedModel> preparedModel, const hidl_handle& cache1,
-                          const hidl_handle& cache2, ErrorStatus* status) {
-        // Save IPreparedModel to cache.
+    void saveModelToCache(const V1_2::Model& model, const hidl_vec<hidl_handle>& modelCache,
+                          const hidl_vec<hidl_handle>& dataCache, bool* supported,
+                          sp<IPreparedModel>* preparedModel = nullptr) {
+        if (preparedModel != nullptr) *preparedModel = nullptr;
+
+        // See if service can handle model.
+        bool fullySupportsModel = false;
+        Return<void> supportedCall = device->getSupportedOperations_1_2(
+                model,
+                [&fullySupportsModel, &model](ErrorStatus status, const hidl_vec<bool>& supported) {
+                    ASSERT_EQ(ErrorStatus::NONE, status);
+                    ASSERT_EQ(supported.size(), model.operations.size());
+                    fullySupportsModel = std::all_of(supported.begin(), supported.end(),
+                                                     [](bool valid) { return valid; });
+                });
+        ASSERT_TRUE(supportedCall.isOk());
+        *supported = fullySupportsModel;
+        if (!fullySupportsModel) return;
+
+        // Launch prepare model.
+        sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
+        ASSERT_NE(nullptr, preparedModelCallback.get());
         hidl_array<uint8_t, sizeof(mToken)> cacheToken(mToken);
-        Return<ErrorStatus> saveToCacheStatus =
-                preparedModel->saveToCache(cache1, cache2, cacheToken);
-        ASSERT_TRUE(saveToCacheStatus.isOk());
-        *status = static_cast<ErrorStatus>(saveToCacheStatus);
+        Return<ErrorStatus> prepareLaunchStatus =
+                device->prepareModel_1_2(model, ExecutionPreference::FAST_SINGLE_ANSWER, modelCache,
+                                         dataCache, cacheToken, preparedModelCallback);
+        ASSERT_TRUE(prepareLaunchStatus.isOk());
+        ASSERT_EQ(static_cast<ErrorStatus>(prepareLaunchStatus), ErrorStatus::NONE);
+
+        // Retrieve prepared model.
+        preparedModelCallback->wait();
+        ASSERT_EQ(preparedModelCallback->getStatus(), ErrorStatus::NONE);
+        if (preparedModel != nullptr) {
+            *preparedModel =
+                    V1_2::IPreparedModel::castFrom(preparedModelCallback->getPreparedModel())
+                            .withDefault(nullptr);
+        }
     }
 
     bool checkEarlyTermination(ErrorStatus status) {
@@ -157,14 +199,27 @@
         return false;
     }
 
-    void prepareModelFromCache(const hidl_handle& cache1, const hidl_handle& cache2,
+    bool checkEarlyTermination(bool supported) {
+        if (!supported) {
+            LOG(INFO) << "NN VTS: Early termination of test because vendor service cannot "
+                         "prepare model that it does not support.";
+            std::cout << "[          ]   Early termination of test because vendor service cannot "
+                         "prepare model that it does not support."
+                      << std::endl;
+            return true;
+        }
+        return false;
+    }
+
+    void prepareModelFromCache(const hidl_vec<hidl_handle>& modelCache,
+                               const hidl_vec<hidl_handle>& dataCache,
                                sp<IPreparedModel>* preparedModel, ErrorStatus* status) {
         // Launch prepare model from cache.
         sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
         ASSERT_NE(nullptr, preparedModelCallback.get());
         hidl_array<uint8_t, sizeof(mToken)> cacheToken(mToken);
-        Return<ErrorStatus> prepareLaunchStatus =
-                device->prepareModelFromCache(cache1, cache2, cacheToken, preparedModelCallback);
+        Return<ErrorStatus> prepareLaunchStatus = device->prepareModelFromCache(
+                modelCache, dataCache, cacheToken, preparedModelCallback);
         ASSERT_TRUE(prepareLaunchStatus.isOk());
         if (static_cast<ErrorStatus>(prepareLaunchStatus) != ErrorStatus::NONE) {
             *preparedModel = nullptr;
@@ -179,49 +234,54 @@
                                  .withDefault(nullptr);
     }
 
+    // Absolute path to the temporary cache directory.
     std::string mCacheDir;
-    std::string mCache1;
-    std::string mCache2;
-    std::string mCache3;
+
+    // Groups of file paths for model and data cache in the tmp cache directory, initialized with
+    // outer_size = mNum{Model|Data}Cache, inner_size = 1. The outer vector corresponds to handles
+    // and the inner vector is for fds held by each handle.
+    std::vector<std::vector<std::string>> mModelCache;
+    std::vector<std::vector<std::string>> mDataCache;
+
+    // A separate temporary file path in the tmp cache directory.
+    std::string mTmpCache;
+
     uint8_t mToken[static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN)] = {};
-    bool mIsCachingSupported;
+    uint32_t mNumModelCache;
+    uint32_t mNumDataCache;
+    uint32_t mIsCachingSupported;
 };
 
 TEST_F(CompilationCachingTest, CacheSavingAndRetrieval) {
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
     sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
     // Save the compilation to cache.
     {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (!mIsCachingSupported) {
-            EXPECT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-        } else {
-            if (checkEarlyTermination(status)) return;
-            ASSERT_EQ(status, ErrorStatus::NONE);
-        }
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        saveModelToCache(testModel, modelCache, dataCache, &supported);
+        if (checkEarlyTermination(supported)) return;
     }
 
     // Retrieve preparedModel from cache.
     {
         preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         if (!mIsCachingSupported) {
             ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
             ASSERT_EQ(preparedModel, nullptr);
             return;
+        } else if (checkEarlyTermination(status)) {
+            ASSERT_EQ(preparedModel, nullptr);
+            return;
         } else {
             ASSERT_EQ(status, ErrorStatus::NONE);
             ASSERT_NE(preparedModel, nullptr);
@@ -238,41 +298,54 @@
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
     sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
     // Save the compilation to cache.
     {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (!mIsCachingSupported) {
-            EXPECT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-        } else {
-            if (checkEarlyTermination(status)) return;
-            ASSERT_EQ(status, ErrorStatus::NONE);
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        uint8_t dummyBytes[] = {0, 0};
+        // Write a dummy integer to the cache.
+        // The driver should be able to handle non-empty cache and non-zero fd offset.
+        for (uint32_t i = 0; i < modelCache.size(); i++) {
+            ASSERT_EQ(write(modelCache[i].getNativeHandle()->data[0], &dummyBytes,
+                            sizeof(dummyBytes)),
+                      sizeof(dummyBytes));
         }
+        for (uint32_t i = 0; i < dataCache.size(); i++) {
+            ASSERT_EQ(
+                    write(dataCache[i].getNativeHandle()->data[0], &dummyBytes, sizeof(dummyBytes)),
+                    sizeof(dummyBytes));
+        }
+        saveModelToCache(testModel, modelCache, dataCache, &supported);
+        if (checkEarlyTermination(supported)) return;
     }
 
     // Retrieve preparedModel from cache.
     {
         preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
         uint8_t dummyByte = 0;
-        // Advance offset by one byte.
-        ASSERT_GE(read(cache1.getNativeHandle()->data[0], &dummyByte, 1), 0);
-        ASSERT_GE(read(cache2.getNativeHandle()->data[0], &dummyByte, 1), 0);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        // Advance the offset of each handle by one byte.
+        // The driver should be able to handle non-zero fd offset.
+        for (uint32_t i = 0; i < modelCache.size(); i++) {
+            ASSERT_GE(read(modelCache[i].getNativeHandle()->data[0], &dummyByte, 1), 0);
+        }
+        for (uint32_t i = 0; i < dataCache.size(); i++) {
+            ASSERT_GE(read(dataCache[i].getNativeHandle()->data[0], &dummyByte, 1), 0);
+        }
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         if (!mIsCachingSupported) {
             ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
             ASSERT_EQ(preparedModel, nullptr);
             return;
+        } else if (checkEarlyTermination(status)) {
+            ASSERT_EQ(preparedModel, nullptr);
+            return;
         } else {
             ASSERT_EQ(status, ErrorStatus::NONE);
             ASSERT_NE(preparedModel, nullptr);
@@ -285,234 +358,512 @@
                                            /*testDynamicOutputShape=*/false);
 }
 
+TEST_F(CompilationCachingTest, SaveToCacheInvalidNumCache) {
+    // Create test HIDL model and compile.
+    Model testModel = createTestModel();
+
+    // Test with number of model cache files greater than mNumModelCache.
+    {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pass an additional cache file for model cache.
+        mModelCache.push_back({mTmpCache});
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache.pop_back();
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
+        ErrorStatus status;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Test with number of model cache files smaller than mNumModelCache.
+    if (mModelCache.size() > 0) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pop out the last cache file.
+        auto tmp = mModelCache.back();
+        mModelCache.pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache.push_back(tmp);
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
+        ErrorStatus status;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Test with number of data cache files greater than mNumDataCache.
+    {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pass an additional cache file for data cache.
+        mDataCache.push_back({mTmpCache});
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache.pop_back();
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
+        ErrorStatus status;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Test with number of data cache files smaller than mNumDataCache.
+    if (mDataCache.size() > 0) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pop out the last cache file.
+        auto tmp = mDataCache.back();
+        mDataCache.pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache.push_back(tmp);
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
+        ErrorStatus status;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+}
+
+TEST_F(CompilationCachingTest, PrepareModelFromCacheInvalidNumCache) {
+    // Create test HIDL model and compile.
+    Model testModel = createTestModel();
+
+    // Save the compilation to cache.
+    {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        saveModelToCache(testModel, modelCache, dataCache, &supported);
+        if (checkEarlyTermination(supported)) return;
+    }
+
+    // Test with number of model cache files greater than mNumModelCache.
+    {
+        sp<IPreparedModel> preparedModel = nullptr;
+        ErrorStatus status;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        mModelCache.push_back({mTmpCache});
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache.pop_back();
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::GENERAL_FAILURE) {
+            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Test with number of model cache files smaller than mNumModelCache.
+    if (mModelCache.size() > 0) {
+        sp<IPreparedModel> preparedModel = nullptr;
+        ErrorStatus status;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        auto tmp = mModelCache.back();
+        mModelCache.pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache.push_back(tmp);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::GENERAL_FAILURE) {
+            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Test with number of data cache files greater than mNumDataCache.
+    {
+        sp<IPreparedModel> preparedModel = nullptr;
+        ErrorStatus status;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        mDataCache.push_back({mTmpCache});
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache.pop_back();
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::GENERAL_FAILURE) {
+            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Test with number of data cache files smaller than mNumDataCache.
+    if (mDataCache.size() > 0) {
+        sp<IPreparedModel> preparedModel = nullptr;
+        ErrorStatus status;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        auto tmp = mDataCache.back();
+        mDataCache.pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache.push_back(tmp);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::GENERAL_FAILURE) {
+            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+}
+
 TEST_F(CompilationCachingTest, SaveToCacheInvalidNumFd) {
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
-    // cache1 with invalid NumFd.
-    {
+    // Go through each handle in model cache, test with NumFd greater than 1.
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pass an invalid number of fds for handle i.
+        mModelCache[i].push_back(mTmpCache);
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache[i].pop_back();
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1, mCache3}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (status != ErrorStatus::GENERAL_FAILURE) {
-            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
         }
+        ASSERT_EQ(preparedModel, nullptr);
     }
 
-    // cache2 with invalid NumFd.
-    {
+    // Go through each handle in model cache, test with NumFd equal to 0.
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pass an invalid number of fds for handle i.
+        auto tmp = mModelCache[i].back();
+        mModelCache[i].pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache[i].push_back(tmp);
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2, mCache3}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (status != ErrorStatus::GENERAL_FAILURE) {
-            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
         }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Go through each handle in data cache, test with NumFd greater than 1.
+    for (uint32_t i = 0; i < mNumDataCache; i++) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pass an invalid number of fds for handle i.
+        mDataCache[i].push_back(mTmpCache);
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache[i].pop_back();
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
+        ErrorStatus status;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Go through each handle in data cache, test with NumFd equal to 0.
+    for (uint32_t i = 0; i < mNumDataCache; i++) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        // Pass an invalid number of fds for handle i.
+        auto tmp = mDataCache[i].back();
+        mDataCache[i].pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache[i].push_back(tmp);
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
+        ErrorStatus status;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
     }
 }
 
 TEST_F(CompilationCachingTest, PrepareModelFromCacheInvalidNumFd) {
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
     // Save the compilation to cache.
     {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (status != ErrorStatus::GENERAL_FAILURE) {
-            ASSERT_EQ(status, ErrorStatus::NONE);
-        }
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        saveModelToCache(testModel, modelCache, dataCache, &supported);
+        if (checkEarlyTermination(supported)) return;
     }
 
-    // cache1 with invalid NumFd.
-    {
-        preparedModel = nullptr;
+    // Go through each handle in model cache, test with NumFd greater than 1.
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        sp<IPreparedModel> preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1, mCache3}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        mModelCache[i].push_back(mTmpCache);
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache[i].pop_back();
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         if (status != ErrorStatus::GENERAL_FAILURE) {
             ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
-            ASSERT_EQ(preparedModel, nullptr);
         }
+        ASSERT_EQ(preparedModel, nullptr);
     }
 
-    // cache2 with invalid NumFd.
-    {
-        preparedModel = nullptr;
+    // Go through each handle in model cache, test with NumFd equal to 0.
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        sp<IPreparedModel> preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2, mCache3}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        auto tmp = mModelCache[i].back();
+        mModelCache[i].pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mModelCache[i].push_back(tmp);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         if (status != ErrorStatus::GENERAL_FAILURE) {
             ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
-            ASSERT_EQ(preparedModel, nullptr);
         }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Go through each handle in data cache, test with NumFd greater than 1.
+    for (uint32_t i = 0; i < mNumDataCache; i++) {
+        sp<IPreparedModel> preparedModel = nullptr;
+        ErrorStatus status;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        mDataCache[i].push_back(mTmpCache);
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache[i].pop_back();
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::GENERAL_FAILURE) {
+            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
+    }
+
+    // Go through each handle in data cache, test with NumFd equal to 0.
+    for (uint32_t i = 0; i < mNumDataCache; i++) {
+        sp<IPreparedModel> preparedModel = nullptr;
+        ErrorStatus status;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        auto tmp = mDataCache[i].back();
+        mDataCache[i].pop_back();
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        mDataCache[i].push_back(tmp);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::GENERAL_FAILURE) {
+            ASSERT_EQ(status, ErrorStatus::INVALID_ARGUMENT);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
     }
 }
 
 TEST_F(CompilationCachingTest, SaveToCacheInvalidAccessMode) {
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
+    std::vector<AccessMode> modelCacheMode(mNumModelCache, AccessMode::READ_WRITE);
+    std::vector<AccessMode> dataCacheMode(mNumDataCache, AccessMode::READ_WRITE);
 
-    // cache1 with invalid access mode.
-    {
+    // Go through each handle in model cache, test with invalid access mode.
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        modelCacheMode[i] = AccessMode::READ_ONLY;
+        createCacheHandles(mModelCache, modelCacheMode, &modelCache);
+        createCacheHandles(mDataCache, dataCacheMode, &dataCache);
+        modelCacheMode[i] = AccessMode::READ_WRITE;
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
     }
 
-    // cache2 with invalid access mode.
-    {
+    // Go through each handle in data cache, test with invalid access mode.
+    for (uint32_t i = 0; i < mNumDataCache; i++) {
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        dataCacheMode[i] = AccessMode::READ_ONLY;
+        createCacheHandles(mModelCache, modelCacheMode, &modelCache);
+        createCacheHandles(mDataCache, dataCacheMode, &dataCache);
+        dataCacheMode[i] = AccessMode::READ_WRITE;
+        sp<IPreparedModel> preparedModel = nullptr;
+        saveModelToCache(testModel, modelCache, dataCache, &supported, &preparedModel);
+        if (checkEarlyTermination(supported)) return;
+        ASSERT_NE(preparedModel, nullptr);
+        // Execute and verify results.
+        generated_tests::EvaluatePreparedModel(preparedModel, [](int) { return false; },
+                                               get_examples(),
+                                               testModel.relaxComputationFloat32toFloat16,
+                                               /*testDynamicOutputShape=*/false);
+        // Check if prepareModelFromCache fails.
+        preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+        if (status != ErrorStatus::INVALID_ARGUMENT) {
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+        }
+        ASSERT_EQ(preparedModel, nullptr);
     }
 }
 
 TEST_F(CompilationCachingTest, PrepareModelFromCacheInvalidAccessMode) {
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
+    std::vector<AccessMode> modelCacheMode(mNumModelCache, AccessMode::READ_WRITE);
+    std::vector<AccessMode> dataCacheMode(mNumDataCache, AccessMode::READ_WRITE);
 
     // Save the compilation to cache.
     {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (status != ErrorStatus::GENERAL_FAILURE) {
-            ASSERT_EQ(status, ErrorStatus::NONE);
-        }
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        saveModelToCache(testModel, modelCache, dataCache, &supported);
+        if (checkEarlyTermination(supported)) return;
     }
 
-    // cache1 with invalid access mode.
-    {
-        preparedModel = nullptr;
+    // Go through each handle in model cache, test with invalid access mode.
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        sp<IPreparedModel> preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        modelCacheMode[i] = AccessMode::WRITE_ONLY;
+        createCacheHandles(mModelCache, modelCacheMode, &modelCache);
+        createCacheHandles(mDataCache, dataCacheMode, &dataCache);
+        modelCacheMode[i] = AccessMode::READ_WRITE;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
         ASSERT_EQ(preparedModel, nullptr);
     }
 
-    // cache2 with invalid access mode.
-    {
-        preparedModel = nullptr;
+    // Go through each handle in data cache, test with invalid access mode.
+    for (uint32_t i = 0; i < mNumDataCache; i++) {
+        sp<IPreparedModel> preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        dataCacheMode[i] = AccessMode::WRITE_ONLY;
+        createCacheHandles(mModelCache, modelCacheMode, &modelCache);
+        createCacheHandles(mDataCache, dataCacheMode, &dataCache);
+        dataCacheMode[i] = AccessMode::READ_WRITE;
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
         ASSERT_EQ(preparedModel, nullptr);
     }
 }
 
-TEST_F(CompilationCachingTest, SaveToCacheInvalidOffset) {
-    // Create test HIDL model and compile.
-    Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
-
-    // cache1 with invalid file descriptor offset.
-    {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        uint8_t dummyByte = 0;
-        // Advance offset by one byte.
-        ASSERT_EQ(write(cache1.getNativeHandle()->data[0], &dummyByte, 1), 1);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-    }
-
-    // cache2 with invalid file descriptor offset.
-    {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        uint8_t dummyByte = 0;
-        // Advance offset by one byte.
-        ASSERT_EQ(write(cache2.getNativeHandle()->data[0], &dummyByte, 1), 1);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-    }
-}
-
-TEST_F(CompilationCachingTest, SaveToCacheInvalidFileSize) {
-    // Create test HIDL model and compile.
-    Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
-
-    // cache1 with invalid file size.
-    {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        uint8_t dummyByte = 0;
-        // Write one byte and seek back to the beginning.
-        ASSERT_EQ(write(cache1.getNativeHandle()->data[0], &dummyByte, 1), 1);
-        ASSERT_EQ(lseek(cache1.getNativeHandle()->data[0], 0, SEEK_SET), 0);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-    }
-
-    // cache2 with invalid file size.
-    {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        uint8_t dummyByte = 0;
-        // Write one byte and seek back to the beginning.
-        ASSERT_EQ(write(cache2.getNativeHandle()->data[0], &dummyByte, 1), 1);
-        ASSERT_EQ(lseek(cache2.getNativeHandle()->data[0], 0, SEEK_SET), 0);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-    }
-}
-
 class CompilationCachingSecurityTest : public CompilationCachingTest,
                                        public ::testing::WithParamInterface<uint32_t> {
   protected:
@@ -537,44 +888,44 @@
 
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
-    // Save the compilation to cache.
-    {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (checkEarlyTermination(status)) return;
-        ASSERT_EQ(status, ErrorStatus::NONE);
-    }
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        // Save the compilation to cache.
+        {
+            bool supported;
+            hidl_vec<hidl_handle> modelCache, dataCache;
+            createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+            createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+            saveModelToCache(testModel, modelCache, dataCache, &supported);
+            if (checkEarlyTermination(supported)) return;
+        }
 
-    // Randomly flip one single bit of the cache entry.
-    FILE* pFile = fopen(mCache1.c_str(), "r+");
-    ASSERT_EQ(fseek(pFile, 0, SEEK_END), 0);
-    long int fileSize = ftell(pFile);
-    ASSERT_GT(fileSize, 0);
-    ASSERT_EQ(fseek(pFile, getRandomInt(0l, fileSize - 1), SEEK_SET), 0);
-    int readByte = fgetc(pFile);
-    ASSERT_NE(readByte, EOF);
-    ASSERT_EQ(fseek(pFile, -1, SEEK_CUR), 0);
-    ASSERT_NE(fputc(static_cast<uint8_t>(readByte) ^ (1U << getRandomInt(0, 7)), pFile), EOF);
-    fclose(pFile);
+        // Randomly flip one single bit of the cache entry.
+        FILE* pFile = fopen(mModelCache[i][0].c_str(), "r+");
+        ASSERT_EQ(fseek(pFile, 0, SEEK_END), 0);
+        long int fileSize = ftell(pFile);
+        if (fileSize == 0) {
+            fclose(pFile);
+            continue;
+        }
+        ASSERT_EQ(fseek(pFile, getRandomInt(0l, fileSize - 1), SEEK_SET), 0);
+        int readByte = fgetc(pFile);
+        ASSERT_NE(readByte, EOF);
+        ASSERT_EQ(fseek(pFile, -1, SEEK_CUR), 0);
+        ASSERT_NE(fputc(static_cast<uint8_t>(readByte) ^ (1U << getRandomInt(0, 7)), pFile), EOF);
+        fclose(pFile);
 
-    // Retrieve preparedModel from cache, expect failure.
-    {
-        preparedModel = nullptr;
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-        ASSERT_EQ(preparedModel, nullptr);
+        // Retrieve preparedModel from cache, expect failure.
+        {
+            sp<IPreparedModel> preparedModel = nullptr;
+            ErrorStatus status;
+            hidl_vec<hidl_handle> modelCache, dataCache;
+            createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+            createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+            prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+            ASSERT_EQ(preparedModel, nullptr);
+        }
     }
 }
 
@@ -583,40 +934,37 @@
 
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
-    // Save the compilation to cache.
-    {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (checkEarlyTermination(status)) return;
-        ASSERT_EQ(status, ErrorStatus::NONE);
-    }
+    for (uint32_t i = 0; i < mNumModelCache; i++) {
+        // Save the compilation to cache.
+        {
+            bool supported;
+            hidl_vec<hidl_handle> modelCache, dataCache;
+            createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+            createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+            saveModelToCache(testModel, modelCache, dataCache, &supported);
+            if (checkEarlyTermination(supported)) return;
+        }
 
-    // Randomly append bytes to the cache entry.
-    FILE* pFile = fopen(mCache1.c_str(), "a");
-    uint32_t appendLength = getRandomInt(1, 256);
-    for (uint32_t i = 0; i < appendLength; i++) {
-        ASSERT_NE(fputc(getRandomInt<uint8_t>(0, 255), pFile), EOF);
-    }
-    fclose(pFile);
+        // Randomly append bytes to the cache entry.
+        FILE* pFile = fopen(mModelCache[i][0].c_str(), "a");
+        uint32_t appendLength = getRandomInt(1, 256);
+        for (uint32_t i = 0; i < appendLength; i++) {
+            ASSERT_NE(fputc(getRandomInt<uint8_t>(0, 255), pFile), EOF);
+        }
+        fclose(pFile);
 
-    // Retrieve preparedModel from cache, expect failure.
-    {
-        preparedModel = nullptr;
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
-        ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
-        ASSERT_EQ(preparedModel, nullptr);
+        // Retrieve preparedModel from cache, expect failure.
+        {
+            sp<IPreparedModel> preparedModel = nullptr;
+            ErrorStatus status;
+            hidl_vec<hidl_handle> modelCache, dataCache;
+            createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+            createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+            prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
+            ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
+            ASSERT_EQ(preparedModel, nullptr);
+        }
     }
 }
 
@@ -625,20 +973,15 @@
 
     // Create test HIDL model and compile.
     Model testModel = createTestModel();
-    sp<IPreparedModel> preparedModel = nullptr;
-    generated_tests::PrepareModel(device, testModel, &preparedModel);
-    // Terminate early if the driver cannot prepare the model.
-    if (preparedModel == nullptr) return;
 
     // Save the compilation to cache.
     {
-        ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::WRITE_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::WRITE_ONLY, &cache2);
-        saveModelToCache(preparedModel, cache1, cache2, &status);
-        if (checkEarlyTermination(status)) return;
-        ASSERT_EQ(status, ErrorStatus::NONE);
+        bool supported;
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        saveModelToCache(testModel, modelCache, dataCache, &supported);
+        if (checkEarlyTermination(supported)) return;
     }
 
     // Randomly flip one single bit in mToken.
@@ -647,12 +990,12 @@
 
     // Retrieve the preparedModel from cache, expect failure.
     {
-        preparedModel = nullptr;
+        sp<IPreparedModel> preparedModel = nullptr;
         ErrorStatus status;
-        hidl_handle cache1, cache2;
-        createCacheHandle({mCache1}, AccessMode::READ_ONLY, &cache1);
-        createCacheHandle({mCache2}, AccessMode::READ_ONLY, &cache2);
-        prepareModelFromCache(cache1, cache2, &preparedModel, &status);
+        hidl_vec<hidl_handle> modelCache, dataCache;
+        createCacheHandles(mModelCache, AccessMode::READ_WRITE, &modelCache);
+        createCacheHandles(mDataCache, AccessMode::READ_WRITE, &dataCache);
+        prepareModelFromCache(modelCache, dataCache, &preparedModel, &status);
         ASSERT_EQ(status, ErrorStatus::GENERAL_FAILURE);
         ASSERT_EQ(preparedModel, nullptr);
     }
diff --git a/neuralnetworks/1.2/vts/functional/ValidateModel.cpp b/neuralnetworks/1.2/vts/functional/ValidateModel.cpp
index 065546d..dc452e9 100644
--- a/neuralnetworks/1.2/vts/functional/ValidateModel.cpp
+++ b/neuralnetworks/1.2/vts/functional/ValidateModel.cpp
@@ -33,6 +33,7 @@
 
 using ::android::hardware::neuralnetworks::V1_2::implementation::ExecutionCallback;
 using ::android::hardware::neuralnetworks::V1_2::implementation::PreparedModelCallback;
+using HidlToken = hidl_array<uint8_t, static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN)>;
 
 ///////////////////////// UTILITY FUNCTIONS /////////////////////////
 
@@ -54,7 +55,8 @@
     sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
     ASSERT_NE(nullptr, preparedModelCallback.get());
     Return<ErrorStatus> prepareLaunchStatus =
-        device->prepareModel_1_2(model, preference, preparedModelCallback);
+            device->prepareModel_1_2(model, preference, hidl_vec<hidl_handle>(),
+                                     hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
     ASSERT_TRUE(prepareLaunchStatus.isOk());
     ASSERT_EQ(ErrorStatus::INVALID_ARGUMENT, static_cast<ErrorStatus>(prepareLaunchStatus));
 
diff --git a/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp b/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp
index d411da4..b15f657 100644
--- a/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp
+++ b/neuralnetworks/1.2/vts/functional/ValidateRequest.cpp
@@ -37,6 +37,7 @@
 using ::android::hardware::neuralnetworks::V1_2::implementation::ExecutionCallback;
 using ::android::hardware::neuralnetworks::V1_2::implementation::PreparedModelCallback;
 using ::android::hidl::memory::V1_0::IMemory;
+using HidlToken = hidl_array<uint8_t, static_cast<uint32_t>(Constant::BYTE_SIZE_OF_CACHE_TOKEN)>;
 using test_helper::for_all;
 using test_helper::MixedTyped;
 using test_helper::MixedTypedExample;
@@ -66,7 +67,8 @@
     sp<PreparedModelCallback> preparedModelCallback = new PreparedModelCallback();
     ASSERT_NE(nullptr, preparedModelCallback.get());
     Return<ErrorStatus> prepareLaunchStatus = device->prepareModel_1_2(
-        model, ExecutionPreference::FAST_SINGLE_ANSWER, preparedModelCallback);
+            model, ExecutionPreference::FAST_SINGLE_ANSWER, hidl_vec<hidl_handle>(),
+            hidl_vec<hidl_handle>(), HidlToken(), preparedModelCallback);
     ASSERT_TRUE(prepareLaunchStatus.isOk());
     ASSERT_EQ(ErrorStatus::NONE, static_cast<ErrorStatus>(prepareLaunchStatus));