diff --git a/.ci/scripts/build-qnn-sdk.sh b/.ci/scripts/build-qnn-sdk.sh
index d912069b06..260072f734 100644
--- a/.ci/scripts/build-qnn-sdk.sh
+++ b/.ci/scripts/build-qnn-sdk.sh
@@ -5,15 +5,43 @@
 # This source code is licensed under the BSD-style license found in the
 # LICENSE file in the root directory of this source tree.
 
-set -ex
+set -eux
 
 build_qnn_backend() {
   echo "Start building qnn backend."
   export ANDROID_NDK_ROOT=/opt/ndk
   export QNN_SDK_ROOT=/tmp/qnn/2.23.0.240531
-  export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/.." && pwd)"
+  export EXECUTORCH_ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd)"
 
   bash backends/qualcomm/scripts/build.sh --skip_aarch64 --job_number 2 --release
 }
 
+set_up_aot() {
+  cd $EXECUTORCH_ROOT
+  if [ ! -d "cmake-out" ]; then
+      mkdir cmake-out
+  fi
+  pushd cmake-out
+  cmake .. \
+      -DCMAKE_INSTALL_PREFIX=$PWD \
+      -DEXECUTORCH_BUILD_QNN=ON \
+      -DQNN_SDK_ROOT=${QNN_SDK_ROOT} \
+      -DEXECUTORCH_BUILD_SDK=ON \
+      -DEXECUTORCH_BUILD_EXTENSION_MODULE=ON \
+      -DEXECUTORCH_ENABLE_EVENT_TRACER=ON \
+      -DPYTHON_EXECUTABLE=python3 \
+      -DEXECUTORCH_SEPARATE_FLATCC_HOST_PROJECT=OFF
+  cmake --build $PWD --target "PyQnnManagerAdaptor" "PyQnnWrapperAdaptor" -j$(nproc)
+  # install Python APIs to correct import path
+  # The filename might vary depending on your Python and host version.
+  cp -f backends/qualcomm/PyQnnManagerAdaptor.cpython-310-x86_64-linux-gnu.so $EXECUTORCH_ROOT/backends/qualcomm/python
+  cp -f backends/qualcomm/PyQnnWrapperAdaptor.cpython-310-x86_64-linux-gnu.so $EXECUTORCH_ROOT/backends/qualcomm/python
+  popd
+
+  # Workaround for fbs files in exir/_serialize
+  cp schema/program.fbs exir/_serialize/program.fbs
+  cp schema/scalar_type.fbs exir/_serialize/scalar_type.fbs
+}
+
 build_qnn_backend
+set_up_aot
diff --git a/.ci/scripts/test.sh b/.ci/scripts/test.sh
index ad02fdc79d..bdb71921bf 100755
--- a/.ci/scripts/test.sh
+++ b/.ci/scripts/test.sh
@@ -28,10 +28,25 @@ if [[ -z "${BACKEND:-}" ]]; then
   exit 1
 fi
 
+UPLOAD_DIR=${4:-}
+
+if [[ -z "${PYTHON_EXECUTABLE:-}" ]]; then
+  PYTHON_EXECUTABLE=python3
+fi
 which "${PYTHON_EXECUTABLE}"
+
 # Just set this variable here, it's cheap even if we use buck2
 CMAKE_OUTPUT_DIR=cmake-out
 
+prepare_artifacts_upload() {
+  if [ -n "$UPLOAD_DIR" ]; then
+    echo "Preparing for uploading generated artifacs"
+    zip -j model.zip "${EXPORTED_MODEL}"
+    mkdir -p "${UPLOAD_DIR}"
+    mv model.zip "${UPLOAD_DIR}"
+  fi
+}
+
 build_cmake_executor_runner() {
   echo "Building executor_runner"
   (rm -rf ${CMAKE_OUTPUT_DIR} \
@@ -129,9 +144,30 @@ test_model_with_xnnpack() {
   fi
 }
 
+test_model_with_qnn() {
+  source "$(dirname "${BASH_SOURCE[0]}")/build-qnn-sdk.sh"
+  echo "ANDROID_NDK_ROOT: $ANDROID_NDK_ROOT"
+  echo "QNN_SDK_ROOT: $QNN_SDK_ROOT"
+  echo "EXECUTORCH_ROOT: $EXECUTORCH_ROOT"
+
+  export LD_LIBRARY_PATH=$QNN_SDK_ROOT/lib/x86_64-linux-clang/
+  export PYTHONPATH=$EXECUTORCH_ROOT/..
+
+  if [[ "${MODEL_NAME}" == "dl3" ]]; then
+    "${PYTHON_EXECUTABLE}" -m examples.qualcomm.scripts.deeplab_v3 -b ${CMAKE_OUTPUT_DIR} -m SM8550 --compile_only --download
+    EXPORTED_MODEL=./${CMAKE_OUTPUT_DIR}/deeplab_v3/dlv3_qnn.pte
+  fi
+}
+
 if [[ "${BACKEND}" == "portable" ]]; then
   echo "Testing ${MODEL_NAME} with portable kernels..."
   test_model
+elif [[ "${BACKEND}" == "qnn" ]]; then
+  echo "Testing ${MODEL_NAME} with qnn..."
+  test_model_with_qnn
+  if [[ $? -eq 0 ]]; then
+    prepare_artifacts_upload
+  fi
 else
   set +e
   if [[ "${BACKEND}" == *"quantization"* ]]; then
@@ -153,5 +189,7 @@ else
   if [[ -n "${Q_ERROR:-}" ]] || [[ -n "${D_ERROR:-}" ]] || [[ -n "${Q_D_ERROR:-}" ]]; then
     echo "Portable q8 ${Q_ERROR:-ok}," "Delegation fp32 ${D_ERROR:-ok}," "Delegation q8 ${Q_D_ERROR:-ok}"
     exit 1
+  else
+    prepare_artifacts_upload
   fi
 fi
diff --git a/.github/workflows/android-perf.yml b/.github/workflows/android-perf.yml
index fcc3421772..86219a817e 100644
--- a/.github/workflows/android-perf.yml
+++ b/.github/workflows/android-perf.yml
@@ -135,24 +135,41 @@ jobs:
       fail-fast: false
     with:
       runner: linux.2xlarge
-      docker-image: executorch-ubuntu-22.04-clang12
+      docker-image: executorch-ubuntu-22.04-clang12-android
       submodules: 'true'
       timeout: 60
       upload-artifact: android-models
       script: |
         # The generic Linux job chooses to use base env, not the one setup by the image
+        echo "::group::Setting up dev environment"
         CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
         conda activate "${CONDA_ENV}"
-
+        if [[ ${{ matrix.delegate }} == "qnn" ]]; then
+            PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
+            PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
+        fi
         PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh "cmake"
-        echo "Exporting model: ${{ matrix.model }}"
-        export ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded/${{ matrix.model }}_${{ matrix.delegate }}
+        ARTIFACTS_DIR_NAME=artifacts-to-be-uploaded/${{ matrix.model }}_${{ matrix.delegate }}
+        echo "::endgroup::"
+
+        echo "::group::Exporting ${{ matrix.delegate }} model: ${{ matrix.model }}"
+        BUILD_MODE="cmake"
+        DTYPE="fp32"
+        if [[ ${{ matrix.delegate }} == "qnn" ]]; then
+            DELEGATE_CONFIG="qnn"
+        elif [[ ${{ matrix.delegate }} == "xnnpack" ]]; then
+            DELEGATE_CONFIG="xnnpack+custom+qe"
+        fi
 
-        # TODO(T197546696): Note that the following scripts/steps only work for llama. It's expected to fail for other models+delegates.
-        # Install requirements for export_llama
-        PYTHON_EXECUTABLE=python bash examples/models/llama2/install_requirements.sh
-        # Test llama2
-        PYTHON_EXECUTABLE=python bash .ci/scripts/test_llama.sh "${{ matrix.model }}.pt" "cmake" "fp32" "xnnpack+custom+qe" "${ARTIFACTS_DIR_NAME}"\
+        if [[ ${{ matrix.model }} == "llama*" ]]; then
+            # Install requirements for export_llama
+            PYTHON_EXECUTABLE=python bash examples/models/llama2/install_requirements.sh
+            # Test llama2
+            PYTHON_EXECUTABLE=python bash .ci/scripts/test_llama.sh "${{ matrix.model }}.pt" "${BUILD_MODE}" "${DTYPE}" "${DELEGATE_CONFIG}" "${ARTIFACTS_DIR_NAME}"
+        else
+            PYTHON_EXECUTABLE=python bash .ci/scripts/test.sh "${{ matrix.model }}" "${BUILD_MODE}" "${DELEGATE_CONFIG}" "${ARTIFACTS_DIR_NAME}"
+        fi
+        echo "::endgroup::"
 
   # Upload models to S3. The artifacts are needed not only by the device farm but also TorchChat
   upload-models:
diff --git a/.github/workflows/trunk.yml b/.github/workflows/trunk.yml
index 9b28d26048..969ea3d361 100644
--- a/.github/workflows/trunk.yml
+++ b/.github/workflows/trunk.yml
@@ -270,3 +270,26 @@ jobs:
         PYTHON_EXECUTABLE=python ${CONDA_RUN} bash examples/models/llama2/install_requirements.sh
         # Test llama2
         PYTHON_EXECUTABLE=python ${CONDA_RUN} bash .ci/scripts/test_llama.sh stories110M.pt "${BUILD_TOOL}" "${DTYPE}" "${MODE}"
+
+  test-qnn-model:
+    name: test-qnn-model
+    uses: pytorch/test-infra/.github/workflows/linux_job.yml@main
+    strategy:
+      matrix:
+        dtype: [fp32]
+        model: [dl3]
+      fail-fast: false
+    with:
+      runner: linux.2xlarge
+      docker-image: executorch-ubuntu-22.04-clang12-android
+      submodules: 'true'
+      ref: ${{ github.event_name == 'pull_request' && github.event.pull_request.head.sha || github.sha }}
+      timeout: 900
+      script: |
+        # The generic Linux job chooses to use base env, not the one setup by the image
+        CONDA_ENV=$(conda env list --json | jq -r ".envs | .[-1]")
+        conda activate "${CONDA_ENV}"
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-linux.sh cmake
+        PYTHON_EXECUTABLE=python bash .ci/scripts/setup-qnn-deps.sh
+        PYTHON_EXECUTABLE=python bash .ci/scripts/build-qnn-sdk.sh
+        PYTHON_EXECUTABLE=python bash .ci/scripts/test.sh ${{ matrix.model }} "cmake" "qnn"