diff --git a/examples/models/llava/main.cpp b/examples/models/llava/main.cpp
index 431f86c906e..171eb77077f 100644
--- a/examples/models/llava/main.cpp
+++ b/examples/models/llava/main.cpp
@@ -103,6 +103,6 @@ int32_t main(int32_t argc, char** argv) {
        .width = static_cast<int32_t>(image_tensor.size(2)),
        .height = static_cast<int32_t>(image_tensor.size(1))}};
   // generate
-  runner.generate(images, prompt, seq_len);
+  runner.generate(std::move(images), prompt, seq_len);
   return 0;
 }
diff --git a/examples/models/llava/runner/llava_runner.cpp b/examples/models/llava/runner/llava_runner.cpp
index b186af892f1..cb968ca88d8 100644
--- a/examples/models/llava/runner/llava_runner.cpp
+++ b/examples/models/llava/runner/llava_runner.cpp
@@ -74,7 +74,7 @@ Error LlavaRunner::load() {
 }
 
 Error LlavaRunner::generate(
-    std::vector<Image>& images,
+    std::vector<Image> images,
     const std::string& prompt,
     int32_t seq_len,
     std::function<void(const std::string&)> token_callback,
diff --git a/examples/models/llava/runner/llava_runner.h b/examples/models/llava/runner/llava_runner.h
index 13d842e30fe..9b14bc9283a 100644
--- a/examples/models/llava/runner/llava_runner.h
+++ b/examples/models/llava/runner/llava_runner.h
@@ -31,7 +31,7 @@ class LlavaRunner : public MultimodalRunner {
   bool is_loaded();
   Error load();
   Error generate(
-      std::vector<Image>& images,
+      std::vector<Image> images,
       const std::string& prompt,
       int32_t seq_len = 1024,
       std::function<void(const std::string&)> token_callback = {},
diff --git a/examples/models/llava/runner/targets.bzl b/examples/models/llava/runner/targets.bzl
index 435ab2a8c70..72942acf16f 100644
--- a/examples/models/llava/runner/targets.bzl
+++ b/examples/models/llava/runner/targets.bzl
@@ -8,6 +8,9 @@ def define_common_targets():
         visibility = [
             "@EXECUTORCH_CLIENTS",
         ],
+        compiler_flags = [
+            "-Wno-global-constructors",
+        ],
         exported_deps = [
             "//executorch/backends/xnnpack:xnnpack_backend",
             "//executorch/extension/llm/runner:runner_lib",
diff --git a/extension/android/jni/jni_layer_llama.cpp b/extension/android/jni/jni_layer_llama.cpp
index 4f67d04396c..dda9ece589d 100644
--- a/extension/android/jni/jni_layer_llama.cpp
+++ b/extension/android/jni/jni_layer_llama.cpp
@@ -165,7 +165,7 @@ class ExecuTorchLlamaJni
         images.push_back(image_runner);
       }
       multi_modal_runner_->generate(
-          images,
+          std::move(images),
           prompt->toStdString(),
           seq_len,
           [callback](std::string result) { callback->onResult(result); },
diff --git a/extension/llm/runner/multimodal_runner.h b/extension/llm/runner/multimodal_runner.h
index 745f086f803..dbffac46fce 100644
--- a/extension/llm/runner/multimodal_runner.h
+++ b/extension/llm/runner/multimodal_runner.h
@@ -56,7 +56,7 @@ class MultimodalRunner {
   virtual bool is_loaded() = 0;
   virtual ::executorch::runtime::Error load() = 0;
   virtual ::executorch::runtime::Error generate(
-      std::vector<Image>& images,
+      std::vector<Image> images,
       const std::string& prompt,
       int32_t seq_len = 1024,
       std::function<void(const std::string&)> token_callback = {},