diff --git a/extension/llm/runner/multimodal_runner.cpp b/extension/llm/runner/multimodal_runner.cpp
index 9e47f48ebf3..f7a27c968d6 100644
--- a/extension/llm/runner/multimodal_runner.cpp
+++ b/extension/llm/runner/multimodal_runner.cpp
@@ -62,7 +62,7 @@ Error MultimodalRunner::load() {
     ET_LOG(Info, format, __VA_ARGS__);     \
   }
 
-Error MultimodalRunner::prefill(std::vector<MultimodalInput>& inputs) {
+Error MultimodalRunner::prefill(const std::vector<MultimodalInput>& inputs) {
   if (!is_loaded()) {
     ET_CHECK_OK_OR_RETURN_ERROR(load());
   }
diff --git a/extension/llm/runner/multimodal_runner.h b/extension/llm/runner/multimodal_runner.h
index 4a824fd4d9c..caf3c296038 100644
--- a/extension/llm/runner/multimodal_runner.h
+++ b/extension/llm/runner/multimodal_runner.h
@@ -126,7 +126,7 @@ class ET_EXPERIMENTAL MultimodalRunner {
    * @return The error code. KV cache position is tracked internally in pos_.
    */
   virtual ::executorch::runtime::Error prefill(
-      std::vector<MultimodalInput>& inputs);
+      const std::vector<MultimodalInput>& inputs);
 
   inline void stop() {
     text_token_generator_->stop();