Rebased on top of recent changes

Reviewed By: cccclai

Author: rohansjoshi

examples/qualcomm/oss_scripts/llama/runner/runner.cpp

@@ -323,7 +323,8 @@ Error Runner<T>::generate(
   return generate_from_pos(prompt, 0, config, token_callback, stats_callback);
 }
 
-Error Runner::generate_from_pos(
+template <typename T>
+Error Runner<T>::generate_from_pos(
     const std::string& prompt,
     int64_t start_pos,
     const llm::GenerationConfig& config,
@@ -334,7 +335,8 @@ Error Runner::generate_from_pos(
       prompt, false, config, token_callback, stats_callback);
 }
 
-Error Runner::generate_from_prompt_or_file(
+template <typename T>
+Error Runner<T>::generate_from_prompt_or_file(
     const std::string& prompt,
     bool tokenized_prompt,
     const llm::GenerationConfig& config,

extension/android/CMakeLists.txt

@@ -180,18 +180,19 @@ if(EXECUTORCH_BUILD_LLAMA_JNI)
   )
 
   target_sources(
-    executorch_jni PRIVATE
-    ${EXECUTORCH_ROOT}/extension/llm/runner/llm_runner_helper.cpp
+    executorch_jni
+    PRIVATE ${EXECUTORCH_ROOT}/extension/llm/runner/llm_runner_helper.cpp
   )
 
   target_include_directories(
-    executorch_jni PRIVATE
-    ${EXECUTORCH_ROOT}/extension/llm/runner
+    executorch_jni
+    PRIVATE ${EXECUTORCH_ROOT}/extension/llm/runner
   )
 
   if(QNN_SDK_ROOT)
     target_sources(
-      executorch_jni PRIVATE
+      executorch_jni
+      PRIVATE
       ${EXECUTORCH_ROOT}/examples/qualcomm/oss_scripts/llama/runner/runner.cpp
       ${EXECUTORCH_ROOT}/examples/qualcomm/oss_scripts/llama/runner/decoder_runner.cpp
       ${EXECUTORCH_ROOT}/examples/qualcomm/oss_scripts/llama/runner/prompt_processor.cpp
@@ -202,8 +203,8 @@ if(EXECUTORCH_BUILD_LLAMA_JNI)
     )
 
     target_include_directories(
-      executorch_jni PRIVATE
-      ${EXECUTORCH_ROOT}/examples/qualcomm/oss_scripts/llama/runner
+      executorch_jni
+      PRIVATE ${EXECUTORCH_ROOT}/examples/qualcomm/oss_scripts/llama/runner
     )
     target_compile_definitions(executorch_jni PRIVATE EXECUTORCH_BUILD_QNN=1)
   endif()

extension/android/jni/jni_layer_llama.cpp

@@ -15,6 +15,7 @@
 
 #include <executorch/examples/models/llama/runner/runner.h>
 #include <executorch/examples/models/llava/runner/llava_runner.h>
+#include <executorch/examples/qualcomm/oss_scripts/llama/runner/runner.h>
 #include <executorch/extension/llm/runner/image.h>
 #include <executorch/extension/llm/runner/irunner.h>
 #include <executorch/runtime/platform/log.h>
@@ -181,10 +182,10 @@ class ExecuTorchLlmJni : public facebook::jni::HybridClass<ExecuTorchLlmJni> {
           data_path_str);
 #if defined(EXECUTORCH_BUILD_QNN)
     } else if (model_type_category == MODEL_TYPE_QNN_LLAMA) {
-      std::unique_ptr<executorch::extension::Module> module =
-        std::make_unique<executorch::extension::Module>(
-            FLAGS_model_path.c_str(),
-            executorch::extension::Module::LoadMode::MmapUseMlockIgnoreErrors);
+      std::unique_ptr<executorch::extension::Module> module = std::make_unique<
+          executorch::extension::Module>(
+          model_path->toStdString().c_str(),
+          executorch::extension::Module::LoadMode::MmapUseMlockIgnoreErrors);
       std::string decoder_model = "llama3"; // use llama3 for now
       runner_ = std::make_unique<example::Runner<uint16_t>>( // QNN runner
           std::move(module),