build: build with executorch 1.1 (#765) (#806)

## Description

Bumped ExecuTorch binaries & headers to 1.1.0 version of the package.

### Introduces a breaking change?

- [ ] Yes
- [x] No

### Type of change

- [ ] Bug fix (change which fixes an issue)
- [ ] New feature (change which adds functionality)
- [ ] Documentation update (improves or adds clarity to existing
documentation)
- [x] Other (chores, tests, code style improvements etc.)

### Tested on

- [x] iOS
- [x] Android

### Testing instructions

Run all the test from
`packages/react-native-executorch/common/rnexecutorch/tests` and verify
the correctness of all models (except Speech-to-Text, see the
**Additional notes** section).

### Screenshots

<!-- Add screenshots here, if applicable -->

### Related issues

<!-- Link related issues here using #issue-number -->

### Checklist

- [ ] I have performed a self-review of my code
- [ ] I have commented my code, particularly in hard-to-understand areas
- [ ] I have updated the documentation accordingly
- [x] My changes generate no new warnings

### Additional notes

I tested most of the models on both Android & iOS platforms. The only
'broken' model seems to be Speech To Text Encoder - but this one will be
fixed in the incoming STT fix PR.

Author: IgorSwat

apps/llm/app/voice_chat/index.tsx

@@ -101,7 +101,7 @@ function VoiceChatScreen() {
   return !llm.isReady || !speechToText.isReady ? (
     <Spinner
       visible={!llm.isReady || !speechToText.isReady}
-      textContent={`Loading the model ${(llm.downloadProgress * 100).toFixed(0)} %\nLoading the speech model ${(speechToText.downloadProgress * 100).toFixed(0)} %`}
+      textContent={`Loading the LLM model ${(llm.downloadProgress * 100).toFixed(0)} %\nLoading the speech model ${(speechToText.downloadProgress * 100).toFixed(0)} %`}
     />
   ) : (
     <TouchableWithoutFeedback onPress={Keyboard.dismiss}>

packages/react-native-executorch/android/libs/classes.jar

@@ -9,7 +9,7 @@
 
 namespace rnexecutorch {
 using namespace facebook;
-using namespace executorch::extension::constants;
+using namespace executorch::extension::llm;
 
 TokenizerModule::TokenizerModule(
     std::string source, std::shared_ptr<react::CallInvoker> callInvoker)

packages/react-native-executorch/common/rnexecutorch/TokenizerModule.cpp

@@ -46,7 +46,7 @@ class BaseModel {
   // (unnecessary copies instead of working on JS memory). In this case
   // CallInvoker can be used to get jsi::Runtime, and use it in a safe manner.
   std::shared_ptr<react::CallInvoker> callInvoker;
-  std::unique_ptr<executorch::extension::Module> module_;
+  std::unique_ptr<Module> module_;
 
   std::size_t memorySizeLowerBound{0};
 

packages/react-native-executorch/common/rnexecutorch/models/BaseModel.h

@@ -74,7 +74,7 @@ TEST(S2TTranscribeTests, TranscribeReturnsValidChars) {
   auto result = model.transcribe(audio, "en", true);
   ASSERT_EQ(result.language, "en");
   EXPECT_GE(result.duration, 20.0f);
-  ASSERT_EQ(result.task, "transcription");
+  ASSERT_EQ(result.task, "transcribe");
   ASSERT_FALSE(result.segments.empty());
   ASSERT_FALSE(result.text.empty());
   for (char c : result.text) {

packages/react-native-executorch/common/rnexecutorch/tests/integration/SpeechToTextTest.cpp

@@ -7,7 +7,7 @@
  */
 #pragma once
 // constants for LLM runtime
-namespace executorch::extension::constants {
+namespace executorch::extension::llm {
 
 // Runtime metadata key constants
 inline constexpr auto kEnableDynamicShape = "enable_dynamic_shape";
@@ -27,4 +27,5 @@ inline constexpr auto kTextModelMethod = "text_decoder";
 
 inline constexpr auto numOfAddedBoSTokens = 0;
 inline constexpr auto numOfAddedEoSTokens = 0;
-} // namespace executorch::extension::constants
+
+} // namespace executorch::extension::llm

packages/react-native-executorch/common/runner/constants.h

@@ -65,6 +65,10 @@ struct GenerationConfig {
 
   // Use KV_CACHE implementation (if implemented) or not
   bool enable_kv_cache = true;
+
+  // Number of eos and bos to add to the prompt
+  int32_t num_bos = 0;
+  int32_t num_eos = 0;
 };
 
 // Base interface for LLM runners

packages/react-native-executorch/common/runner/irunner.h

@@ -15,8 +15,8 @@
 #pragma once
 
 // This list should be very conservative since most kernel .cpp files will
-// include these and depend on their transitive deps. Only add a header if 99%
-// of kernels would have included it anyway.
+// include these and depend on their transitive deps. Only add a header if
+// 99% of kernels would have included it anyway.
 #include <executorch/runtime/core/exec_aten/exec_aten.h> // IWYU pragma: export
 #include <executorch/runtime/core/exec_aten/util/scalar_type_util.h> // IWYU pragma: export
 #include <executorch/runtime/core/exec_aten/util/tensor_util.h> // IWYU pragma: export

packages/react-native-executorch/common/runner/kernel_includes.h

@@ -19,7 +19,7 @@
 
 namespace example {
 
-using namespace executorch::extension::constants;
+using namespace executorch::extension::llm;
 using ::executorch::extension::Module;
 using ::executorch::runtime::Error;
 using ::executorch::runtime::Result;

packages/react-native-executorch/common/runner/runner.cpp

@@ -26,6 +26,8 @@ namespace extension {
 namespace llm {
 // A simple llama2 sampler.
 
+inline constexpr auto kTopp = 0.9f;
+
 template <typename T> struct ProbIndex {
   T prob;
   int32_t index;
@@ -65,3 +67,9 @@ using ::executorch::extension::llm::ProbIndex;
 using ::executorch::extension::llm::Sampler;
 } // namespace executor
 } // namespace torch
+
+namespace executorch::llm {
+// TODO(T197294990): Remove these deprecated aliases once all users have moved
+// to the new `::executorch::extension::llm` namespaces.
+using ::executorch::extension::llm::kTopp;
+} // namespace executorch::llm