fix: requested changes

Author: NorbertKlockiewicz

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

@@ -24,24 +24,23 @@ BaseLLMRunner::BaseLLMRunner(std::unique_ptr<Module> module,
       }) {}
 
 Error BaseLLMRunner::load() {
-  if (is_loaded())
+  if (is_loaded()) {
     return Error::Ok;
+  }
 
   auto status = tokenizer_->load(tokenizer_path_);
   if (status != tokenizers::Error::Ok) {
     throw rnexecutorch::RnExecutorchError(
         rnexecutorch::RnExecutorchErrorCode::TokenizerError,
         "Unexpected issue occurred while loading tokenizer (error code: " +
-            std::to_string(static_cast<int>(status)) + ")");
+            std::to_string(static_cast<int32_t>(status)) + ")");
   }
 
   const auto method_names =
       ET_UNWRAP(module_->method_names(), "Failed reading method names");
 
   metadata_[kVocabSize] = tokenizer_->vocab_size();
-  for (auto &pair : metadata_) {
-    const auto &method_name = pair.first;
-    auto &value = pair.second;
+  for (auto &[method_name, value] : metadata_) {
     if (method_names.count(method_name)) {
       value = ET_UNWRAP(module_->get(method_name))
                   .toScalar()
@@ -71,7 +70,10 @@ Error BaseLLMRunner::load() {
     }
   }
   if (eos_ids_->empty()) {
-    eos_ids_->emplace(7); // fallback <|im_end|>
+    throw rnexecutorch::RnExecutorchError(
+        rnexecutorch::RnExecutorchErrorCode::InvalidModelOutput,
+        "Model did not provide any EOS token IDs via 'get_eos_ids'. "
+        "The .pte file may be misconfigured.");
   }
 
   io_manager_ = std::make_unique<IOManager>(*module_);
@@ -89,8 +91,9 @@ Error BaseLLMRunner::generate(
   std::vector<MultimodalInput> inputs = {make_text_input(prompt)};
   auto err = generate_internal(inputs, token_callback);
 
-  if (stats_callback)
+  if (stats_callback) {
     stats_callback(stats_);
+  }
 
   return err;
 }
@@ -102,8 +105,9 @@ Error BaseLLMRunner::generate(
 
   auto err = generate_internal(inputs, token_callback);
 
-  if (stats_callback)
+  if (stats_callback) {
     stats_callback(stats_);
+  }
 
   return err;
 }

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

@@ -40,7 +40,7 @@ bool VisionEncoder::is_loaded() const noexcept {
   return module_->is_method_loaded(kVisionEncoderMethod);
 }
 
-int32_t VisionEncoder::encoderTokenCount() const noexcept {
+int32_t VisionEncoder::encoderTokenCount() const {
   if (!is_loaded()) {
     return 0;
   }

packages/react-native-executorch/common/runner/encoders/vision_encoder.h

@@ -18,7 +18,7 @@ class VisionEncoder : public IEncoder {
   bool is_loaded() const noexcept override;
   ::executorch::runtime::Result<::executorch::runtime::EValue>
   encode(const MultimodalInput &input) override;
-  int32_t encoderTokenCount() const noexcept override;
+  int32_t encoderTokenCount() const override;
 
 private:
   struct ImageShape {

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

@@ -22,11 +22,8 @@ struct ImagePath {
 
 class MultimodalInput {
 public:
-  explicit MultimodalInput(const std::string &text) : data_(text) {}
-  explicit MultimodalInput(std::string &&text) : data_(std::move(text)) {}
-  explicit MultimodalInput(const std::vector<uint64_t> &tokens)
-      : data_(tokens) {}
-  explicit MultimodalInput(std::vector<uint64_t> &&tokens)
+  explicit MultimodalInput(std::string text) : data_(std::move(text)) {}
+  explicit MultimodalInput(std::vector<uint64_t> tokens)
       : data_(std::move(tokens)) {}
   explicit MultimodalInput(ImagePath image_path)
       : data_(std::move(image_path)) {}

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

@@ -12,6 +12,7 @@
 #include "multimodal_prefiller.h"
 #include "constants.h"
 #include "util.h"
+#include <algorithm>
 
 namespace executorch::extension::llm {
 
@@ -21,11 +22,10 @@ using ::executorch::runtime::EValue;
 using ::executorch::runtime::Result;
 
 MultimodalPrefiller::MultimodalPrefiller(
-    Module *module, MultimodalDecoderRunner *decoder_runner,
-    tokenizers::HFTokenizer *tokenizer, IOManager *io_manager,
-    IEncoder *image_encoder)
-    : module_(module), decoder_runner_(decoder_runner), tokenizer_(tokenizer),
-      io_manager_(io_manager), image_encoder_(image_encoder) {}
+    Module &module, MultimodalDecoderRunner &decoder_runner,
+    tokenizers::HFTokenizer &tokenizer, IEncoder *image_encoder)
+    : module_(&module), decoder_runner_(&decoder_runner),
+      tokenizer_(&tokenizer), image_encoder_(image_encoder) {}
 
 Result<uint64_t> MultimodalPrefiller::prefill(const MultimodalInput &input,
                                               int64_t &start_pos) {
@@ -65,7 +65,7 @@ Result<uint64_t> MultimodalPrefiller::prefill(const MultimodalInput &input,
     }
 
     padded_tokens_storage.assign(max_seq_len, 0);
-    std::copy(tokens.begin(), tokens.end(), padded_tokens_storage.begin());
+    std::ranges::copy(tokens, padded_tokens_storage.begin());
 
     auto text_tensor = ::executorch::extension::from_blob(
         padded_tokens_storage.data(), {1, static_cast<SizesType>(max_seq_len)},

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

@@ -20,10 +20,9 @@ namespace executorch::extension::llm {
 
 class MultimodalPrefiller {
 public:
-  explicit MultimodalPrefiller(Module *module,
-                               MultimodalDecoderRunner *decoder_runner,
-                               tokenizers::HFTokenizer *tokenizer,
-                               IOManager *io_manager,
+  explicit MultimodalPrefiller(Module &module,
+                               MultimodalDecoderRunner &decoder_runner,
+                               tokenizers::HFTokenizer &tokenizer,
                                IEncoder *image_encoder = nullptr);
 
   // Prefill one input segment. Updates start_pos in-place.
@@ -38,7 +37,6 @@ class MultimodalPrefiller {
   Module *module_;
   MultimodalDecoderRunner *decoder_runner_;
   tokenizers::HFTokenizer *tokenizer_;
-  IOManager *io_manager_;
   IEncoder *image_encoder_;
 };
 

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

@@ -26,13 +26,16 @@ int32_t MultimodalRunner::get_visual_token_count() const {
 }
 
 bool MultimodalRunner::is_loaded() const {
-  if (!mm_prefiller_ || !mm_token_generator_)
+  if (!mm_prefiller_ || !mm_token_generator_) {
     return false;
-  if (!mm_prefiller_->is_method_loaded() || !mm_token_generator_->is_loaded())
+  }
+  if (!mm_prefiller_->is_method_loaded() || !mm_token_generator_->is_loaded()) {
     return false;
+  }
   for (const auto &[type, encoder] : encoders_) {
-    if (!encoder->is_loaded())
+    if (!encoder->is_loaded()) {
       return false;
+    }
   }
   return true;
 }
@@ -52,8 +55,7 @@ Error MultimodalRunner::load_subcomponents() {
     image_encoder = enc_it->second.get();
   }
   mm_prefiller_ = std::make_unique<MultimodalPrefiller>(
-      module_.get(), mm_decoder_runner_.get(), tokenizer_.get(),
-      io_manager_.get(), image_encoder);
+      *module_, *mm_decoder_runner_, *tokenizer_, image_encoder);
   mm_token_generator_ = std::make_unique<TextTokenGenerator>(
       tokenizer_.get(), mm_decoder_runner_.get(), /*use_kv_cache=*/true,
       std::move(eos_ids_), stats_ptr);
@@ -68,10 +70,12 @@ Error MultimodalRunner::generate_internal(
     const std::vector<MultimodalInput> &inputs,
     std::function<void(const std::string &)> token_callback) {
 
-  if (inputs.empty())
+  if (inputs.empty()) {
     return Error::InvalidArgument;
-  if (!is_loaded())
+  }
+  if (!is_loaded()) {
     ET_CHECK_OK_OR_RETURN_ERROR(load());
+  }
 
   stats_.inference_start_ms = time_in_ms();
 
@@ -116,18 +120,21 @@ Error MultimodalRunner::generate_internal(
 }
 
 void MultimodalRunner::stop_impl() {
-  if (mm_token_generator_)
+  if (mm_token_generator_) {
     mm_token_generator_->stop();
+  }
 }
 
 void MultimodalRunner::set_count_interval_impl(size_t count_interval) {
-  if (mm_token_generator_)
+  if (mm_token_generator_) {
     mm_token_generator_->set_count_interval(count_interval);
+  }
 }
 
 void MultimodalRunner::set_time_interval_impl(size_t time_interval) {
-  if (mm_token_generator_)
+  if (mm_token_generator_) {
     mm_token_generator_->set_time_interval(time_interval);
+  }
 }
 
 } // namespace executorch::extension::llm

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

@@ -4,7 +4,6 @@
 #include "util.h"
 #include <cstdint>
 #include <rnexecutorch/Error.h>
-#include <rnexecutorch/Log.h>
 
 namespace executorch::extension::llm {
 
@@ -140,13 +139,15 @@ void TextRunner::set_topp_impl(float topp) {
 }
 
 void TextRunner::set_count_interval_impl(size_t count_interval) {
-  if (text_token_generator_)
+  if (text_token_generator_) {
     text_token_generator_->set_count_interval(count_interval);
+  }
 }
 
 void TextRunner::set_time_interval_impl(size_t time_interval) {
-  if (text_token_generator_)
+  if (text_token_generator_) {
     text_token_generator_->set_time_interval(time_interval);
+  }
 }
 
 } // namespace executorch::extension::llm