react-native-executorch 0.5.15 → 0.6.0-nightly-897eae9-20251213

package/common/runner/text_decoder_runner.h

@@ -10,20 +10,17 @@
 
 #pragma once
 
+#include "io_manager.h"
 #include "sampler.h"
-#include <executorch/extension/module/module.h>
-#include <executorch/extension/tensor/tensor.h>
-#include <executorch/runtime/platform/compiler.h>
-#include <functional>
 
 namespace executorch {
 namespace extension {
 namespace llm {
 
-class ET_EXPERIMENTAL TextDecoderRunner {
+class TextDecoderRunner {
 public:
-  TextDecoderRunner(Module *module, bool use_kv_cache, int32_t vocab_size,
-                    float temperature);
+  explicit TextDecoderRunner(Module *module, IOManager *io_manager,
+                             float temperature = 0.8F, float topp = 0.9F);
 
   virtual ~TextDecoderRunner() = default;
 
@@ -35,7 +32,7 @@ public:
    * @return The output of the LLM Module. This will be a tensor of logits.
    */
   virtual ::executorch::runtime::Result<executorch::aten::Tensor>
-  step(TensorPtr &input, TensorPtr &start_pos);
+  step(TensorPtr &input, int64_t start_pos);
 
   /**
    * Load the Module for text decode purpose.
@@ -53,43 +50,67 @@ public:
     return module_->is_method_loaded("forward");
   }
 
+  virtual void set_temperature(float temperature) noexcept {
+    temperature_ = temperature;
+  }
+
+  virtual void set_topp(float topp) noexcept { topp_ = topp; }
+
   inline void stop() { should_stop_ = true; }
 
   /**
    * Sample the next token from the logits tensor.
    * @param logits_tensor The logits tensor.
+   * @param temperature The temperature parameter used to control randomness in
+   * sampling.
    * @return The next token.
    */
-  inline int32_t
-  logits_to_token(const executorch::aten::Tensor &logits_tensor) {
+  inline int32_t logits_to_token(const executorch::aten::Tensor &logits_tensor,
+                                 float temperature = -1.F, float topp = -1.F) {
     int32_t result = 0;
-    ET_SWITCH_THREE_TYPES(Float, Half, BFloat16, logits_tensor.scalar_type(),
-                          unused, "logits_to_token", CTYPE, [&]() {
-                            // If the logit_tensor rank is 3, the shape is
-                            // [batch, seq_length, vocab_size], get the last
-                            // logits, sample and return. Else the model outputs
-                            // the last logit, directly sample and return.
-                            auto *logits =
-                                logits_tensor.mutable_data_ptr<CTYPE>();
-                            if (logits_tensor.dim() == 3) {
-                              auto num_tokens = logits_tensor.size(1);
-                              auto vocab_size = logits_tensor.size(2);
-                              auto *logits_last = logits;
-                              logits_last += (num_tokens - 1) * vocab_size;
-                              result = sampler_->sample(logits_last);
-                            } else {
-                              result = sampler_->sample(logits);
-                            }
-                          });
+
+    temperature = temperature < 0.F ? temperature_ : temperature;
+    topp = topp < 0.F ? topp_ : topp;
+
+    // Create a minimal context for error handling in ET_SWITCH
+    struct {
+      [[noreturn]] void fail(torch::executor::Error /* error */) {
+        ET_CHECK_MSG(false, "Unsupported dtype in logits_to_token");
+      }
+    } ctx;
+
+    ET_SWITCH_FOUR_TYPES(
+        Float, Half, BFloat16, UInt16, logits_tensor.scalar_type(), ctx,
+        "logits_to_token", CTYPE, [&]() {
+          // If the logit_tensor rank is 3, the shape is [batch, seq_length,
+          // vocab_size], get the last logits, sample and return. Else the model
+          // outputs the last logit, directly sample and return.
+          auto *logits = logits_tensor.mutable_data_ptr<CTYPE>();
+          ssize_t vocab_size = logits_tensor.size(logits_tensor.dim() - 1);
+          if (logits_tensor.dim() == 3) {
+            auto num_tokens = logits_tensor.size(1);
+            logits += (num_tokens - 1) * vocab_size;
+          }
+          // @lint-ignore CLANGTIDY facebook-hte-Deprecated
+          Sampler sampler(vocab_size, temperature, topp);
+          result = sampler.sample(logits);
+        });
     return result;
   }
 
 protected:
-  // TODO: use shared_ptr for module
+  /**
+   * Note: TextDecoderRunner does not own the Module or IOManager instance. It
+   * is expected that the outer class (likely Runner) manages the lifecycle of
+   * them. This means that the responsibility for creating, maintaining, and
+   * destroying the Module lies outside of TextDecoderRunner. Ensure that the
+   * Module remains valid for the duration of TextDecoderRunner's usage.
+   */
   Module *module_;
-  std::unique_ptr<Sampler> sampler_;
-  bool use_kv_cache_;
+  IOManager *io_manager_;
   bool should_stop_{false};
+  float temperature_;
+  float topp_;
 };
 
 } // namespace llm

package/common/runner/text_prefiller.cpp

@@ -10,15 +10,18 @@
 // LLM.
 
 #include "text_prefiller.h"
+#include <algorithm>
 
 namespace executorch {
 namespace extension {
 namespace llm {
 
 TextPrefiller::TextPrefiller(TextDecoderRunner *text_decoder_runner,
-                             bool use_kv_cache, bool enable_parallel_prefill)
+                             bool use_kv_cache, bool enable_parallel_prefill,
+                             int64_t max_seq_len)
     : text_decoder_runner_(text_decoder_runner), use_kv_cache_(use_kv_cache),
-      enable_parallel_prefill_(enable_parallel_prefill) {}
+      enable_parallel_prefill_(enable_parallel_prefill),
+      max_seq_len_(max_seq_len > 0 ? max_seq_len : 128) {}
 
 ::executorch::runtime::Result<uint64_t>
 TextPrefiller::prefill(std::vector<uint64_t> &prompt_tokens,
@@ -27,6 +30,44 @@ TextPrefiller::prefill(std::vector<uint64_t> &prompt_tokens,
   if (!text_decoder_runner_->is_method_loaded()) {
     ET_CHECK_OK_OR_RETURN_ERROR(text_decoder_runner_->load());
   }
+
+  // Check if we need to chunk the prompt tokens
+  int32_t num_prompt_tokens = prompt_tokens.size();
+
+  // If prompt tokens exceed max_seq_len_, we need to chunk them
+  if (num_prompt_tokens > max_seq_len_) {
+    uint64_t cur_token = 0;
+    int num_tokens_to_process = 0;
+
+    while (num_tokens_to_process < num_prompt_tokens) {
+      auto num_tokens_to_prefill_with = std::min<int>(
+          num_prompt_tokens - num_tokens_to_process, max_seq_len_);
+
+      std::vector<uint64_t> prompt_tokens_to_process(
+          num_tokens_to_prefill_with);
+      std::copy(prompt_tokens.begin() + num_tokens_to_process,
+                prompt_tokens.begin() + num_tokens_to_process +
+                    num_tokens_to_prefill_with,
+                prompt_tokens_to_process.begin());
+
+      // Process this chunk
+      auto chunk_result = prefill_chunk(prompt_tokens_to_process, start_pos);
+      ET_CHECK_OK_OR_RETURN_ERROR(chunk_result.error());
+      cur_token = chunk_result.get();
+
+      num_tokens_to_process += num_tokens_to_prefill_with;
+    }
+
+    return cur_token;
+  } else {
+    // If prompt tokens don't exceed max_seq_len_, process them directly
+    return prefill_chunk(prompt_tokens, start_pos);
+  }
+}
+
+::executorch::runtime::Result<uint64_t>
+TextPrefiller::prefill_chunk(std::vector<uint64_t> &prompt_tokens,
+                             int64_t &start_pos) {
   // enable_parallel_prefill_ maybe set even when not using kv cache
   // When kv cache is not used, start pos is ignored
   int32_t num_prompt_tokens = prompt_tokens.size();
@@ -38,10 +79,7 @@ TextPrefiller::prefill(std::vector<uint64_t> &prompt_tokens,
     auto tokens = from_blob(prompt_tokens.data(), {1, num_prompt_tokens},
                             executorch::aten::ScalarType::Long);
 
-    auto start_pos_tensor =
-        from_blob(&start_pos, {1}, executorch::aten::ScalarType::Long);
-
-    auto outputs_res = text_decoder_runner_->step(tokens, start_pos_tensor);
+    auto outputs_res = text_decoder_runner_->step(tokens, start_pos);
 
     ET_CHECK_OK_OR_RETURN_ERROR(outputs_res.error());
     ET_LOG(Info, "Prefill token result numel(): %zu",
@@ -58,13 +96,10 @@ TextPrefiller::prefill(std::vector<uint64_t> &prompt_tokens,
     auto tokens =
         from_blob(&cur_token, {1, 1}, executorch::aten::ScalarType::Long);
 
-    auto start_pos_tensor =
-        from_blob(&start_pos, {1}, executorch::aten::ScalarType::Long);
-
     // run the first token and get back logits tensor. Assuming the first token
     // is bos so don't callback.
     auto logits_tensor =
-        ET_UNWRAP(text_decoder_runner_->step(tokens, start_pos_tensor));
+        ET_UNWRAP(text_decoder_runner_->step(tokens, start_pos));
 
     pos += 1; // start the loop from index 1
     start_pos += 1;
@@ -74,8 +109,7 @@ TextPrefiller::prefill(std::vector<uint64_t> &prompt_tokens,
       // NOLINTNEXTLINE(facebook-hte-ParameterUncheckedArrayBounds)
       cur_token = prompt_tokens[pos];
 
-      logits_tensor =
-          ET_UNWRAP(text_decoder_runner_->step(tokens, start_pos_tensor));
+      logits_tensor = ET_UNWRAP(text_decoder_runner_->step(tokens, start_pos));
 
       pos++;
       start_pos++;

package/common/runner/text_prefiller.h

@@ -17,10 +17,12 @@ namespace executorch {
 namespace extension {
 namespace llm {
 
-class ET_EXPERIMENTAL TextPrefiller {
+class TextPrefiller {
 public:
-  TextPrefiller(TextDecoderRunner *text_decoder_runner, bool use_kv_cache_,
-                bool enable_parallel_prefill);
+  TextPrefiller(TextDecoderRunner *text_decoder_runner, bool use_kv_cache,
+                bool enable_parallel_prefill, int64_t max_seq_len = 128);
+
+  virtual ~TextPrefiller() = default;
   /**
    * Prefill an LLM Module with the given text input.
    * @param prompt_tokens The text prompt tokens to the LLM Module. Encoded by
@@ -29,13 +31,45 @@ public:
    * Module.
    * @return The next token of the LLM Module after prefill.
    */
-  ::executorch::runtime::Result<uint64_t>
+  virtual ::executorch::runtime::Result<uint64_t>
   prefill(std::vector<uint64_t> &prompt_tokens, int64_t &start_pos);
 
+  /**
+   * Helper method to prefill a chunk of tokens.
+   * @param prompt_tokens The chunk of text prompt tokens to process.
+   * @param start_pos The starting position in KV cache of the input in the LLM
+   * Module.
+   * @return The next token of the LLM Module after prefilling this chunk.
+   */
+  virtual ::executorch::runtime::Result<uint64_t>
+  prefill_chunk(std::vector<uint64_t> &prompt_tokens, int64_t &start_pos);
+
+  /**
+   * Load the necessary resources for the TextPrefiller.
+   * This method should be called before using the prefill methods.
+   */
+  ::executorch::runtime::Error load() { return text_decoder_runner_->load(); }
+
+  /**
+   * Check if the TextPrefiller has been successfully loaded.
+   * @return True if the resources are loaded, false otherwise.
+   */
+  bool inline is_loaded() const {
+    // Implementation to check if resources are loaded
+    return text_decoder_runner_->is_method_loaded();
+  }
+
 private:
+  /**
+   * Note: TextPrefiller does not own the TextDecoderRunner instance.
+   * The responsibility of managing the lifecycle of TextDecoderRunner
+   * lies with the outer class or entity (likely Runner) that creates
+   * and passes the TextDecoderRunner instance to TextPrefiller.
+   */
   TextDecoderRunner *text_decoder_runner_;
   bool use_kv_cache_;
   bool enable_parallel_prefill_;
+  int64_t max_seq_len_;
 };
 
 } // namespace llm

package/common/runner/text_token_generator.h

@@ -11,39 +11,40 @@
 
 #include "stats.h"
 #include "text_decoder_runner.h"
-#include <chrono>
+#include "util.h"
 #include <executorch/extension/tensor/tensor.h>
-#include <iostream>
 #include <tokenizers-cpp/tokenizers_cpp.h>
 
 namespace executorch {
 namespace extension {
 namespace llm {
 
-class ET_EXPERIMENTAL TextTokenGenerator {
+class TextTokenGenerator {
 public:
-  TextTokenGenerator(tokenizers::Tokenizer *tokenizer,
+  TextTokenGenerator(::tokenizers::Tokenizer *tokenizer,
                      TextDecoderRunner *text_decoder_runner, bool use_kv_cache,
                      std::unique_ptr<std::unordered_set<uint64_t>> &&eos_ids,
                      Stats *stats)
       : tokenizer_(tokenizer), text_decoder_runner_(text_decoder_runner),
         eos_ids_(std::move(eos_ids)), use_kv_cache_(use_kv_cache),
-        stats_(stats), timestamp_(std::chrono::high_resolution_clock::now()) {}
+        timestamp_(std::chrono::high_resolution_clock::now()), stats_(stats) {}
+
+  virtual ~TextTokenGenerator() = default;
 
   /**
    * Token generation loop.
-   * @param tokens prompt tokens as well as the first token generated by
-   * prefill.
-   * @param start_pos the start position of the new tokens, based on how many
+   * @param tokens The first token generated by prefill, if using kv cache. Else
+   * the prompt tokens + the first token generated by prefill.
+   * @param start_pos The start position of the new tokens, based on how many
    * prompt tokens is prefilled.
-   * @param seq_len the total sequence length, including the prompt tokens, next
-   * token from prefill and new tokens.
+   * @param max_new_tokens Maximum number of new tokens to generate.
    * @param token_callback what to do after a token is generated.
    * @return how many tokens are generated.
    */
-  inline ::executorch::runtime::Result<int64_t>
-  generate(std::vector<uint64_t> tokens, int64_t start_pos, int32_t seq_len,
-           std::function<void(const std::string &)> token_callback) {
+  inline ::executorch::runtime::Result<int64_t> generate(
+      std::vector<uint64_t> tokens, int64_t start_pos, int32_t max_new_tokens,
+      float temperature, float topp,
+      const std::function<void(const std::string &)> &token_callback = {}) {
     ET_CHECK_MSG(!tokens.empty(),
                  "Token generation loop shouldn't take empty tokens");
     int64_t pos = start_pos; // position in the sequence
@@ -53,7 +54,8 @@ public:
 
     // Token after prefill
     uint64_t cur_token = tokens.back();
-    uint64_t prev_token;
+    [[maybe_unused]] uint64_t prev_token;
+
     // cache to keep tokens if they were decoded into illegal character
     std::vector<int32_t> token_cache;
     // add first token after prefill to cache here
@@ -72,16 +74,14 @@ public:
     // initialize tensor wrappers
     auto tokens_managed = from_blob(token_data.data(), token_shape,
                                     executorch::aten::ScalarType::Long);
-    auto start_pos_managed =
-        from_blob(&pos, {1}, executorch::aten::ScalarType::Long);
 
     should_stop_ = false;
     timestamp_ = std::chrono::high_resolution_clock::now();
+
     // Generate our tokens
-    while (pos < seq_len - 1) {
+    while (pos < start_pos + max_new_tokens) {
       // Run the model
-      auto logits_res =
-          text_decoder_runner_->step(tokens_managed, start_pos_managed);
+      auto logits_res = text_decoder_runner_->step(tokens_managed, pos);
 
       ET_CHECK_OK_OR_RETURN_ERROR(logits_res.error());
       executorch::aten::Tensor &logits_tensor = logits_res.get();
@@ -89,7 +89,8 @@ public:
       prev_token = cur_token;
 
       stats_->on_sampling_begin();
-      cur_token = text_decoder_runner_->logits_to_token(logits_tensor);
+      cur_token = text_decoder_runner_->logits_to_token(logits_tensor,
+                                                        temperature, topp);
       stats_->on_sampling_end();
 
       pos++;
@@ -105,8 +106,9 @@ public:
             tokens_managed, {1, static_cast<int>(token_data.size())}));
       }
 
-      // print the tokens as string, decode it with the Tokenizer object
       token_cache.push_back(static_cast<int32_t>(cur_token));
+
+      // print the token as string, decode it with the Tokenizer object
       const std::string cache_decoded = tokenizer_->Decode(token_cache);
 
       const auto timeIntervalElapsed =
@@ -129,7 +131,7 @@ public:
       }
 
       // data-dependent terminating condition: we have n_eos_ number of EOS
-      if (eos_reached) {
+      if (eos_ids_->find(cur_token) != eos_ids_->end()) {
         printf("\n");
         ET_LOG(Info, "\nReached to the end of generation");
         break;
@@ -143,6 +145,23 @@ public:
    */
   inline void stop() { should_stop_ = true; }
 
+  /**
+   * Load the necessary resources for TextTokenGenerator.
+   * This method should be called before using the generate() method.
+   */
+  ::executorch::runtime::Error load() { return text_decoder_runner_->load(); }
+
+  /**
+   * Check if the TextTokenGenerator has been successfully loaded.
+   * @return True if the resources are loaded, false otherwise.
+   */
+  bool inline is_loaded() const {
+    // Implementation to check if resources are loaded
+    // return tokenizer_->is_loaded() &&
+    // text_decoder_runner_->is_method_loaded();
+    return text_decoder_runner_->is_method_loaded();
+  }
+
   void set_count_interval(size_t count_interval) {
     count_interval_ = count_interval;
   }
@@ -152,16 +171,22 @@ public:
   }
 
 private:
-  tokenizers::Tokenizer *tokenizer_;
+  /**
+   * Note: TextTokenGenerator does not own the tokenizer_ and
+   * text_decoder_runner_. The lifecycle of these objects should be managed
+   * externally, likely in the Runner. This class assumes that the provided
+   * pointers remain valid for the duration of its use.
+   */
+  ::tokenizers::Tokenizer *tokenizer_;
   TextDecoderRunner *text_decoder_runner_;
   std::unique_ptr<std::unordered_set<uint64_t>> eos_ids_;
   bool use_kv_cache_;
-  size_t count_interval_{10};
-  std::chrono::milliseconds time_interval_{120};
-  std::chrono::high_resolution_clock::time_point timestamp_;
 
   // state machine
   bool should_stop_ = false;
+  size_t count_interval_{10};
+  std::chrono::milliseconds time_interval_{120};
+  std::chrono::high_resolution_clock::time_point timestamp_;
 
   // stats
   Stats *stats_;

package/common/runner/util.h

@@ -7,7 +7,10 @@
  */
 
 #pragma once
+#include "constants.h"
+#include "text_prefiller.h"
 #include <cctype>
+#include <executorch/extension/tensor/tensor.h>
 #include <executorch/runtime/platform/compiler.h>
 #include <stdio.h>
 #include <time.h>
@@ -41,7 +44,7 @@ namespace executorch {
 namespace extension {
 namespace llm {
 
-ET_EXPERIMENTAL void inline safe_printf(const char *piece) {
+void inline safe_printf(const char *piece) {
   // piece might be a raw byte token, and we only want to print printable chars
   // or whitespace because some of the other bytes can be various control codes,
   // backspace, etc.
@@ -63,10 +66,16 @@ ET_EXPERIMENTAL void inline safe_printf(const char *piece) {
 // ----------------------------------------------------------------------------
 // utilities: time
 
-ET_EXPERIMENTAL long inline time_in_ms() {
+long inline time_in_ms() {
   // return time in milliseconds, for benchmarking the model speed
   struct timespec time;
+  // The `timespec_get` function is for windows time access. Some AOSP OS does
+  // not have timespec_get support.
+#if defined(__ANDROID_API__)
   clock_gettime(CLOCK_REALTIME, &time);
+#else
+  timespec_get(&time, TIME_UTC);
+#endif
   return time.tv_sec * 1000 + time.tv_nsec / 1000000;
 }
 
@@ -77,20 +86,56 @@ ET_EXPERIMENTAL long inline time_in_ms() {
 // RSS: Resident Set Size, the amount of memory currently in the RAM for this
 // process. These values are approximate, and are only used for logging
 // purposes.
-ET_EXPERIMENTAL size_t inline get_rss_bytes() {
+size_t inline get_rss_bytes() {
 #if defined(__linux__) || defined(__ANDROID__) || defined(__unix__)
   struct rusage r_usage;
   if (getrusage(RUSAGE_SELF, &r_usage) == 0) {
     return r_usage.ru_maxrss * 1024;
   }
 #endif // __linux__ || __ANDROID__ || __unix__
-       // Unsupported platform like Windows, or getrusage() failed.
-       // __APPLE__ and __MACH__ are not supported because r_usage.ru_maxrss
-       // does not consistently return kbytes on macOS. On older versions of
-       // macOS, it returns bytes, but on newer versions it returns kbytes. Need
-       // to figure out when this changed.
+  // Unsupported platform like Windows, or getrusage() failed.
+  // __APPLE__ and __MACH__ are not supported because r_usage.ru_maxrss does not
+  // consistently return kbytes on macOS. On older versions of macOS, it
+  // returns bytes, but on newer versions it returns kbytes. Need to figure out
+  // when this changed.
   return 0;
 }
+
+// Returns the cache position tensor, which can be either a single start_pos
+// (when the method_name [`text_decoder` or `forward`] expects a tensor with
+// size 1 because model will populate the cache position tensor underneath), or
+// a populated tensor for cache position, for the given start_pos and seq_len.
+inline runtime::Result<TensorPtr>
+populate_start_pos_or_cache_position(Module *module, int64_t &start_pos,
+                                     std::vector<int64_t> &cache_positions_vec,
+                                     int seq_len,
+                                     const char *method_name = "forward") {
+  // Get expected shape of cache position tensor, which should be the second
+  // argument
+  auto method_meta = ET_UNWRAP(module->method_meta(method_name));
+  auto second_input_info = ET_UNWRAP(method_meta.input_tensor_meta(1));
+  auto second_input_sizes = second_input_info.sizes();
+  auto numel = second_input_sizes[0];
+
+  TensorPtr start_pos_tensor;
+  if (numel > 1) {
+    // `cache_position` goes from start_pos to start_pos +
+    // encoder_output.size(1). e.g. if start_pos = 2 and encoder_output.size(1)
+    // = 5, cache_position_tensor should be [2, 3, 4, 5, 6].
+    cache_positions_vec.resize(seq_len);
+    for (int64_t i = 0; i < seq_len; ++i) {
+      cache_positions_vec[i] = start_pos + i;
+    }
+    return ::executorch::extension::from_blob(
+        cache_positions_vec.data(), {static_cast<int>(seq_len)},
+        executorch::aten::ScalarType::Long);
+  } else {
+    // Cache position is size 1.
+    return ::executorch::extension::from_blob(
+        &start_pos, {1}, executorch::aten::ScalarType::Long);
+  }
+}
+
 } // namespace llm
 } // namespace extension
 } // namespace executorch

package/ios/RnExecutorch.xcodeproj/project.pbxproj

@@ -7,7 +7,6 @@
 	objects = {
 
 /* Begin PBXBuildFile section */
-		55D6EA8C2D0987D2009BA408 /* ExecutorchLib.xcframework in Frameworks */ = {isa = PBXBuildFile; fileRef = 55D6EA8B2D0987D2009BA408 /* ExecutorchLib.xcframework */; };
 		8C9A9BD02DB0CE800027DD32 /* ImageSegmentation.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 8C9A9BCA2DB0CE800027DD32 /* ImageSegmentation.cpp */; };
 		8C9A9BD12DB0CE800027DD32 /* RuntimeLifecycleMonitor.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 8C9A9BC72DB0CE800027DD32 /* RuntimeLifecycleMonitor.cpp */; };
 		8C9A9BD22DB0CE800027DD32 /* JsiHostObject.cpp in Sources */ = {isa = PBXBuildFile; fileRef = 8C9A9BC22DB0CE800027DD32 /* JsiHostObject.cpp */; };
@@ -43,30 +42,9 @@
 		8C9A9BCD2DB0CE800027DD32 /* RnExecutorchInstaller.h */ = {isa = PBXFileReference; lastKnownFileType = sourcecode.c.h; path = RnExecutorchInstaller.h; sourceTree = "<group>"; };
 /* End PBXFileReference section */
 
-/* Begin PBXFileSystemSynchronizedGroupBuildPhaseMembershipExceptionSet section */
-		550986902CEF541900FECBB8 /* Exceptions for "RnExecutorch" folder in "Copy Files" phase from "RnExecutorch" target */ = {
-			isa = PBXFileSystemSynchronizedGroupBuildPhaseMembershipExceptionSet;
-			buildPhase = 550986872CEF541900FECBB8 /* CopyFiles */;
-			membershipExceptions = (
-				LLM.h,
-			);
-		};
-		552754CC2D394AC9006B38A2 /* Exceptions for "RnExecutorch" folder in "Compile Sources" phase from "RnExecutorch" target */ = {
-			isa = PBXFileSystemSynchronizedGroupBuildPhaseMembershipExceptionSet;
-			buildPhase = 550986852CEF541900FECBB8 /* Sources */;
-			membershipExceptions = (
-				models/ocr/utils/DetectorUtils.h,
-			);
-		};
-/* End PBXFileSystemSynchronizedGroupBuildPhaseMembershipExceptionSet section */
-
 /* Begin PBXFileSystemSynchronizedRootGroup section */
 		5509868B2CEF541900FECBB8 /* RnExecutorch */ = {
 			isa = PBXFileSystemSynchronizedRootGroup;
-			exceptions = (
-				552754CC2D394AC9006B38A2 /* Exceptions for "RnExecutorch" folder in "Compile Sources" phase from "RnExecutorch" target */,
-				550986902CEF541900FECBB8 /* Exceptions for "RnExecutorch" folder in "Copy Files" phase from "RnExecutorch" target */,
-			);
 			path = RnExecutorch;
 			sourceTree = "<group>";
 		};
@@ -77,7 +55,6 @@
 			isa = PBXFrameworksBuildPhase;
 			buildActionMask = 2147483647;
 			files = (
-				55D6EA8C2D0987D2009BA408 /* ExecutorchLib.xcframework in Frameworks */,
 			);
 			runOnlyForDeploymentPostprocessing = 0;
 		};

package/lib/module/Error.js

@@ -7,6 +7,7 @@ export let ETError = /*#__PURE__*/function (ETError) {
   ETError[ETError["FileWriteFailed"] = 103] = "FileWriteFailed";
   ETError[ETError["ModelGenerating"] = 104] = "ModelGenerating";
   ETError[ETError["LanguageNotSupported"] = 105] = "LanguageNotSupported";
+  ETError[ETError["InvalidConfig"] = 112] = "InvalidConfig";
   ETError[ETError["InvalidModelSource"] = 255] = "InvalidModelSource";
   // SpeechToText errors
   ETError[ETError["MultilingualConfiguration"] = 160] = "MultilingualConfiguration";

package/lib/module/Error.js.map

@@ -1 +1 @@
-{"version":3,"names":["ETError","getError","e","UndefinedError","error","errorCode","parseInt","message","Number","isNaN"],"sourceRoot":"../../src","sources":["Error.ts"],"mappings":";;AAAA,WAAYA,OAAO,0BAAPA,OAAO;EACjB;EADUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EASjB;EATUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAcjB;EACA;EACA;EAhBUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAsBjB;EAtBUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EA6BjB;EA7BUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAqCjB;EArCUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAA,OAAPA,OAAO;AAAA;AA2CnB,OAAO,MAAMC,QAAQ,GAAIC,CAA4B,IAAa;EAChE,IAAI,OAAOA,CAAC,KAAK,QAAQ,EAAE;IACzB,OAAOF,OAAO,CAACE,CAAC,CAAC,IAAIF,OAAO,CAACA,OAAO,CAACG,cAAc,CAAC;EACtD;;EAEA;EACA,MAAMC,KAAK,GAAGF,CAAU;EACxB,MAAMG,SAAS,GAAGC,QAAQ,CAACF,KAAK,CAACG,OAAO,EAAE,EAAE,CAAC;EAE7C,IAAIC,MAAM,CAACC,KAAK,CAACJ,SAAS,CAAC,EAAE;IAC3B,OAAOD,KAAK,CAACG,OAAO;EACtB;EAEA,OAAOP,OAAO,CAACK,SAAS,CAAC,IAAIL,OAAO,CAACA,OAAO,CAACG,cAAc,CAAC;AAC9D,CAAC","ignoreList":[]}
+{"version":3,"names":["ETError","getError","e","UndefinedError","error","errorCode","parseInt","message","Number","isNaN"],"sourceRoot":"../../src","sources":["Error.ts"],"mappings":";;AAAA,WAAYA,OAAO,0BAAPA,OAAO;EACjB;EADUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAUjB;EAVUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAejB;EACA;EACA;EAjBUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAuBjB;EAvBUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EA8BjB;EA9BUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAsCjB;EAtCUA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAPA,OAAO,CAAPA,OAAO;EAAA,OAAPA,OAAO;AAAA;AA4CnB,OAAO,MAAMC,QAAQ,GAAIC,CAA4B,IAAa;EAChE,IAAI,OAAOA,CAAC,KAAK,QAAQ,EAAE;IACzB,OAAOF,OAAO,CAACE,CAAC,CAAC,IAAIF,OAAO,CAACA,OAAO,CAACG,cAAc,CAAC;EACtD;;EAEA;EACA,MAAMC,KAAK,GAAGF,CAAU;EACxB,MAAMG,SAAS,GAAGC,QAAQ,CAACF,KAAK,CAACG,OAAO,EAAE,EAAE,CAAC;EAE7C,IAAIC,MAAM,CAACC,KAAK,CAACJ,SAAS,CAAC,EAAE;IAC3B,OAAOD,KAAK,CAACG,OAAO;EACtB;EAEA,OAAOP,OAAO,CAACK,SAAS,CAAC,IAAIL,OAAO,CAACA,OAAO,CAACG,cAAc,CAAC;AAC9D,CAAC","ignoreList":[]}

package/lib/module/constants/directories.js

@@ -1,5 +1,5 @@
 "use strict";
 
-import { documentDirectory } from 'expo-file-system';
+import { documentDirectory } from 'expo-file-system/legacy';
 export const RNEDirectory = `${documentDirectory}react-native-executorch/`;
 //# sourceMappingURL=directories.js.map

package/lib/module/constants/directories.js.map

@@ -1 +1 @@
-{"version":3,"names":["documentDirectory","RNEDirectory"],"sourceRoot":"../../../src","sources":["constants/directories.ts"],"mappings":";;AAAA,SAASA,iBAAiB,QAAQ,kBAAkB;AAEpD,OAAO,MAAMC,YAAY,GAAG,GAAGD,iBAAiB,0BAA0B","ignoreList":[]}
+{"version":3,"names":["documentDirectory","RNEDirectory"],"sourceRoot":"../../../src","sources":["constants/directories.ts"],"mappings":";;AAAA,SAASA,iBAAiB,QAAQ,yBAAyB;AAE3D,OAAO,MAAMC,YAAY,GAAG,GAAGD,iBAAiB,0BAA0B","ignoreList":[]}