react-native-executorch 0.10.2 → 0.10.3

package/android/build.gradle.kts

@@ -144,8 +144,13 @@ android {
         getByName("main") {
             // Prebuilt executorch + backend .so live under
             // third-party/android/libs/executorch/<abi>/. Pointing jniLibs here
-            // makes Gradle package them (libexecutorch.so + the enabled
-            // lib*_executorch_backend.so) into the APK alongside our own .so.
+            // makes Gradle package them into the APK alongside our own .so.
+            //
+            // Gradle packages every .so it finds under a jniLibs source dir, so
+            // this alone ships backends the app opted out of: -DRNE_ENABLE_* only
+            // stops CMake importing and linking them, which is enough for
+            // behaviour but has no bearing on packaging. `packaging.jniLibs
+            // .excludes` below drops the disabled ones back out.
             jniLibs.srcDirs("../third-party/android/libs/executorch")
             // Include generated codegen if using TurboModules
             java.srcDirs("${project.buildDir}/generated/source/codegen/java")
@@ -160,6 +165,25 @@ android {
     packaging {
         // Prevents "Duplicate Library" errors with the React Native JSI engine
         resources.excludes.add("**/libjsi.so")
+
+        // Ship only the backends the resolved config asks for. Without this an
+        // app declaring `backends: ["vulkan"]` still carries 2.67 MB of XNNPACK
+        // that nothing links, registers or loads, because jniLibs packages the
+        // directory rather than the link graph. This is the Android counterpart
+        // of the podspec deciding its -force_load entries from the same config.
+        //
+        // Keyed off the backend .so names rather than a staging dir so that a
+        // user who provisions third-party/ by hand gets the same treatment.
+        val backendLibs = mapOf(
+            "enableXnnpack" to "libxnnpack_executorch_backend.so",
+            "enableVulkan" to "libvulkan_executorch_backend.so"
+        )
+        for ((flag, soName) in backendLibs) {
+            if (rneConfig[flag] == false) {
+                logger.lifecycle("[RnExecutorch] $flag is off; excluding $soName from the APK")
+                jniLibs.excludes.add("**/$soName")
+            }
+        }
     }
 
     compileOptions {

package/cpp/extensions/llm/llm_runner.cpp

@@ -11,6 +11,7 @@
 #include <algorithm>
 #include <cstring>
 #include <format>
+#include <optional>
 
 #include <executorch/extension/llm/runner/constants.h>
 #include <executorch/extension/llm/runner/llm_runner_helper.h>
@@ -18,6 +19,8 @@
 #include <executorch/extension/llm/runner/multimodal_runner.h>
 #include <executorch/extension/llm/runner/stats.h>
 #include <executorch/extension/llm/runner/text_llm_runner.h>
+#include <executorch/extension/llm/runner/text_prefiller.h>
+#include <executorch/extension/module/module.h>
 #include <executorch/runtime/core/error.h>
 
 #include "core/conversions.h"
@@ -56,8 +59,8 @@ namespace {
 // COUPLING / MAINTENANCE WARNING:
 // This creates a direct compile-time coupling with the internal implementation
 // details of ExecuTorch's `TextLLMRunner` and `MultimodalRunner`. If upstream
-// ExecuTorch renames or alters `pos_`, `prefill_next_token_`, or `metadata_`,
-// this code will fail to compile and must be adjusted accordingly.
+// ExecuTorch renames or alters any member accessed below, this code will fail
+// to compile and must be adjusted accordingly.
 // ============================================================================
 
 template <typename Tag, auto MemberPtr>
@@ -95,6 +98,96 @@ constexpr auto getPrivateMember(MMRunnerPrefillTag /*tag*/);
 template struct PrivateMemberAccessor<MMRunnerMetadataTag, &MultimodalRunner::metadata_>;
 constexpr auto getPrivateMember(MMRunnerMetadataTag /*tag*/);
 
+// Members that decide how a prompt is split into prefill chunks.
+using executorch::extension::llm::TextPrefiller;
+
+struct TextRunnerModuleTag {};
+struct TextRunnerPrefillerTag {};
+struct PrefillerMaxSeqLenTag {};
+struct MMRunnerModuleTag {};
+struct MMRunnerTokenizerTag {};
+
+template struct PrivateMemberAccessor<TextRunnerModuleTag, &TextLLMRunner::module_>;
+constexpr auto getPrivateMember(TextRunnerModuleTag /*tag*/);
+
+template struct PrivateMemberAccessor<TextRunnerPrefillerTag, &TextLLMRunner::text_prefiller_>;
+constexpr auto getPrivateMember(TextRunnerPrefillerTag /*tag*/);
+
+template struct PrivateMemberAccessor<PrefillerMaxSeqLenTag, &TextPrefiller::max_seq_len_>;
+constexpr auto getPrivateMember(PrefillerMaxSeqLenTag /*tag*/);
+
+template struct PrivateMemberAccessor<MMRunnerModuleTag, &MultimodalRunner::module_>;
+constexpr auto getPrivateMember(MMRunnerModuleTag /*tag*/);
+
+template struct PrivateMemberAccessor<MMRunnerTokenizerTag, &MultimodalRunner::tokenizer_>;
+constexpr auto getPrivateMember(MMRunnerTokenizerTag /*tag*/);
+
+// Upstream sizes prefill chunks from `get_max_seq_len`, which is the KV budget,
+// not the widest token input the graph accepts (e.g. gemma4 MLX: 2048 vs 511).
+// Prompts past the real bound then fail with `Error::NotSupported`. Returns the
+// bound declared by the method's first input, or nullopt if it is not a
+// rank>=2 integer tensor.
+std::optional<int64_t> readPrefillBound(executorch::extension::Module *module, const std::string &methodName) {
+    if (module == nullptr) {
+        return std::nullopt;
+    }
+    auto meta = module->method_meta(methodName);
+    if (!meta.ok() || meta->num_inputs() == 0) {
+        return std::nullopt;
+    }
+    auto inputMeta = meta->input_tensor_meta(0);
+    if (!inputMeta.ok()) {
+        return std::nullopt;
+    }
+    const auto dtype = inputMeta->scalar_type();
+    if (dtype != executorch::aten::ScalarType::Long && dtype != executorch::aten::ScalarType::Int) {
+        return std::nullopt;
+    }
+    const auto sizes = inputMeta->sizes();
+    if (sizes.size() < 2) {
+        return std::nullopt;
+    }
+    const auto bound = static_cast<int64_t>(sizes[sizes.size() - 1]);
+    if (bound <= 0) {
+        return std::nullopt;
+    }
+    return bound;
+}
+
+::tokenizers::Tokenizer *getRunnerTokenizer(executorch::extension::llm::IRunner *runner) {
+    auto *r = dynamic_cast<MultimodalRunner *>(runner);
+    if (r == nullptr) {
+        return nullptr;
+    }
+    return (r->*getPrivateMember(MMRunnerTokenizerTag{})).get();
+}
+
+// `MultimodalPrefiller` does not chunk, so split text into token inputs that fit
+// the bound; `MultimodalRunner` prefills them in order.
+void appendTextInput(std::vector<executorch::extension::llm::MultimodalInput> &inputs,
+                     std::string text,
+                     ::tokenizers::Tokenizer *tokenizer,
+                     int64_t prefillBound) {
+    if (tokenizer == nullptr || prefillBound <= 0) {
+        inputs.emplace_back(std::move(text));
+        return;
+    }
+
+    auto encoded = tokenizer->encode(text, 0, 0);
+    if (!encoded.ok()) {
+        inputs.emplace_back(std::move(text));
+        return;
+    }
+
+    const auto &tokens = encoded.get();
+    const auto chunk = static_cast<size_t>(prefillBound);
+    for (size_t offset = 0; offset < tokens.size(); offset += chunk) {
+        const auto end = std::min(offset + chunk, tokens.size());
+        inputs.emplace_back(std::vector<uint64_t>(tokens.begin() + static_cast<ptrdiff_t>(offset),
+                                                  tokens.begin() + static_cast<ptrdiff_t>(end)));
+    }
+}
+
 int64_t getRunnerPos(executorch::extension::llm::IRunner *runner, bool isMultimodal) {
     if (runner == nullptr) {
         return 0;
@@ -178,10 +271,14 @@ std::vector<executorch::extension::llm::MultimodalInput> parsePrompt(
     jsi::Runtime &rt,
     const std::string &ctx,
     const jsi::Value &value,
-    const std::vector<std::string> &supportedModalities) {
+    const std::vector<std::string> &supportedModalities,
+    ::tokenizers::Tokenizer *tokenizer,
+    int64_t prefillBound) {
 
     if (value.isString()) {
-        return {executorch::extension::llm::MultimodalInput(conversions::asType<std::string>(rt, ctx, value))};
+        std::vector<executorch::extension::llm::MultimodalInput> inputs;
+        appendTextInput(inputs, conversions::asType<std::string>(rt, ctx, value), tokenizer, prefillBound);
+        return inputs;
     }
 
     auto arr = conversions::asType<jsi::Array>(rt, ctx, value);
@@ -195,7 +292,7 @@ std::vector<executorch::extension::llm::MultimodalInput> parsePrompt(
         std::string itemCtx = std::format("{}[{}]", ctx, i);
 
         if (elem.isString()) {
-            inputs.emplace_back(conversions::asType<std::string>(rt, itemCtx, elem));
+            appendTextInput(inputs, conversions::asType<std::string>(rt, itemCtx, elem), tokenizer, prefillBound);
             continue;
         }
         auto mediaObj = conversions::asType<jsi::Object>(rt, itemCtx, elem);
@@ -254,10 +351,16 @@ LLMRunnerHostObject::LLMRunnerHostObject(const std::string &modelPath,
         throw error::LoadFailed(std::format("LLMRunner: Failed to load runner tokenizer at path: {}", tokenizerPath));
     }
 
+    // Plain mmap leaves the weights as clean, file-backed pages the kernel
+    // can evict, which is what the legacy binding has always passed.
+    constexpr auto kLoadMode = executorch::extension::Module::LoadMode::Mmap;
+
     if (modalities_.empty()) {
-        runner_ = executorch::extension::llm::create_text_llm_runner(modelPath, std::move(tokenizer));
+        runner_ = executorch::extension::llm::create_text_llm_runner(
+            modelPath, std::move(tokenizer), std::nullopt, -1.0f, "forward", kLoadMode);
     } else {
-        runner_ = executorch::extension::llm::create_multimodal_runner(modelPath, std::move(tokenizer));
+        runner_ = executorch::extension::llm::create_multimodal_runner(
+            modelPath, std::move(tokenizer), std::nullopt, kLoadMode);
     }
 
     if (!runner_) {
@@ -269,6 +372,33 @@ LLMRunnerHostObject::LLMRunnerHostObject(const std::string &modelPath,
         std::string errorMsg = executorch::runtime::to_string(loadStatus);
         throw error::LoadFailed(std::format("LLMRunner: Failed to load model: {}", errorMsg), loadStatus);
     }
+
+    // Clamp prefill chunks to the graph's real bound (see `readPrefillBound`).
+    // Text models take tokens on `forward`, multimodal ones on `token_embedding`.
+    if (modalities_.empty()) {
+        auto *textRunner = dynamic_cast<TextLLMRunner *>(runner_.get());
+        if (textRunner != nullptr) {
+            const auto &module = textRunner->*getPrivateMember(TextRunnerModuleTag{});
+            const auto bound = readPrefillBound(module.get(), "forward");
+            auto &prefiller = textRunner->*getPrivateMember(TextRunnerPrefillerTag{});
+            if (bound.has_value() && prefiller) {
+                // Upstream allows max_seq_len < max_context_len and chunks itself.
+                auto &maxSeqLen = (*prefiller).*getPrivateMember(PrefillerMaxSeqLenTag{});
+                maxSeqLen = std::min(maxSeqLen, *bound);
+                prefillBound_ = maxSeqLen;
+            }
+        }
+    } else {
+        auto *mmRunner = dynamic_cast<MultimodalRunner *>(runner_.get());
+        if (mmRunner != nullptr) {
+            const auto &module = mmRunner->*getPrivateMember(MMRunnerModuleTag{});
+            const auto bound =
+                readPrefillBound(module.get(), executorch::extension::llm::kTokenEmbeddingMethod);
+            if (bound.has_value()) {
+                prefillBound_ = *bound;
+            }
+        }
+    }
 }
 
 std::unique_lock<std::mutex> LLMRunnerHostObject::tryLockUnique(std::string_view ctx) {
@@ -305,11 +435,9 @@ jsi::Value LLMRunnerHostObject::get(jsi::Runtime &rt, const jsi::PropNameID &nam
             }
 
             executorch::extension::llm::GenerationConfig config;
+            config.echo = false;
             if (count > 1 && !args[1].isUndefined() && !args[1].isNull()) {
                 auto configObj = conversions::asType<jsi::Object>(rt, "LLMRunner.generate: config", args[1]);
-                if (auto echoOpt = conversions::getOptionalProperty<bool>(rt, "LLMRunner.generate: config", configObj, "echo")) {
-                    config.echo = *echoOpt;
-                }
                 if (auto ignoreEosOpt = conversions::getOptionalProperty<bool>(rt, "LLMRunner.generate: config", configObj, "ignoreEos")) {
                     config.ignore_eos = *ignoreEosOpt;
                 }
@@ -348,7 +476,8 @@ jsi::Value LLMRunnerHostObject::get(jsi::Runtime &rt, const jsi::PropNameID &nam
                 auto prompt = conversions::asType<std::string>(rt, "LLMRunner.generate: prompt", args[0]);
                 genStatus = self->runner_->generate(prompt, config, tokenCallback, statsCallback);
             } else {
-                auto inputs = parsePrompt(rt, "LLMRunner.generate", args[0], self->modalities_);
+                auto inputs = parsePrompt(rt, "LLMRunner.generate", args[0], self->modalities_,
+                                          getRunnerTokenizer(self->runner_.get()), self->prefillBound_);
                 auto *multimodalRunner = dynamic_cast<executorch::extension::llm::MultimodalRunner *>(self->runner_.get());
                 if (multimodalRunner == nullptr) {
                     throw error::InvalidArgument("LLMRunner.generate: Runner instance is not a multimodal model");
@@ -375,7 +504,8 @@ jsi::Value LLMRunnerHostObject::get(jsi::Runtime &rt, const jsi::PropNameID &nam
 
             auto lock = self->tryLockUnique("LLMRunner.prefill");
 
-            auto inputs = parsePrompt(rt, "LLMRunner.prefill", args[0], self->modalities_);
+            auto inputs = parsePrompt(rt, "LLMRunner.prefill", args[0], self->modalities_,
+                                      getRunnerTokenizer(self->runner_.get()), self->prefillBound_);
             auto result = self->runner_->prefill(inputs);
 
             if (!result.ok()) {

package/cpp/extensions/llm/llm_runner.h

@@ -8,6 +8,7 @@
 #pragma clang diagnostic ignored "-Wdeprecated-declarations"
 #endif
 
+#include <cstdint>
 #include <memory>
 #include <mutex>
 #include <string>
@@ -77,6 +78,12 @@ private:
     std::string tokenizerPath_;
     /** List of supported non-text input modalities. */
     std::vector<std::string> modalities_;
+    /**
+     * Longest token sequence the loaded graph accepts in one prefill step, read
+     * off the model's own method metadata. 0 when the model does not declare
+     * one, in which case prompts are passed through unsplit.
+     */
+    int64_t prefillBound_ = 0;
 };
 
 void install_createLLMRunner(facebook::jsi::Runtime &rt, facebook::jsi::Object &module);

package/lib/module/extensions/llm/llmRunner.js.map

@@ -1 +1 @@
-{"version":3,"names":["rnexecutorchJsi","createLLMRunner","modelPath","tokenizerPath","modalities","llm"],"sourceRoot":"../../../../src","sources":["extensions/llm/llmRunner.ts"],"mappings":";;AAAA;AACA;AACA;;AAGA,SAASA,eAAe,QAAQ,wBAAqB;;AAIrD;AACA;AACA;AACA;AACA;;AAYA;AACA;AACA;AACA;AACA;;AAoBA;AACA;AACA;AACA;AACA;;AAKA;AACA;AACA;AACA;AACA;;AAGA;AACA;AACA;AACA;AACA;;AAGA;AACA;AACA;AACA;AACA;;AAYA;AACA;AACA;AACA;AACA;AACA;AACA;;AAyDA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,SAASC,eAAeA,CAC7BC,SAAiB,EACjBC,aAAqB,EACrBC,UAAgC,EACrB;EACX,SAAS;;EACT,OAAOJ,eAAe,CAACK,GAAG,CAACJ,eAAe,CAACC,SAAS,EAAEC,aAAa,EAAEC,UAAU,IAAI,EAAE,CAAC;AACxF","ignoreList":[]}
+{"version":3,"names":["rnexecutorchJsi","createLLMRunner","modelPath","tokenizerPath","modalities","llm"],"sourceRoot":"../../../../src","sources":["extensions/llm/llmRunner.ts"],"mappings":";;AAAA;AACA;AACA;;AAGA,SAASA,eAAe,QAAQ,wBAAqB;;AAIrD;AACA;AACA;AACA;AACA;;AAUA;AACA;AACA;AACA;AACA;;AAoBA;AACA;AACA;AACA;AACA;;AAKA;AACA;AACA;AACA;AACA;;AAGA;AACA;AACA;AACA;AACA;;AAGA;AACA;AACA;AACA;AACA;;AAYA;AACA;AACA;AACA;AACA;AACA;AACA;;AAyDA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,SAASC,eAAeA,CAC7BC,SAAiB,EACjBC,aAAqB,EACrBC,UAAgC,EACrB;EACX,SAAS;;EACT,OAAOJ,eAAe,CAACK,GAAG,CAACJ,eAAe,CAACC,SAAS,EAAEC,aAAa,EAAEC,UAAU,IAAI,EAAE,CAAC;AACxF","ignoreList":[]}

package/lib/module/extensions/llm/tasks/llmChatSession.js

@@ -43,13 +43,13 @@ function generateChatTurnWorklet(runner, prompt, options) {
 
   const {
     genConfig,
-    eosToken,
+    stopTokens,
     stopRegex,
     onToken
   } = options;
   let response = '';
   const callback = token => {
-    if (token === eosToken) return;
+    if (stopTokens.includes(token)) return;
     response += token;
     if (onToken) scheduleOnRN(onToken, token);
     if (stopRegex?.test(response)) runner.stop();
@@ -99,7 +99,7 @@ export async function createLLMChatSession(config, options = {}, runtime) {
     const tokenizerConfig = parseTokenizerConfig(JSON.parse(tokenizerConfigStr));
     const {
       chatTemplate,
-      eosToken
+      stopTokens
     } = tokenizerConfig;
 
     // Prepare chat preprocessor
@@ -139,7 +139,7 @@ export async function createLLMChatSession(config, options = {}, runtime) {
       };
       const generationOpts = {
         genConfig: turnGenConfig,
-        eosToken,
+        stopTokens,
         stopRegex,
         onToken
       };

package/lib/module/extensions/llm/tasks/llmChatSession.js.map

@@ -1 +1 @@
-{"version":3,"names":["scheduleOnRN","RNBlobUtil","wrapAsync","createResourceScope","createLLMRunner","parseTokenizerConfig","createChatPreprocessor","generateChatTurnWorklet","runner","prompt","options","genConfig","eosToken","stopRegex","onToken","response","callback","token","test","stop","stats","generate","DEFAULT_MAX_TURNS","createLLMChatSession","config","runtime","scope","dispose","generationConfig","defaultGenerationConfig","initialMessages","toolOpts","resetOnTurn","modelPath","tokenizerPath","tokenizerConfigPath","modalities","preprocessorConfig","tools","parseToolCalls","maxToolTurns","tokenizerConfigStr","fs","readFile","tokenizerConfig","JSON","parse","chatTemplate","chatPreprocessorConfig","chatPreprocessor","track","prefill","history","committed","length","push","process","addGenPrompt","clear","generateChatTurn","sendMessage","message","turnGenConfig","generationOpts","initialCommitted","initialPos","getKVCacheState","pos","turnStartIdx","generationStatsList","role","content","reset","prefillStartMs","Date","now","toCommit","userPrompt","posAtEndOfUser","finishReason","currentTurn","uncommitted","prefillDurationMs","parsedTools","toolCalls","textContent","toolCall","tool","find","t","function","name","toolContent","execute","arguments","err","String","toolCallId","id","messages","slice","getHistory","error"],"sourceRoot":"../../../../../src","sources":["extensions/llm/tasks/llmChatSession.ts"],"mappings":";;AAAA;AACA;AACA;AACA;;AAEA,SAASA,YAAY,QAA6B,uBAAuB;AACzE,OAAOC,UAAU,MAAM,wBAAwB;AAE/C,SAASC,SAAS,QAAQ,0BAAuB;AACjD,SAASC,mBAAmB,QAAQ,2BAAwB;AAC5D,SACEC,eAAe,QAOV,iBAAc;AACrB,SAASC,oBAAoB,QAAQ,6BAA0B;AAC/D,SACEC,sBAAsB,QAIjB,8BAA2B;;AAGlC;AACA;AACA;AACA;;AAcA;AACA;AACA;AACA;;AAUA;AACA;AACA;AACA;;AAiBA;AACA;AACA;AACA;;AAcA;AACA;AACA;AACA;;AAsCA,SAASC,uBAAuBA,CAC9BC,MAAiB,EACjBC,MAAc,EACdC,OAKC,EACkE;EACnE,SAAS;;EACT,MAAM;IAAEC,SAAS;IAAEC,QAAQ;IAAEC,SAAS;IAAEC;EAAQ,CAAC,GAAGJ,OAAO;EAE3D,IAAIK,QAAQ,GAAG,EAAE;EAEjB,MAAMC,QAAQ,GAAIC,KAAa,IAAK;IAClC,IAAIA,KAAK,KAAKL,QAAQ,EAAE;IACxBG,QAAQ,IAAIE,KAAK;IAEjB,IAAIH,OAAO,EAAEd,YAAY,CAACc,OAAO,EAAEG,KAAK,CAAC;IACzC,IAAIJ,SAAS,EAAEK,IAAI,CAACH,QAAQ,CAAC,EAAEP,MAAM,CAACW,IAAI,CAAC,CAAC;EAC9C,CAAC;EAED,MAAMC,KAAK,GAAGZ,MAAM,CAACa,QAAQ,CAACZ,MAAM,EAAEE,SAAS,EAAEK,QAAQ,CAAC;EAE1D,OAAO;IAAED,QAAQ;IAAEK;EAAM,CAAC;AAC5B;AAEA,MAAME,iBAAiB,GAAG,CAAC;;AAE3B;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,eAAeC,oBAAoBA,CACxCC,MAAgB,EAChBd,OAA8B,GAAG,CAAC,CAAC,EACnCe,OAAwB,EACC;EACzB,MAAMC,KAAK,GAAGvB,mBAAmB,CAAC,CAAC;EACnC,MAAMwB,OAAO,GAAGD,KAAK,CAACC,OAAO;EAE7B,IAAI;IACF,MAAM;MACJC,gBAAgB,EAAEC,uBAAuB;MACzCC,eAAe,GAAG,EAAE;MACpBjB,SAAS;MACTkB,QAAQ;MACRC,WAAW,GAAG;IAChB,CAAC,GAAGtB,OAAO;IAEX,MAAM;MAAEuB,SAAS;MAAEC,aAAa;MAAEC,mBAAmB;MAAEC,UAAU;MAAEC;IAAmB,CAAC,GACrFb,MAAM;IACR,MAAM;MAAEc,KAAK;MAAEC,cAAc;MAAEC,YAAY,GAAGlB;IAAkB,CAAC,GAAGS,QAAQ,IAAI,CAAC,CAAC;;IAElF;IACA,MAAMU,kBAAkB,GAAG,MAAMxC,UAAU,CAACyC,EAAE,CAACC,QAAQ,CAACR,mBAAmB,EAAE,MAAM,CAAC;IACpF,MAAMS,eAAe,GAAGvC,oBAAoB,CAACwC,IAAI,CAACC,KAAK,CAACL,kBAAkB,CAAC,CAAC;IAC5E,MAAM;MAAEM,YAAY;MAAEnC;IAAS,CAAC,GAAGgC,eAAe;;IAElD;IACA,MAAMI,sBAAsB,GAAG;MAAED,YAAY;MAAEX,UAAU;MAAEC,kBAAkB;MAAEC;IAAM,CAAC;IACtF,MAAMW,gBAAgB,GAAGvB,KAAK,CAACwB,KAAK,CAAC5C,sBAAsB,CAAC0C,sBAAsB,CAAC,CAAC;;IAEpF;IACA,MAAMxC,MAAM,GAAGkB,KAAK,CAACwB,KAAK,CACxB,MAAMhD,SAAS,CAACE,eAAe,EAAEqB,OAAO,CAAC,CAACQ,SAAS,EAAEC,aAAa,EAAEE,UAAU,CAChF,CAAC;IACD,MAAMe,OAAO,GAAGjD,SAAS,CAACM,MAAM,CAAC2C,OAAO,EAAE1B,OAAO,CAAC;IAElD,MAAM2B,OAAsB,GAAG,EAAE;;IAEjC;IACA;IACA,IAAIC,SAAS,GAAG,CAAC;;IAEjB;IACA,IAAIvB,eAAe,CAACwB,MAAM,GAAG,CAAC,EAAE;MAC9BF,OAAO,CAACG,IAAI,CAAC,GAAGzB,eAAe,CAAC;MAChC,MAAMrB,MAAM,GAAGwC,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAEA,OAAO,CAACE,MAAM,EAAE;QAAEG,YAAY,EAAE;MAAM,CAAC,CAAC;MACzF,MAAMN,OAAO,CAAC1C,MAAM,CAAC;MACrBwC,gBAAgB,CAACS,KAAK,CAAC,CAAC;MACxBL,SAAS,GAAGD,OAAO,CAACE,MAAM;IAC5B;IAEA,MAAMnC,IAAI,GAAGA,CAAA,KAAMX,MAAM,CAACW,IAAI,CAAC,CAAC;IAEhC,MAAMwC,gBAAgB,GAAGzD,SAAS,CAACK,uBAAuB,EAAEkB,OAAO,CAAC;IAEpE,MAAMmC,WAAW,GAAG,MAAAA,CAClBC,OAA2B,EAC3B/C,OAAiC,EACjCH,SAA+B,KACA;MAC/B,MAAMmD,aAAa,GAAG;QAAE,GAAGjC,uBAAuB;QAAE,GAAGlB;MAAU,CAAC;MAClE,MAAMoD,cAAc,GAAG;QAAEpD,SAAS,EAAEmD,aAAa;QAAElD,QAAQ;QAAEC,SAAS;QAAEC;MAAQ,CAAC;MAEjF,MAAMkD,gBAAgB,GAAGX,SAAS;MAClC,MAAMY,UAAU,GAAGzD,MAAM,CAAC0D,eAAe,CAAC,CAAC,CAACC,GAAG;MAE/C,MAAMC,YAAY,GAAGhB,OAAO,CAACE,MAAM;MACnC,MAAMe,mBAAyC,GAAG,EAAE;MAEpDjB,OAAO,CAACG,IAAI,CAAC;QAAEe,IAAI,EAAE,MAAM;QAAEC,OAAO,EAAEV;MAAQ,CAAC,CAAC;MAEhD,IAAI7B,WAAW,EAAE;QACfxB,MAAM,CAACgE,KAAK,CAAC,CAAC;QACdnB,SAAS,GAAG,CAAC;MACf;MAEA,IAAI;QACF,IAAIoB,cAAc,GAAGC,IAAI,CAACC,GAAG,CAAC,CAAC;;QAE/B;QACA,MAAMC,QAAQ,GAAGxB,OAAO,CAACE,MAAM,GAAGD,SAAS;QAC3C,MAAMwB,UAAU,GAAG5B,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAEwB,QAAQ,EAAE;UAAEnB,YAAY,EAAE;QAAM,CAAC,CAAC;QACvF,MAAMN,OAAO,CAAC0B,UAAU,CAAC;QACzB5B,gBAAgB,CAACS,KAAK,CAAC,CAAC;;QAExB;QACA,MAAMoB,cAAc,GAAGtE,MAAM,CAAC0D,eAAe,CAAC,CAAC,CAACC,GAAG;QACnDd,SAAS,GAAGD,OAAO,CAACE,MAAM;QAE1B,IAAIyB,YAAqC,GAAG,cAAc;QAE1D,KAAK,IAAIC,WAAW,GAAG,CAAC,EAAEA,WAAW,GAAGxC,YAAY,EAAE,EAAEwC,WAAW,EAAE;UACnE,MAAMC,WAAW,GAAG7B,OAAO,CAACE,MAAM,GAAGD,SAAS;UAC9C,MAAM5C,MAAM,GAAGwC,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAE6B,WAAW,EAAE;YAAExB,YAAY,EAAE;UAAK,CAAC,CAAC;UAErF,MAAMyB,iBAAiB,GAAGR,IAAI,CAACC,GAAG,CAAC,CAAC,GAAGF,cAAc;UAErD,MAAM;YAAE1D,QAAQ;YAAEK;UAAM,CAAC,GAAG,MAAMuC,gBAAgB,CAACnD,MAAM,EAAEC,MAAM,EAAEsD,cAAc,CAAC;UAClFd,gBAAgB,CAACS,KAAK,CAAC,CAAC;UACxBW,mBAAmB,CAACd,IAAI,CAAC;YAAE,GAAGnC,KAAK;YAAE8D;UAAkB,CAAC,CAAC;;UAEzD;UACA;UACA1E,MAAM,CAACgE,KAAK,CAACM,cAAc,CAAC;;UAE5B;UACA,MAAMK,WAAW,GAAG5C,cAAc,GAAGxB,QAAQ,CAAC;UAE9C,IAAI,CAACoE,WAAW,IAAIA,WAAW,CAACC,SAAS,CAAC9B,MAAM,KAAK,CAAC,EAAE;YACtDF,OAAO,CAACG,IAAI,CAAC;cAAEe,IAAI,EAAE,WAAW;cAAEC,OAAO,EAAExD;YAAS,CAAC,CAAC;YACtDgE,YAAY,GAAG,MAAM;YACrB;UACF;;UAEA;UACA3B,OAAO,CAACG,IAAI,CAAC;YACXe,IAAI,EAAE,WAAW;YACjBC,OAAO,EAAEY,WAAW,CAACE,WAAW;YAChCD,SAAS,EAAED,WAAW,CAACC;UACzB,CAAC,CAAC;UAEF,KAAK,MAAME,QAAQ,IAAIH,WAAW,CAACC,SAAS,EAAE;YAC5C,MAAMG,IAAI,GAAGjD,KAAK,EAAEkD,IAAI,CAAEC,CAAC,IAAKA,CAAC,CAACC,QAAQ,CAACC,IAAI,KAAKL,QAAQ,CAACI,QAAQ,CAACC,IAAI,CAAC;YAE3E,IAAIC,WAA+B;YACnC,IAAI,CAACL,IAAI,EAAE;cACTK,WAAW,GAAG,gBAAgBN,QAAQ,CAACI,QAAQ,CAACC,IAAI,uCAAuC;YAC7F,CAAC,MAAM;cACL,IAAI;gBACFC,WAAW,GAAG,MAAML,IAAI,CAACM,OAAO,CAACP,QAAQ,CAACI,QAAQ,CAACI,SAAS,CAAC;cAC/D,CAAC,CAAC,OAAOC,GAAG,EAAE;gBACZH,WAAW,GAAG,wBAAwBN,QAAQ,CAACI,QAAQ,CAACC,IAAI,KAAKK,MAAM,CAACD,GAAG,CAAC,EAAE;cAChF;YACF;YAEA3C,OAAO,CAACG,IAAI,CAAC;cACXe,IAAI,EAAE,MAAM;cACZ2B,UAAU,EAAEX,QAAQ,CAACY,EAAE;cACvBP,IAAI,EAAEL,QAAQ,CAACI,QAAQ,CAACC,IAAI;cAC5BpB,OAAO,EAAEqB;YACX,CAAC,CAAC;UACJ;UAEAnB,cAAc,GAAGC,IAAI,CAACC,GAAG,CAAC,CAAC;QAC7B;;QAEA;QACA;QACA,MAAMM,WAAW,GAAG7B,OAAO,CAACE,MAAM,GAAGD,SAAS;QAC9C,IAAI4B,WAAW,GAAG,CAAC,EAAE;UACnB,MAAMxE,MAAM,GAAGwC,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAE6B,WAAW,EAAE;YAAExB,YAAY,EAAE;UAAM,CAAC,CAAC;UACtF,MAAMN,OAAO,CAAC1C,MAAM,CAAC;UACrBwC,gBAAgB,CAACS,KAAK,CAAC,CAAC;UACxBL,SAAS,GAAGD,OAAO,CAACE,MAAM;QAC5B;QAEA,OAAO;UAAE6C,QAAQ,EAAE/C,OAAO,CAACgD,KAAK,CAAChC,YAAY,CAAC;UAAEhD,KAAK,EAAEiD,mBAAmB;UAAEU;QAAa,CAAC;MAC5F,CAAC,CAAC,OAAOgB,GAAG,EAAE;QACZ;QACA;QACA;QACA;QACA;QACA;QACA3C,OAAO,CAACE,MAAM,GAAGc,YAAY;QAC7Bf,SAAS,GAAGW,gBAAgB;QAC5BxD,MAAM,CAACgE,KAAK,CAACxC,WAAW,GAAG,CAAC,GAAGiC,UAAU,CAAC;QAC1ChB,gBAAgB,CAACS,KAAK,CAAC,CAAC;QACxB,MAAMqC,GAAG;MACX;IACF,CAAC;IAED,OAAO;MACL5E,IAAI;MACJQ,OAAO;MACPiC,WAAW;MACXyC,UAAU,EAAEA,CAAA,KAAM,CAAC,GAAGjD,OAAO,CAAC;MAC9Bc,eAAe,EAAEA,CAAA,KAAM1D,MAAM,CAAC0D,eAAe,CAAC;IAChD,CAAC;EACH,CAAC,CAAC,OAAOoC,KAAK,EAAE;IACd3E,OAAO,CAAC,CAAC;IACT,MAAM2E,KAAK;EACb;AACF","ignoreList":[]}
+{"version":3,"names":["scheduleOnRN","RNBlobUtil","wrapAsync","createResourceScope","createLLMRunner","parseTokenizerConfig","createChatPreprocessor","generateChatTurnWorklet","runner","prompt","options","genConfig","stopTokens","stopRegex","onToken","response","callback","token","includes","test","stop","stats","generate","DEFAULT_MAX_TURNS","createLLMChatSession","config","runtime","scope","dispose","generationConfig","defaultGenerationConfig","initialMessages","toolOpts","resetOnTurn","modelPath","tokenizerPath","tokenizerConfigPath","modalities","preprocessorConfig","tools","parseToolCalls","maxToolTurns","tokenizerConfigStr","fs","readFile","tokenizerConfig","JSON","parse","chatTemplate","chatPreprocessorConfig","chatPreprocessor","track","prefill","history","committed","length","push","process","addGenPrompt","clear","generateChatTurn","sendMessage","message","turnGenConfig","generationOpts","initialCommitted","initialPos","getKVCacheState","pos","turnStartIdx","generationStatsList","role","content","reset","prefillStartMs","Date","now","toCommit","userPrompt","posAtEndOfUser","finishReason","currentTurn","uncommitted","prefillDurationMs","parsedTools","toolCalls","textContent","toolCall","tool","find","t","function","name","toolContent","execute","arguments","err","String","toolCallId","id","messages","slice","getHistory","error"],"sourceRoot":"../../../../../src","sources":["extensions/llm/tasks/llmChatSession.ts"],"mappings":";;AAAA;AACA;AACA;AACA;;AAEA,SAASA,YAAY,QAA6B,uBAAuB;AACzE,OAAOC,UAAU,MAAM,wBAAwB;AAE/C,SAASC,SAAS,QAAQ,0BAAuB;AACjD,SAASC,mBAAmB,QAAQ,2BAAwB;AAC5D,SACEC,eAAe,QAOV,iBAAc;AACrB,SAASC,oBAAoB,QAAQ,6BAA0B;AAC/D,SACEC,sBAAsB,QAIjB,8BAA2B;;AAGlC;AACA;AACA;AACA;;AAcA;AACA;AACA;AACA;;AAUA;AACA;AACA;AACA;;AAiBA;AACA;AACA;AACA;;AAcA;AACA;AACA;AACA;;AAsCA,SAASC,uBAAuBA,CAC9BC,MAAiB,EACjBC,MAAc,EACdC,OAKC,EACkE;EACnE,SAAS;;EACT,MAAM;IAAEC,SAAS;IAAEC,UAAU;IAAEC,SAAS;IAAEC;EAAQ,CAAC,GAAGJ,OAAO;EAE7D,IAAIK,QAAQ,GAAG,EAAE;EAEjB,MAAMC,QAAQ,GAAIC,KAAa,IAAK;IAClC,IAAIL,UAAU,CAACM,QAAQ,CAACD,KAAK,CAAC,EAAE;IAChCF,QAAQ,IAAIE,KAAK;IAEjB,IAAIH,OAAO,EAAEd,YAAY,CAACc,OAAO,EAAEG,KAAK,CAAC;IACzC,IAAIJ,SAAS,EAAEM,IAAI,CAACJ,QAAQ,CAAC,EAAEP,MAAM,CAACY,IAAI,CAAC,CAAC;EAC9C,CAAC;EAED,MAAMC,KAAK,GAAGb,MAAM,CAACc,QAAQ,CAACb,MAAM,EAAEE,SAAS,EAAEK,QAAQ,CAAC;EAE1D,OAAO;IAAED,QAAQ;IAAEM;EAAM,CAAC;AAC5B;AAEA,MAAME,iBAAiB,GAAG,CAAC;;AAE3B;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,eAAeC,oBAAoBA,CACxCC,MAAgB,EAChBf,OAA8B,GAAG,CAAC,CAAC,EACnCgB,OAAwB,EACC;EACzB,MAAMC,KAAK,GAAGxB,mBAAmB,CAAC,CAAC;EACnC,MAAMyB,OAAO,GAAGD,KAAK,CAACC,OAAO;EAE7B,IAAI;IACF,MAAM;MACJC,gBAAgB,EAAEC,uBAAuB;MACzCC,eAAe,GAAG,EAAE;MACpBlB,SAAS;MACTmB,QAAQ;MACRC,WAAW,GAAG;IAChB,CAAC,GAAGvB,OAAO;IAEX,MAAM;MAAEwB,SAAS;MAAEC,aAAa;MAAEC,mBAAmB;MAAEC,UAAU;MAAEC;IAAmB,CAAC,GACrFb,MAAM;IACR,MAAM;MAAEc,KAAK;MAAEC,cAAc;MAAEC,YAAY,GAAGlB;IAAkB,CAAC,GAAGS,QAAQ,IAAI,CAAC,CAAC;;IAElF;IACA,MAAMU,kBAAkB,GAAG,MAAMzC,UAAU,CAAC0C,EAAE,CAACC,QAAQ,CAACR,mBAAmB,EAAE,MAAM,CAAC;IACpF,MAAMS,eAAe,GAAGxC,oBAAoB,CAACyC,IAAI,CAACC,KAAK,CAACL,kBAAkB,CAAC,CAAC;IAC5E,MAAM;MAAEM,YAAY;MAAEpC;IAAW,CAAC,GAAGiC,eAAe;;IAEpD;IACA,MAAMI,sBAAsB,GAAG;MAAED,YAAY;MAAEX,UAAU;MAAEC,kBAAkB;MAAEC;IAAM,CAAC;IACtF,MAAMW,gBAAgB,GAAGvB,KAAK,CAACwB,KAAK,CAAC7C,sBAAsB,CAAC2C,sBAAsB,CAAC,CAAC;;IAEpF;IACA,MAAMzC,MAAM,GAAGmB,KAAK,CAACwB,KAAK,CACxB,MAAMjD,SAAS,CAACE,eAAe,EAAEsB,OAAO,CAAC,CAACQ,SAAS,EAAEC,aAAa,EAAEE,UAAU,CAChF,CAAC;IACD,MAAMe,OAAO,GAAGlD,SAAS,CAACM,MAAM,CAAC4C,OAAO,EAAE1B,OAAO,CAAC;IAElD,MAAM2B,OAAsB,GAAG,EAAE;;IAEjC;IACA;IACA,IAAIC,SAAS,GAAG,CAAC;;IAEjB;IACA,IAAIvB,eAAe,CAACwB,MAAM,GAAG,CAAC,EAAE;MAC9BF,OAAO,CAACG,IAAI,CAAC,GAAGzB,eAAe,CAAC;MAChC,MAAMtB,MAAM,GAAGyC,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAEA,OAAO,CAACE,MAAM,EAAE;QAAEG,YAAY,EAAE;MAAM,CAAC,CAAC;MACzF,MAAMN,OAAO,CAAC3C,MAAM,CAAC;MACrByC,gBAAgB,CAACS,KAAK,CAAC,CAAC;MACxBL,SAAS,GAAGD,OAAO,CAACE,MAAM;IAC5B;IAEA,MAAMnC,IAAI,GAAGA,CAAA,KAAMZ,MAAM,CAACY,IAAI,CAAC,CAAC;IAEhC,MAAMwC,gBAAgB,GAAG1D,SAAS,CAACK,uBAAuB,EAAEmB,OAAO,CAAC;IAEpE,MAAMmC,WAAW,GAAG,MAAAA,CAClBC,OAA2B,EAC3BhD,OAAiC,EACjCH,SAA+B,KACA;MAC/B,MAAMoD,aAAa,GAAG;QAAE,GAAGjC,uBAAuB;QAAE,GAAGnB;MAAU,CAAC;MAClE,MAAMqD,cAAc,GAAG;QAAErD,SAAS,EAAEoD,aAAa;QAAEnD,UAAU;QAAEC,SAAS;QAAEC;MAAQ,CAAC;MAEnF,MAAMmD,gBAAgB,GAAGX,SAAS;MAClC,MAAMY,UAAU,GAAG1D,MAAM,CAAC2D,eAAe,CAAC,CAAC,CAACC,GAAG;MAE/C,MAAMC,YAAY,GAAGhB,OAAO,CAACE,MAAM;MACnC,MAAMe,mBAAyC,GAAG,EAAE;MAEpDjB,OAAO,CAACG,IAAI,CAAC;QAAEe,IAAI,EAAE,MAAM;QAAEC,OAAO,EAAEV;MAAQ,CAAC,CAAC;MAEhD,IAAI7B,WAAW,EAAE;QACfzB,MAAM,CAACiE,KAAK,CAAC,CAAC;QACdnB,SAAS,GAAG,CAAC;MACf;MAEA,IAAI;QACF,IAAIoB,cAAc,GAAGC,IAAI,CAACC,GAAG,CAAC,CAAC;;QAE/B;QACA,MAAMC,QAAQ,GAAGxB,OAAO,CAACE,MAAM,GAAGD,SAAS;QAC3C,MAAMwB,UAAU,GAAG5B,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAEwB,QAAQ,EAAE;UAAEnB,YAAY,EAAE;QAAM,CAAC,CAAC;QACvF,MAAMN,OAAO,CAAC0B,UAAU,CAAC;QACzB5B,gBAAgB,CAACS,KAAK,CAAC,CAAC;;QAExB;QACA,MAAMoB,cAAc,GAAGvE,MAAM,CAAC2D,eAAe,CAAC,CAAC,CAACC,GAAG;QACnDd,SAAS,GAAGD,OAAO,CAACE,MAAM;QAE1B,IAAIyB,YAAqC,GAAG,cAAc;QAE1D,KAAK,IAAIC,WAAW,GAAG,CAAC,EAAEA,WAAW,GAAGxC,YAAY,EAAE,EAAEwC,WAAW,EAAE;UACnE,MAAMC,WAAW,GAAG7B,OAAO,CAACE,MAAM,GAAGD,SAAS;UAC9C,MAAM7C,MAAM,GAAGyC,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAE6B,WAAW,EAAE;YAAExB,YAAY,EAAE;UAAK,CAAC,CAAC;UAErF,MAAMyB,iBAAiB,GAAGR,IAAI,CAACC,GAAG,CAAC,CAAC,GAAGF,cAAc;UAErD,MAAM;YAAE3D,QAAQ;YAAEM;UAAM,CAAC,GAAG,MAAMuC,gBAAgB,CAACpD,MAAM,EAAEC,MAAM,EAAEuD,cAAc,CAAC;UAClFd,gBAAgB,CAACS,KAAK,CAAC,CAAC;UACxBW,mBAAmB,CAACd,IAAI,CAAC;YAAE,GAAGnC,KAAK;YAAE8D;UAAkB,CAAC,CAAC;;UAEzD;UACA;UACA3E,MAAM,CAACiE,KAAK,CAACM,cAAc,CAAC;;UAE5B;UACA,MAAMK,WAAW,GAAG5C,cAAc,GAAGzB,QAAQ,CAAC;UAE9C,IAAI,CAACqE,WAAW,IAAIA,WAAW,CAACC,SAAS,CAAC9B,MAAM,KAAK,CAAC,EAAE;YACtDF,OAAO,CAACG,IAAI,CAAC;cAAEe,IAAI,EAAE,WAAW;cAAEC,OAAO,EAAEzD;YAAS,CAAC,CAAC;YACtDiE,YAAY,GAAG,MAAM;YACrB;UACF;;UAEA;UACA3B,OAAO,CAACG,IAAI,CAAC;YACXe,IAAI,EAAE,WAAW;YACjBC,OAAO,EAAEY,WAAW,CAACE,WAAW;YAChCD,SAAS,EAAED,WAAW,CAACC;UACzB,CAAC,CAAC;UAEF,KAAK,MAAME,QAAQ,IAAIH,WAAW,CAACC,SAAS,EAAE;YAC5C,MAAMG,IAAI,GAAGjD,KAAK,EAAEkD,IAAI,CAAEC,CAAC,IAAKA,CAAC,CAACC,QAAQ,CAACC,IAAI,KAAKL,QAAQ,CAACI,QAAQ,CAACC,IAAI,CAAC;YAE3E,IAAIC,WAA+B;YACnC,IAAI,CAACL,IAAI,EAAE;cACTK,WAAW,GAAG,gBAAgBN,QAAQ,CAACI,QAAQ,CAACC,IAAI,uCAAuC;YAC7F,CAAC,MAAM;cACL,IAAI;gBACFC,WAAW,GAAG,MAAML,IAAI,CAACM,OAAO,CAACP,QAAQ,CAACI,QAAQ,CAACI,SAAS,CAAC;cAC/D,CAAC,CAAC,OAAOC,GAAG,EAAE;gBACZH,WAAW,GAAG,wBAAwBN,QAAQ,CAACI,QAAQ,CAACC,IAAI,KAAKK,MAAM,CAACD,GAAG,CAAC,EAAE;cAChF;YACF;YAEA3C,OAAO,CAACG,IAAI,CAAC;cACXe,IAAI,EAAE,MAAM;cACZ2B,UAAU,EAAEX,QAAQ,CAACY,EAAE;cACvBP,IAAI,EAAEL,QAAQ,CAACI,QAAQ,CAACC,IAAI;cAC5BpB,OAAO,EAAEqB;YACX,CAAC,CAAC;UACJ;UAEAnB,cAAc,GAAGC,IAAI,CAACC,GAAG,CAAC,CAAC;QAC7B;;QAEA;QACA;QACA,MAAMM,WAAW,GAAG7B,OAAO,CAACE,MAAM,GAAGD,SAAS;QAC9C,IAAI4B,WAAW,GAAG,CAAC,EAAE;UACnB,MAAMzE,MAAM,GAAGyC,gBAAgB,CAACO,OAAO,CAACJ,OAAO,EAAE6B,WAAW,EAAE;YAAExB,YAAY,EAAE;UAAM,CAAC,CAAC;UACtF,MAAMN,OAAO,CAAC3C,MAAM,CAAC;UACrByC,gBAAgB,CAACS,KAAK,CAAC,CAAC;UACxBL,SAAS,GAAGD,OAAO,CAACE,MAAM;QAC5B;QAEA,OAAO;UAAE6C,QAAQ,EAAE/C,OAAO,CAACgD,KAAK,CAAChC,YAAY,CAAC;UAAEhD,KAAK,EAAEiD,mBAAmB;UAAEU;QAAa,CAAC;MAC5F,CAAC,CAAC,OAAOgB,GAAG,EAAE;QACZ;QACA;QACA;QACA;QACA;QACA;QACA3C,OAAO,CAACE,MAAM,GAAGc,YAAY;QAC7Bf,SAAS,GAAGW,gBAAgB;QAC5BzD,MAAM,CAACiE,KAAK,CAACxC,WAAW,GAAG,CAAC,GAAGiC,UAAU,CAAC;QAC1ChB,gBAAgB,CAACS,KAAK,CAAC,CAAC;QACxB,MAAMqC,GAAG;MACX;IACF,CAAC;IAED,OAAO;MACL5E,IAAI;MACJQ,OAAO;MACPiC,WAAW;MACXyC,UAAU,EAAEA,CAAA,KAAM,CAAC,GAAGjD,OAAO,CAAC;MAC9Bc,eAAe,EAAEA,CAAA,KAAM3D,MAAM,CAAC2D,eAAe,CAAC;IAChD,CAAC;EACH,CAAC,CAAC,OAAOoC,KAAK,EAAE;IACd3E,OAAO,CAAC,CAAC;IACT,MAAM2E,KAAK;EACb;AACF","ignoreList":[]}

package/lib/module/extensions/llm/utils/tokenizerConfig.js

@@ -43,9 +43,11 @@ export function parseTokenizerConfig(config) {
   if (!eosToken) {
     throw RnExecuTorchError('LOAD_FAILED', 'tokenizer_config.json does not define required `eos_token` string');
   }
+  const stopTokens = [eosToken, resolveToken(config.eot_token), resolveToken(config.pad_token)].filter((token, index, all) => !!token && all.indexOf(token) === index);
   return {
     chatTemplate,
-    eosToken
+    eosToken,
+    stopTokens
   };
 }
 //# sourceMappingURL=tokenizerConfig.js.map

package/lib/module/extensions/llm/utils/tokenizerConfig.js.map

@@ -1 +1 @@
-{"version":3,"names":["RnExecuTorchError","resolveToken","token","content","undefined","parseTokenizerConfig","config","chatTemplate","chat_template","Array","isArray","entry","find","t","name","template","eosToken","eos_token"],"sourceRoot":"../../../../../src","sources":["extensions/llm/utils/tokenizerConfig.ts"],"mappings":";;AAAA;AACA;AACA;AACA;;AAEA,SAASA,iBAAiB,QAAQ,wBAAqB;;AAEvD;AACA;AACA;AACA;;AAQA,SAASC,YAAYA,CAACC,KAAc,EAAsB;EACxD,IAAI,OAAOA,KAAK,KAAK,QAAQ,EAAE,OAAOA,KAAK;EAC3C,IAAIA,KAAK,IAAI,OAAOA,KAAK,KAAK,QAAQ,IAAI,OAAQA,KAAK,CAASC,OAAO,KAAK,QAAQ,EAAE;IACpF,OAAQD,KAAK,CAASC,OAAO;EAC/B;EACA,OAAOC,SAAS;AAClB;;AAEA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,SAASC,oBAAoBA,CAACC,MAAW,EAAuB;EACrE,IAAIC,YAAY,GAAGD,MAAM,CAACE,aAAa;;EAEvC;EACA,IAAIC,KAAK,CAACC,OAAO,CAACH,YAAY,CAAC,EAAE;IAC/B,MAAMI,KAAK,GAAGJ,YAAY,CAACK,IAAI,CAAEC,CAAC,IAAKA,CAAC,EAAEC,IAAI,KAAK,SAAS,CAAC,IAAIP,YAAY,CAAC,CAAC,CAAC;IAChFA,YAAY,GAAGI,KAAK,EAAEI,QAAQ;EAChC;EAEA,IAAI,OAAOR,YAAY,KAAK,QAAQ,EAAE;IACpC,MAAMP,iBAAiB,CACrB,aAAa,EACb,iEACF,CAAC;EACH;EAEA,MAAMgB,QAAQ,GAAGf,YAAY,CAACK,MAAM,CAACW,SAAS,CAAC;EAE/C,IAAI,CAACD,QAAQ,EAAE;IACb,MAAMhB,iBAAiB,CACrB,aAAa,EACb,mEACF,CAAC;EACH;EAEA,OAAO;IAAEO,YAAY;IAAES;EAAS,CAAC;AACnC","ignoreList":[]}
+{"version":3,"names":["RnExecuTorchError","resolveToken","token","content","undefined","parseTokenizerConfig","config","chatTemplate","chat_template","Array","isArray","entry","find","t","name","template","eosToken","eos_token","stopTokens","eot_token","pad_token","filter","index","all","indexOf"],"sourceRoot":"../../../../../src","sources":["extensions/llm/utils/tokenizerConfig.ts"],"mappings":";;AAAA;AACA;AACA;AACA;;AAEA,SAASA,iBAAiB,QAAQ,wBAAqB;;AAEvD;AACA;AACA;AACA;;AAYA,SAASC,YAAYA,CAACC,KAAc,EAAsB;EACxD,IAAI,OAAOA,KAAK,KAAK,QAAQ,EAAE,OAAOA,KAAK;EAC3C,IAAIA,KAAK,IAAI,OAAOA,KAAK,KAAK,QAAQ,IAAI,OAAQA,KAAK,CAASC,OAAO,KAAK,QAAQ,EAAE;IACpF,OAAQD,KAAK,CAASC,OAAO;EAC/B;EACA,OAAOC,SAAS;AAClB;;AAEA;AACA;AACA;AACA;AACA;AACA;AACA;AACA;AACA,OAAO,SAASC,oBAAoBA,CAACC,MAAW,EAAuB;EACrE,IAAIC,YAAY,GAAGD,MAAM,CAACE,aAAa;;EAEvC;EACA,IAAIC,KAAK,CAACC,OAAO,CAACH,YAAY,CAAC,EAAE;IAC/B,MAAMI,KAAK,GAAGJ,YAAY,CAACK,IAAI,CAAEC,CAAC,IAAKA,CAAC,EAAEC,IAAI,KAAK,SAAS,CAAC,IAAIP,YAAY,CAAC,CAAC,CAAC;IAChFA,YAAY,GAAGI,KAAK,EAAEI,QAAQ;EAChC;EAEA,IAAI,OAAOR,YAAY,KAAK,QAAQ,EAAE;IACpC,MAAMP,iBAAiB,CACrB,aAAa,EACb,iEACF,CAAC;EACH;EAEA,MAAMgB,QAAQ,GAAGf,YAAY,CAACK,MAAM,CAACW,SAAS,CAAC;EAE/C,IAAI,CAACD,QAAQ,EAAE;IACb,MAAMhB,iBAAiB,CACrB,aAAa,EACb,mEACF,CAAC;EACH;EAEA,MAAMkB,UAAU,GAAG,CACjBF,QAAQ,EACRf,YAAY,CAACK,MAAM,CAACa,SAAS,CAAC,EAC9BlB,YAAY,CAACK,MAAM,CAACc,SAAS,CAAC,CAC/B,CAACC,MAAM,CAAC,CAACnB,KAAK,EAAEoB,KAAK,EAAEC,GAAG,KAAsB,CAAC,CAACrB,KAAK,IAAIqB,GAAG,CAACC,OAAO,CAACtB,KAAK,CAAC,KAAKoB,KAAK,CAAC;EAEzF,OAAO;IAAEf,YAAY;IAAES,QAAQ;IAAEE;EAAW,CAAC;AAC/C","ignoreList":[]}

package/lib/module/fetcher/telemetry.js

@@ -12,7 +12,7 @@ const DOWNLOAD_EVENT_ENDPOINT = 'https://ai.swmansion.com/telemetry/downloads/ap
 // self-referencing import would need package `exports` support in the consuming
 // bundler to load at all. A wrong value here is analytics noise; a failed
 // import would break the bundle.
-const LIB_VERSION = '0.10.2';
+const LIB_VERSION = '0.10.3';
 
 // Anonymous analytics are on by default; apps opt out via setTelemetryEnabled.
 let telemetryEnabled = true;

package/lib/typescript/src/extensions/llm/llmRunner.d.ts

@@ -8,8 +8,6 @@ import type { Tensor } from '../../core/tensor';
  * @category LLM / Types
  */
 export type LLMGenerationConfig = {
-    /** Whether to echo the prompt in the generated output. */
-    readonly echo?: boolean;
     /** Whether to ignore EOS tokens during generation. */
     readonly ignoreEos?: boolean;
     /** Maximum number of new tokens to generate. */

package/lib/typescript/src/extensions/llm/llmRunner.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"llmRunner.d.ts","sourceRoot":"","sources":["../../../../../src/extensions/llm/llmRunner.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AAKhD;;;;GAIG;AACH,MAAM,MAAM,mBAAmB,GAAG;IAChC,0DAA0D;IAC1D,QAAQ,CAAC,IAAI,CAAC,EAAE,OAAO,CAAC;IACxB,sDAAsD;IACtD,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC;IAC7B,gDAAgD;IAChD,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAC/B,gDAAgD;IAChD,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;CAC/B,CAAC;AAEF;;;;GAIG;AACH,MAAM,MAAM,kBAAkB,GAAG;IAC/B,4CAA4C;IAC5C,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAC;IACjC,wCAAwC;IACxC,QAAQ,CAAC,kBAAkB,EAAE,MAAM,CAAC;IACpC,oEAAoE;IACpE,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,6EAA6E;IAC7E,QAAQ,CAAC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IACpC,wDAAwD;IACxD,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,0DAA0D;IAC1D,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC,4DAA4D;IAC5D,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,8DAA8D;IAC9D,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;CACjC,CAAC;AAEF;;;;GAIG;AACH,MAAM,MAAM,UAAU,GAClB;IAAE,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAClD;IAAE,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvD;;;;GAIG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,MAAM,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC,MAAM,GAAG,UAAU,CAAC,EAAE,CAAC;AAE/D;;;;GAIG;AACH,MAAM,MAAM,eAAe,GAAG;IAC5B,gFAAgF;IAChF,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,sEAAsE;IACtE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,kEAAkE;IAClE,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAC;IACjC,sEAAsE;IACtE,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;CAC7B,CAAC;AAEF;;;;;;GAMG;AACH,MAAM,MAAM,SAAS,GAAG;IACtB,oCAAoC;IACpC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,sDAAsD;IACtD,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;IAC/B,sFAAsF;IACtF,QAAQ,CAAC,UAAU,EAAE,SAAS,QAAQ,EAAE,CAAC;IAEzC;;OAEG;IACH,OAAO,IAAI,IAAI,CAAC;IAEhB;;OAEG;IACH,IAAI,IAAI,IAAI,CAAC;IAEb;;;;OAIG;IACH,KAAK,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAEhC;;OAEG;IACH,eAAe,IAAI,eAAe,CAAC;IAEnC;;;OAGG;IACH,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAE9B;;;;;;OAMG;IACH,QAAQ,CACN,MAAM,EAAE,MAAM,EACd,MAAM,CAAC,EAAE,mBAAmB,EAC5B,OAAO,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,GAChC,kBAAkB,CAAC;CAOvB,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,wBAAgB,eAAe,CAC7B,SAAS,EAAE,MAAM,EACjB,aAAa,EAAE,MAAM,EACrB,UAAU,CAAC,EAAE,SAAS,QAAQ,EAAE,GAC/B,SAAS,CAGX"}
+{"version":3,"file":"llmRunner.d.ts","sourceRoot":"","sources":["../../../../../src/extensions/llm/llmRunner.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,mBAAmB,CAAC;AAKhD;;;;GAIG;AACH,MAAM,MAAM,mBAAmB,GAAG;IAChC,sDAAsD;IACtD,QAAQ,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC;IAC7B,gDAAgD;IAChD,QAAQ,CAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAC/B,gDAAgD;IAChD,QAAQ,CAAC,WAAW,CAAC,EAAE,MAAM,CAAC;CAC/B,CAAC;AAEF;;;;GAIG;AACH,MAAM,MAAM,kBAAkB,GAAG;IAC/B,4CAA4C;IAC5C,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAC;IACjC,wCAAwC;IACxC,QAAQ,CAAC,kBAAkB,EAAE,MAAM,CAAC;IACpC,oEAAoE;IACpE,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,6EAA6E;IAC7E,QAAQ,CAAC,iBAAiB,CAAC,EAAE,MAAM,CAAC;IACpC,wDAAwD;IACxD,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,0DAA0D;IAC1D,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;IAChC,4DAA4D;IAC5D,QAAQ,CAAC,gBAAgB,EAAE,MAAM,CAAC;IAClC,8DAA8D;IAC9D,QAAQ,CAAC,cAAc,EAAE,MAAM,CAAC;CACjC,CAAC;AAEF;;;;GAIG;AACH,MAAM,MAAM,UAAU,GAClB;IAAE,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;CAAE,GAClD;IAAE,QAAQ,CAAC,IAAI,EAAE,OAAO,CAAC;IAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC;AAEvD;;;;GAIG;AACH,MAAM,MAAM,QAAQ,GAAG,UAAU,CAAC,MAAM,CAAC,CAAC;AAE1C;;;;GAIG;AACH,MAAM,MAAM,MAAM,GAAG,MAAM,GAAG,SAAS,CAAC,MAAM,GAAG,UAAU,CAAC,EAAE,CAAC;AAE/D;;;;GAIG;AACH,MAAM,MAAM,eAAe,GAAG;IAC5B,gFAAgF;IAChF,QAAQ,CAAC,GAAG,EAAE,MAAM,CAAC;IACrB,sEAAsE;IACtE,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,kEAAkE;IAClE,QAAQ,CAAC,eAAe,EAAE,MAAM,CAAC;IACjC,sEAAsE;IACtE,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;CAC7B,CAAC;AAEF;;;;;;GAMG;AACH,MAAM,MAAM,SAAS,GAAG;IACtB,oCAAoC;IACpC,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAC;IAC3B,sDAAsD;IACtD,QAAQ,CAAC,aAAa,EAAE,MAAM,CAAC;IAC/B,sFAAsF;IACtF,QAAQ,CAAC,UAAU,EAAE,SAAS,QAAQ,EAAE,CAAC;IAEzC;;OAEG;IACH,OAAO,IAAI,IAAI,CAAC;IAEhB;;OAEG;IACH,IAAI,IAAI,IAAI,CAAC;IAEb;;;;OAIG;IACH,KAAK,CAAC,SAAS,CAAC,EAAE,MAAM,GAAG,IAAI,CAAC;IAEhC;;OAEG;IACH,eAAe,IAAI,eAAe,CAAC;IAEnC;;;OAGG;IACH,OAAO,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAE9B;;;;;;OAMG;IACH,QAAQ,CACN,MAAM,EAAE,MAAM,EACd,MAAM,CAAC,EAAE,mBAAmB,EAC5B,OAAO,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,KAAK,IAAI,GAChC,kBAAkB,CAAC;CAOvB,CAAC;AAEF;;;;;;;;;;;GAWG;AACH,wBAAgB,eAAe,CAC7B,SAAS,EAAE,MAAM,EACjB,aAAa,EAAE,MAAM,EACrB,UAAU,CAAC,EAAE,SAAS,QAAQ,EAAE,GAC/B,SAAS,CAGX"}

package/lib/typescript/src/extensions/llm/utils/tokenizerConfig.d.ts

@@ -11,6 +11,10 @@ export type TokenizerChatConfig = {
     readonly chatTemplate: string;
     /** End-of-sequence token string. */
     readonly eosToken: string;
+    /**
+     * Every terminal token the model may decode into its output.
+     */
+    readonly stopTokens: readonly string[];
 };
 /**
  * Parses raw JSON configuration from `tokenizer_config.json` into a normalized format.

package/lib/typescript/src/extensions/llm/utils/tokenizerConfig.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"tokenizerConfig.d.ts","sourceRoot":"","sources":["../../../../../../src/extensions/llm/utils/tokenizerConfig.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH;;;GAGG;AACH,MAAM,MAAM,mBAAmB,GAAG;IAChC,uDAAuD;IACvD,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,oCAAoC;IACpC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;CAC3B,CAAC;AAUF;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,GAAG,GAAG,mBAAmB,CA0BrE"}
+{"version":3,"file":"tokenizerConfig.d.ts","sourceRoot":"","sources":["../../../../../../src/extensions/llm/utils/tokenizerConfig.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH;;;GAGG;AACH,MAAM,MAAM,mBAAmB,GAAG;IAChC,uDAAuD;IACvD,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC;IAC9B,oCAAoC;IACpC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAC;IAC1B;;OAEG;IACH,QAAQ,CAAC,UAAU,EAAE,SAAS,MAAM,EAAE,CAAC;CACxC,CAAC;AAUF;;;;;;;GAOG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,GAAG,GAAG,mBAAmB,CAgCrE"}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "react-native-executorch",
-  "version": "0.10.2",
-  "nativeLibsVersion": "0.10.0",
+  "version": "0.10.3",
+  "nativeLibsVersion": "0.10.4",
   "description": "An easy way to run AI models in React Native with ExecuTorch",
   "main": "./lib/module/index.js",
   "module": "./lib/module/index.js",

package/scripts/download-libs.js

@@ -11,9 +11,10 @@
  *
  *   headers.tar.gz                   -- ExecuTorch + c10 + torch + tokenizers + opencv
  *                                       headers (platform-independent; always downloaded)
- *   core-android-arm64-v8a.tar.gz    -- executorch, pthreadpool, cpuinfo for arm64
- *   core-android-x86_64.tar.gz       -- executorch for x86_64
- *   core-ios.tar.gz                  -- ExecutorchLib.xcframework (without xnnpack/coreml)
+ *   core-android-arm64-v8a.tar.gz    -- libexecutorch.so (+ executorch.jar) for arm64;
+ *                                       pthreadpool and cpuinfo are statically linked in
+ *   core-android-x86_64.tar.gz       -- libexecutorch.so for x86_64
+ *   core-ios.tar.gz                  -- ExecutorchLib.xcframework + libthreadpool_*.a
  *   opencv-android-arm64-v8a.tar.gz  -- OpenCV for arm64
  *   opencv-android-x86_64.tar.gz     -- OpenCV for x86_64
  *   opencv-ios.tar.gz                -- OpenCV xcframework
@@ -74,7 +75,9 @@ const https = require('https');
 const http = require('http');
 const fs = require('fs');
 const path = require('path');
-const { execSync } = require('child_process');
+const { spawnSync } = require('child_process');
+const { createHash } = require('crypto');
+const { Buffer } = require('buffer');
 
 // ---- Config ----------------------------------------------------------------
 
@@ -151,8 +154,10 @@ const FEATURE_MAP = {
   // YOLO is xnnpack-only, ssdlite/rf_detr add coreml → union.
   objectDetection: { backends: ['xnnpack', 'coreml'], libs: ['opencv'] },
   // Keypoint detection (#1280): BlazeFace + YOLO26-pose ship xnnpack; RF-DETR
-  // keypoint adds coreml + mlx → union. (Named to track the useKeypointDetector
-  // hook; main calls this poseEstimation.)
+  // keypoint adds coreml → union. mlx stays for the legacy API, whose
+  // RF_DETR_KEYPOINT_PREVIEW_MLX_FP32_MODEL outlived the new registry's MLX
+  // keypoint export (dropped in #1418).
+  // (Named to track the useKeypointDetector hook; main calls this poseEstimation.)
   keypointDetection: { backends: ['xnnpack', 'coreml', 'mlx'], libs: ['opencv'] },
   // DeepLab/FCN/LR-ASPP/selfie all ship xnnpack + coreml.
   semanticSegmentation: { backends: ['xnnpack', 'coreml'], libs: ['opencv'] },
@@ -431,15 +436,33 @@ function download(url, dest) {
     };
     get(url);
     file.on('error', (err) => {
-      fs.unlinkSync(dest);
+      // Close before removing: on Windows an open handle makes the unlink fail
+      // with EBUSY, which then replaces the real error with the cleanup's. And
+      // `force` covers the case where the file was never created.
+      file.close(() => fs.rmSync(dest, { force: true }));
       reject(err);
     });
   });
 }
 
+// Hashes in-process rather than shelling out to `sha256sum` / `shasum`. Those
+// are not on PATH everywhere, and GNU `sha256sum` escapes a filename holding a
+// backslash by prefixing its whole output line with one -- so on Windows every
+// hash came back as `\<hex>` and no artifact could ever validate. Read in
+// chunks: some artifacts are hundreds of megabytes and this runs at install.
 function sha256(filePath) {
-  const result = execSync(`sha256sum "${filePath}" || shasum -a 256 "${filePath}"`);
-  return result.toString().split(' ')[0].trim();
+  const hash = createHash('sha256');
+  const buffer = Buffer.alloc(1024 * 1024);
+  const fd = fs.openSync(filePath, 'r');
+  try {
+    let bytesRead;
+    while ((bytesRead = fs.readSync(fd, buffer, 0, buffer.length, null)) > 0) {
+      hash.update(buffer.subarray(0, bytesRead));
+    }
+  } finally {
+    fs.closeSync(fd);
+  }
+  return hash.digest('hex');
 }
 
 async function isCacheValid(artifact) {
@@ -461,6 +484,68 @@ async function isCacheValid(artifact) {
   return expectedChecksum === actualChecksum;
 }
 
+// The backend binaries each artifact owns, relative to the extraction dir. Used
+// to clear stale copies: nothing else removes them, and `isCacheValid()`
+// short-circuits the download, so turning a backend OFF after an install would
+// otherwise leave the old binary in place and (on Android) keep shipping it.
+const BACKEND_FILES = {
+  android: {
+    xnnpack: ['executorch/*/libxnnpack_executorch_backend.so'],
+    vulkan: ['executorch/*/libvulkan_executorch_backend.so'],
+  },
+  ios: {
+    xnnpack: ['XnnpackBackend.xcframework'],
+    coreml: ['CoreMLBackend.xcframework'],
+    mlx: ['MLXBackend.xcframework', 'libs/executorch/mlx.metallib'],
+  },
+};
+
+// Removes the binaries of backends the resolved config does not ask for, so a
+// config change takes effect without anyone deleting third-party/ by hand.
+function pruneDisabledBackends(targets, { backends }) {
+  for (const target of targets) {
+    const platform = target.startsWith('android') ? 'android' : 'ios';
+    const destDir =
+      platform === 'android'
+        ? path.join(THIRD_PARTY_DIR, 'android', 'libs')
+        : path.join(THIRD_PARTY_DIR, 'ios');
+    if (!fs.existsSync(destDir)) continue;
+
+    for (const [backend, patterns] of Object.entries(BACKEND_FILES[platform])) {
+      if (backends.includes(backend)) continue;
+      for (const pattern of patterns) {
+        for (const stale of expandPattern(destDir, pattern)) {
+          console.log(
+            `[react-native-executorch] Removing ${backend} binary (not enabled): ${path.relative(THIRD_PARTY_DIR, stale)}`
+          );
+          fs.rmSync(stale, { recursive: true, force: true });
+        }
+      }
+    }
+  }
+}
+
+// Resolves a path that may contain a single `*` segment (the Android ABI dir).
+// A dependency-free stand-in for a glob, since this script runs at postinstall
+// with nothing but Node's stdlib available.
+function expandPattern(root, pattern) {
+  const segments = pattern.split('/');
+  let candidates = [root];
+  for (const segment of segments) {
+    const next = [];
+    for (const base of candidates) {
+      if (segment === '*') {
+        if (!fs.existsSync(base)) continue;
+        for (const entry of fs.readdirSync(base)) next.push(path.join(base, entry));
+      } else {
+        next.push(path.join(base, segment));
+      }
+    }
+    candidates = next;
+  }
+  return candidates.filter((candidate) => fs.existsSync(candidate));
+}
+
 function extract(tarball, destDir) {
   ensureDir(destDir);
   // `-m` stamps extracted files with the extraction time instead of the mtime
@@ -471,7 +556,36 @@ function extract(tarball, destDir) {
   // and they get archived and linked against the new libraries -- which shows
   // up as an undefined symbol for whatever API changed between the two
   // releases, far away from the actual cause.
-  execSync(`tar -xzmf "${tarball}" -C "${destDir}"`);
+  //
+  // No path reaches tar's argv, because GNU tar (Git for Windows' is the one
+  // on PATH under Git Bash) mangles Windows paths two ways: `-f C:\Users\...`
+  // is the `host:file` form, so it tries to reach a remote host named `C`; and
+  // `-C` is unquoted by default, so the `\r` and `\t` in
+  // `node_modules\react-native-executorch\third-party` become a carriage
+  // return and a tab. The archive arrives on stdin and `cwd` does the chdir.
+  const fd = fs.openSync(tarball, 'r');
+  let result;
+  try {
+    result = spawnSync('tar', ['-xzm', '-f', '-'], {
+      cwd: destDir,
+      stdio: [fd, 'inherit', 'pipe'],
+      encoding: 'utf8',
+    });
+  } finally {
+    fs.closeSync(fd);
+  }
+  if (result.error) {
+    throw new Error(
+      `Could not run \`tar\` to extract ${tarball}: ${result.error.message}. ` +
+        'Install tar (it ships with Windows 10 1803+, Git for Windows and every Unix) and reinstall.'
+    );
+  }
+  if (result.status !== 0) {
+    throw new Error(
+      `tar exited with ${result.status ?? result.signal} while extracting ${tarball}` +
+        `${result.stderr ? `:\n${result.stderr.trim()}` : ''}`
+    );
+  }
 }
 
 // ---- Main ------------------------------------------------------------------
@@ -525,6 +639,10 @@ async function main() {
     console.log(`  ✓ Done`);
   }
 
+  // Belt and braces to core no longer carrying backends: this also clears
+  // binaries left by an EARLIER install that had the backend enabled.
+  pruneDisabledBackends(targets, config);
+
   console.log('[react-native-executorch] Native libs ready.');
 }
 
@@ -536,4 +654,14 @@ if (require.main === module) {
   });
 }
 
-module.exports = { ALL_BACKENDS, ALL_LIBS, FEATURE_MAP, findUserConfig, readUserConfig };
+module.exports = {
+  ALL_BACKENDS,
+  ALL_LIBS,
+  BACKEND_FILES,
+  FEATURE_MAP,
+  findUserConfig,
+  readUserConfig,
+  pruneDisabledBackends,
+  sha256,
+  extract,
+};

package/src/extensions/llm/llmRunner.ts

@@ -13,8 +13,6 @@ declare const llmRunnerBrand: unique symbol;
  * @category LLM / Types
  */
 export type LLMGenerationConfig = {
-  /** Whether to echo the prompt in the generated output. */
-  readonly echo?: boolean;
   /** Whether to ignore EOS tokens during generation. */
   readonly ignoreEos?: boolean;
   /** Maximum number of new tokens to generate. */

package/src/extensions/llm/tasks/llmChatSession.ts

@@ -139,18 +139,18 @@ function generateChatTurnWorklet(
   prompt: Prompt,
   options: {
     readonly genConfig: LLMGenerationConfig;
-    readonly eosToken: string;
+    readonly stopTokens: readonly string[];
     readonly stopRegex?: RegExp;
     readonly onToken?: (token: string) => void;
   }
 ): { readonly response: string; readonly stats: LLMGenerationStats } {
   'worklet';
-  const { genConfig, eosToken, stopRegex, onToken } = options;
+  const { genConfig, stopTokens, stopRegex, onToken } = options;
 
   let response = '';
 
   const callback = (token: string) => {
-    if (token === eosToken) return;
+    if (stopTokens.includes(token)) return;
     response += token;
 
     if (onToken) scheduleOnRN(onToken, token);
@@ -196,7 +196,7 @@ export async function createLLMChatSession(
     // Read and parse tokenizer_config.json
     const tokenizerConfigStr = await RNBlobUtil.fs.readFile(tokenizerConfigPath, 'utf8');
     const tokenizerConfig = parseTokenizerConfig(JSON.parse(tokenizerConfigStr));
-    const { chatTemplate, eosToken } = tokenizerConfig;
+    const { chatTemplate, stopTokens } = tokenizerConfig;
 
     // Prepare chat preprocessor
     const chatPreprocessorConfig = { chatTemplate, modalities, preprocessorConfig, tools };
@@ -233,7 +233,7 @@ export async function createLLMChatSession(
       genConfig?: LLMGenerationConfig
     ): Promise<LLMChatTurnResult> => {
       const turnGenConfig = { ...defaultGenerationConfig, ...genConfig };
-      const generationOpts = { genConfig: turnGenConfig, eosToken, stopRegex, onToken };
+      const generationOpts = { genConfig: turnGenConfig, stopTokens, stopRegex, onToken };
 
       const initialCommitted = committed;
       const initialPos = runner.getKVCacheState().pos;

package/src/extensions/llm/utils/tokenizerConfig.ts

@@ -14,6 +14,10 @@ export type TokenizerChatConfig = {
   readonly chatTemplate: string;
   /** End-of-sequence token string. */
   readonly eosToken: string;
+  /**
+   * Every terminal token the model may decode into its output.
+   */
+  readonly stopTokens: readonly string[];
 };
 
 function resolveToken(token: unknown): string | undefined {
@@ -57,5 +61,11 @@ export function parseTokenizerConfig(config: any): TokenizerChatConfig {
     );
   }
 
-  return { chatTemplate, eosToken };
+  const stopTokens = [
+    eosToken,
+    resolveToken(config.eot_token),
+    resolveToken(config.pad_token),
+  ].filter((token, index, all): token is string => !!token && all.indexOf(token) === index);
+
+  return { chatTemplate, eosToken, stopTokens };
 }

package/src/fetcher/telemetry.ts

@@ -10,7 +10,7 @@ const DOWNLOAD_EVENT_ENDPOINT = 'https://ai.swmansion.com/telemetry/downloads/ap
 // self-referencing import would need package `exports` support in the consuming
 // bundler to load at all. A wrong value here is analytics noise; a failed
 // import would break the bundle.
-const LIB_VERSION = '0.10.2';
+const LIB_VERSION = '0.10.3';
 
 // Anonymous analytics are on by default; apps opt out via setTelemetryEnabled.
 let telemetryEnabled = true;