node-llama-cpp 3.0.0-beta.38 → 3.0.0-beta.39

package/llama/addon/AddonModel.h

@@ -0,0 +1,61 @@
+#pragma once
+#include "llama.h"
+#include "napi.h"
+#include "addonGlobals.h"
+#include "globals/addonProgress.h"
+
+class AddonModel : public Napi::ObjectWrap<AddonModel> {
+    public:
+        llama_model_params model_params;
+        std::vector<llama_model_kv_override> kv_overrides;
+        llama_model* model;
+        uint64_t loadedModelSize;
+        Napi::Reference<Napi::Object> addonExportsRef;
+        bool hasAddonExportsRef;
+        AddonModelData* data;
+
+        std::string modelPath;
+        bool modelLoaded;
+        bool abortModelLoad;
+        bool model_load_stopped;
+        float rawModelLoadPercentage;
+        unsigned modelLoadPercentage;
+        AddonThreadSafeProgressEventCallbackFunction addonThreadSafeOnLoadProgressEventCallback;
+        bool onLoadProgressEventCallbackSet;
+        bool hasLoadAbortSignal;
+
+        bool disposed;
+
+        AddonModel(const Napi::CallbackInfo& info);
+        ~AddonModel();
+        void dispose();
+
+        Napi::Value Init(const Napi::CallbackInfo& info);
+        Napi::Value LoadLora(const Napi::CallbackInfo& info);
+        Napi::Value AbortActiveModelLoad(const Napi::CallbackInfo& info);
+        Napi::Value Dispose(const Napi::CallbackInfo& info);
+        Napi::Value Tokenize(const Napi::CallbackInfo& info);
+        Napi::Value Detokenize(const Napi::CallbackInfo& info);
+        Napi::Value GetTrainContextSize(const Napi::CallbackInfo& info);
+        Napi::Value GetEmbeddingVectorSize(const Napi::CallbackInfo& info);
+        Napi::Value GetTotalSize(const Napi::CallbackInfo& info);
+        Napi::Value GetTotalParameters(const Napi::CallbackInfo& info);
+        Napi::Value GetModelDescription(const Napi::CallbackInfo& info);
+
+        Napi::Value TokenBos(const Napi::CallbackInfo& info);
+        Napi::Value TokenEos(const Napi::CallbackInfo& info);
+        Napi::Value TokenNl(const Napi::CallbackInfo& info);
+        Napi::Value PrefixToken(const Napi::CallbackInfo& info);
+        Napi::Value MiddleToken(const Napi::CallbackInfo& info);
+        Napi::Value SuffixToken(const Napi::CallbackInfo& info);
+        Napi::Value EotToken(const Napi::CallbackInfo& info);
+        Napi::Value GetTokenString(const Napi::CallbackInfo& info);
+
+        Napi::Value GetTokenAttributes(const Napi::CallbackInfo& info);
+        Napi::Value IsEogToken(const Napi::CallbackInfo& info);
+        Napi::Value GetVocabularyType(const Napi::CallbackInfo& info);
+        Napi::Value ShouldPrependBosToken(const Napi::CallbackInfo& info);
+        Napi::Value GetModelSize(const Napi::CallbackInfo& info);
+
+        static void init(Napi::Object exports);
+};

package/llama/addon/AddonModelData.cpp

@@ -0,0 +1,25 @@
+#include <iostream>
+
+#include "addonGlobals.h"
+#include "AddonModelData.h"
+#include "AddonModelLora.h"
+
+AddonModelData::AddonModelData() {
+
+}
+AddonModelData::~AddonModelData() {
+    std::set<AddonModelLora *> currentLoraAdapters;
+    currentLoraAdapters.swap(loraAdapters);
+
+    for (auto lora : currentLoraAdapters) {
+        lora->dispose(true);
+    }
+    currentLoraAdapters.clear();
+}
+
+void AddonModelData::removeLora(AddonModelLora* lora) {
+    auto pos = loraAdapters.find(lora);
+    if (pos != loraAdapters.end()) {
+        loraAdapters.erase(pos);
+    }
+}

package/llama/addon/AddonModelData.h

@@ -0,0 +1,15 @@
+#pragma once
+#include <set>
+#include "llama.h"
+#include "napi.h"
+#include "addonGlobals.h"
+
+class AddonModelData {
+    public:
+        std::set<AddonModelLora *> loraAdapters;
+
+        AddonModelData();
+        ~AddonModelData();
+
+        void removeLora(AddonModelLora* lora);
+};

package/llama/addon/AddonModelLora.cpp

@@ -0,0 +1,107 @@
+#include "addonGlobals.h"
+#include "AddonModel.h"
+#include "AddonModelData.h"
+#include "AddonModelLora.h"
+
+class AddonModelLoraUnloadLoraWorker : public Napi::AsyncWorker {
+    public:
+        AddonModelLora* addonLora;
+
+        AddonModelLoraUnloadLoraWorker(const Napi::Env& env, AddonModelLora* addonLora)
+            : Napi::AsyncWorker(env, "AddonModelLoraUnloadLoraWorker"),
+              addonLora(addonLora),
+              deferred(Napi::Promise::Deferred::New(env)) {
+            addonLora->Ref();
+        }
+        ~AddonModelLoraUnloadLoraWorker() {
+            addonLora->Unref();
+        }
+
+        Napi::Promise GetPromise() {
+            return deferred.Promise();
+        }
+
+    protected:
+        Napi::Promise::Deferred deferred;
+
+        void Execute() {
+            try {
+                addonLora->dispose();
+            } catch (const std::exception& e) {
+                SetError(e.what());
+            } catch(...) {
+                SetError("Unknown error when calling \"llama_lora_adapter_free\"");
+            }
+        }
+        void OnOK() {
+            deferred.Resolve(Env().Undefined());
+        }
+        void OnError(const Napi::Error& err) {
+            deferred.Reject(err.Value());
+        }
+};
+
+AddonModelLora::AddonModelLora(const Napi::CallbackInfo& info) : Napi::ObjectWrap<AddonModelLora>(info) {
+    usages = 0;
+
+    model = Napi::ObjectWrap<AddonModel>::Unwrap(info[0].As<Napi::Object>());
+    loraFilePath = info[1].As<Napi::String>().Utf8Value();
+    lora_adapter = nullptr;
+}
+
+AddonModelLora::~AddonModelLora() {
+    dispose();
+}
+
+void AddonModelLora::dispose(bool skipErase) {
+    if (lora_adapter != nullptr) {
+        auto loraAdapterToDispose = lora_adapter;
+        lora_adapter = nullptr;
+        llama_lora_adapter_free(loraAdapterToDispose);
+        
+        if (!skipErase && model->data != nullptr) {
+            model->data->removeLora(this);
+        }
+
+        model->Unref();
+    }
+}
+
+Napi::Value AddonModelLora::GetFilePath(const Napi::CallbackInfo& info) {
+    return Napi::String::New(info.Env(), loraFilePath);
+}
+
+
+Napi::Value AddonModelLora::GetUsages(const Napi::CallbackInfo& info) {
+    return Napi::Number::From(info.Env(), usages);
+}
+
+void AddonModelLora::SetUsages(const Napi::CallbackInfo& info, const Napi::Value &value) {
+    usages = value.As<Napi::Number>().Uint32Value();
+}
+
+Napi::Value AddonModelLora::Dispose(const Napi::CallbackInfo& info) {
+    AddonModelLoraUnloadLoraWorker* worker = new AddonModelLoraUnloadLoraWorker(this->Env(), this);
+    worker->Queue();
+    return worker->GetPromise();
+}
+
+Napi::Value AddonModelLora::GetDisposed(const Napi::CallbackInfo& info) {
+    return Napi::Boolean::New(info.Env(), lora_adapter == nullptr);
+}
+
+void AddonModelLora::init(Napi::Object exports) {
+    exports.Set(
+        "AddonModelLora",
+        DefineClass(
+            exports.Env(),
+            "AddonModelLora",
+            {
+                InstanceAccessor("usages", &AddonModelLora::GetUsages, &AddonModelLora::SetUsages),
+                InstanceAccessor("filePath", &AddonModelLora::GetFilePath, nullptr),
+                InstanceAccessor("disposed", &AddonModelLora::GetDisposed, nullptr),
+                InstanceMethod("dispose", &AddonModelLora::Dispose),
+            }
+        )
+    );
+}

package/llama/addon/AddonModelLora.h

@@ -0,0 +1,28 @@
+#pragma once
+#include "llama.h"
+#include "napi.h"
+#include "addonGlobals.h"
+
+class AddonModelLora : public Napi::ObjectWrap<AddonModelLora> {
+    public:
+        AddonModel* model;
+        llama_lora_adapter * lora_adapter;
+        std::string loraFilePath;
+        uint32_t usages;
+
+        AddonModelLora(const Napi::CallbackInfo& info);
+        ~AddonModelLora();
+
+        void dispose(bool skipErase = false);
+
+        Napi::Value GetFilePath(const Napi::CallbackInfo& info);
+
+        Napi::Value GetUsages(const Napi::CallbackInfo& info);
+        void SetUsages(const Napi::CallbackInfo& info, const Napi::Value &value);
+
+        Napi::Value GetDisposed(const Napi::CallbackInfo& info);
+
+        Napi::Value Dispose(const Napi::CallbackInfo& info);
+
+        static void init(Napi::Object exports);
+};

package/llama/addon/addon.cpp

@@ -0,0 +1,217 @@
+#include "addonGlobals.h"
+#include "AddonModel.h"
+#include "AddonModelLora.h"
+#include "AddonGrammar.h"
+#include "AddonGrammarEvaluationState.h"
+#include "AddonContext.h"
+#include "globals/addonLog.h"
+#include "globals/addonProgress.h"
+#include "globals/getGpuInfo.h"
+
+bool backendInitialized = false;
+bool backendDisposed = false;
+
+Napi::Value systemInfo(const Napi::CallbackInfo& info) {
+    return Napi::String::From(info.Env(), llama_print_system_info());
+}
+
+Napi::Value addonGetSupportsGpuOffloading(const Napi::CallbackInfo& info) {
+    return Napi::Boolean::New(info.Env(), llama_supports_gpu_offload());
+}
+
+Napi::Value addonGetSupportsMmap(const Napi::CallbackInfo& info) {
+    return Napi::Boolean::New(info.Env(), llama_supports_mmap());
+}
+
+Napi::Value addonGetSupportsMlock(const Napi::CallbackInfo& info) {
+    return Napi::Boolean::New(info.Env(), llama_supports_mlock());
+}
+
+Napi::Value addonGetBlockSizeForGgmlType(const Napi::CallbackInfo& info) {
+    const int ggmlType = info[0].As<Napi::Number>().Int32Value();
+
+    if (ggmlType < 0 || ggmlType > GGML_TYPE_COUNT) {
+        return info.Env().Undefined();
+    }
+
+    const auto blockSize = ggml_blck_size(static_cast<ggml_type>(ggmlType));
+
+    return Napi::Number::New(info.Env(), blockSize);
+}
+
+Napi::Value addonGetTypeSizeForGgmlType(const Napi::CallbackInfo& info) {
+    const int ggmlType = info[0].As<Napi::Number>().Int32Value();
+
+    if (ggmlType < 0 || ggmlType > GGML_TYPE_COUNT) {
+        return info.Env().Undefined();
+    }
+
+    const auto typeSize = ggml_type_size(static_cast<ggml_type>(ggmlType));
+
+    return Napi::Number::New(info.Env(), typeSize);
+}
+
+Napi::Value addonGetConsts(const Napi::CallbackInfo& info) {
+    Napi::Object consts = Napi::Object::New(info.Env());
+    consts.Set("ggmlMaxDims", Napi::Number::New(info.Env(), GGML_MAX_DIMS));
+    consts.Set("ggmlTypeF16Size", Napi::Number::New(info.Env(), ggml_type_size(GGML_TYPE_F16)));
+    consts.Set("ggmlTypeF32Size", Napi::Number::New(info.Env(), ggml_type_size(GGML_TYPE_F32)));
+    consts.Set("ggmlTensorOverhead", Napi::Number::New(info.Env(), ggml_tensor_overhead()));
+    consts.Set("llamaMaxRngState", Napi::Number::New(info.Env(), LLAMA_MAX_RNG_STATE));
+    consts.Set("llamaPosSize", Napi::Number::New(info.Env(), sizeof(llama_pos)));
+    consts.Set("llamaSeqIdSize", Napi::Number::New(info.Env(), sizeof(llama_seq_id)));
+
+    return consts;
+}
+
+class AddonBackendLoadWorker : public Napi::AsyncWorker {
+    public:
+        AddonBackendLoadWorker(const Napi::Env& env)
+            : Napi::AsyncWorker(env, "AddonBackendLoadWorker"),
+              deferred(Napi::Promise::Deferred::New(env)) {
+        }
+        ~AddonBackendLoadWorker() {
+        }
+
+        Napi::Promise GetPromise() {
+            return deferred.Promise();
+        }
+
+    protected:
+        Napi::Promise::Deferred deferred;
+
+        void Execute() {
+            try {
+                llama_backend_init();
+
+                try {
+                    if (backendDisposed) {
+                        llama_backend_free();
+                    } else {
+                        backendInitialized = true;
+                    }
+                } catch (const std::exception& e) {
+                    SetError(e.what());
+                } catch(...) {
+                    SetError("Unknown error when calling \"llama_backend_free\"");
+                }
+            } catch (const std::exception& e) {
+                SetError(e.what());
+            } catch(...) {
+                SetError("Unknown error when calling \"llama_backend_init\"");
+            }
+        }
+        void OnOK() {
+            deferred.Resolve(Env().Undefined());
+        }
+        void OnError(const Napi::Error& err) {
+            deferred.Reject(err.Value());
+        }
+};
+
+
+class AddonBackendUnloadWorker : public Napi::AsyncWorker {
+    public:
+        AddonBackendUnloadWorker(const Napi::Env& env)
+            : Napi::AsyncWorker(env, "AddonBackendUnloadWorker"),
+              deferred(Napi::Promise::Deferred::New(env)) {
+        }
+        ~AddonBackendUnloadWorker() {
+        }
+
+        Napi::Promise GetPromise() {
+            return deferred.Promise();
+        }
+
+    protected:
+        Napi::Promise::Deferred deferred;
+
+        void Execute() {
+            try {
+                if (backendInitialized) {
+                    backendInitialized = false;
+                    llama_backend_free();
+                }
+            } catch (const std::exception& e) {
+                SetError(e.what());
+            } catch(...) {
+                SetError("Unknown error when calling \"llama_backend_free\"");
+            }
+        }
+        void OnOK() {
+            deferred.Resolve(Env().Undefined());
+        }
+        void OnError(const Napi::Error& err) {
+            deferred.Reject(err.Value());
+        }
+};
+
+Napi::Value addonInit(const Napi::CallbackInfo& info) {
+    if (backendInitialized) {
+        Napi::Promise::Deferred deferred = Napi::Promise::Deferred::New(info.Env());
+        deferred.Resolve(info.Env().Undefined());
+        return deferred.Promise();
+    }
+
+    AddonBackendLoadWorker* worker = new AddonBackendLoadWorker(info.Env());
+    worker->Queue();
+    return worker->GetPromise();
+}
+
+Napi::Value addonDispose(const Napi::CallbackInfo& info) {
+    if (backendDisposed) {
+        Napi::Promise::Deferred deferred = Napi::Promise::Deferred::New(info.Env());
+        deferred.Resolve(info.Env().Undefined());
+        return deferred.Promise();
+    }
+
+    backendDisposed = true;
+
+    AddonBackendUnloadWorker* worker = new AddonBackendUnloadWorker(info.Env());
+    worker->Queue();
+    return worker->GetPromise();
+}
+
+static void addonFreeLlamaBackend(Napi::Env env, int* data) {
+    if (backendDisposed) {
+        return;
+    }
+
+    backendDisposed = true;
+    if (backendInitialized) {
+        backendInitialized = false;
+        llama_backend_free();
+    }
+}
+
+Napi::Object registerCallback(Napi::Env env, Napi::Object exports) {
+    exports.DefineProperties({
+        Napi::PropertyDescriptor::Function("systemInfo", systemInfo),
+        Napi::PropertyDescriptor::Function("getSupportsGpuOffloading", addonGetSupportsGpuOffloading),
+        Napi::PropertyDescriptor::Function("getSupportsMmap", addonGetSupportsMmap),
+        Napi::PropertyDescriptor::Function("getSupportsMlock", addonGetSupportsMlock),
+        Napi::PropertyDescriptor::Function("getBlockSizeForGgmlType", addonGetBlockSizeForGgmlType),
+        Napi::PropertyDescriptor::Function("getTypeSizeForGgmlType", addonGetTypeSizeForGgmlType),
+        Napi::PropertyDescriptor::Function("getConsts", addonGetConsts),
+        Napi::PropertyDescriptor::Function("setLogger", setLogger),
+        Napi::PropertyDescriptor::Function("setLoggerLogLevel", setLoggerLogLevel),
+        Napi::PropertyDescriptor::Function("getGpuVramInfo", getGpuVramInfo),
+        Napi::PropertyDescriptor::Function("getGpuDeviceInfo", getGpuDeviceInfo),
+        Napi::PropertyDescriptor::Function("getGpuType", getGpuType),
+        Napi::PropertyDescriptor::Function("init", addonInit),
+        Napi::PropertyDescriptor::Function("dispose", addonDispose),
+    });
+    AddonModel::init(exports);
+    AddonModelLora::init(exports);
+    AddonGrammar::init(exports);
+    AddonGrammarEvaluationState::init(exports);
+    AddonContext::init(exports);
+
+    llama_log_set(addonLlamaCppLogCallback, nullptr);
+
+    exports.AddFinalizer(addonFreeLlamaBackend, static_cast<int*>(nullptr));
+
+    return exports;
+}
+
+NODE_API_MODULE(NODE_GYP_MODULE_NAME, registerCallback)

package/llama/addon/addonGlobals.cpp

@@ -0,0 +1,22 @@
+#include <sstream>
+#include <vector>
+#include "addonGlobals.h"
+#include "napi.h"
+
+void adjustNapiExternalMemoryAdd(Napi::Env env, uint64_t size) {
+    const uint64_t chunkSize = std::numeric_limits<int64_t>::max();
+    while (size > 0) {
+        int64_t adjustSize = std::min(size, chunkSize);
+        Napi::MemoryManagement::AdjustExternalMemory(env, adjustSize);
+        size -= adjustSize;
+    }
+}
+
+void adjustNapiExternalMemorySubtract(Napi::Env env, uint64_t size) {
+    const uint64_t chunkSize = std::numeric_limits<int64_t>::max();
+    while (size > 0) {
+        int64_t adjustSize = std::min(size, chunkSize);
+        Napi::MemoryManagement::AdjustExternalMemory(env, -adjustSize);
+        size -= adjustSize;
+    }
+}

package/llama/addon/addonGlobals.h

@@ -0,0 +1,12 @@
+#pragma once
+#include "napi.h"
+
+class AddonModel;
+class AddonModelLora;
+class AddonModelData;
+class AddonContext;
+class AddonGrammar;
+class AddonGrammarEvaluationState;
+
+void adjustNapiExternalMemoryAdd(Napi::Env env, uint64_t size);
+void adjustNapiExternalMemorySubtract(Napi::Env env, uint64_t size);

package/llama/addon/globals/addonLog.cpp

@@ -0,0 +1,135 @@
+#include <sstream>
+
+#include "addonLog.h"
+
+AddonThreadSafeLogCallbackFunction addonThreadSafeLoggerCallback;
+bool addonJsLoggerCallbackSet = false;
+int addonLoggerLogLevel = 5;
+
+static int addonGetGgmlLogLevelNumber(ggml_log_level level) {
+    switch (level) {
+        case GGML_LOG_LEVEL_ERROR: return 2;
+        case GGML_LOG_LEVEL_WARN: return 3;
+        case GGML_LOG_LEVEL_INFO: return 4;
+        case GGML_LOG_LEVEL_DEBUG: return 5;
+    }
+
+    return 1;
+}
+
+void addonCallJsLogCallback(
+    Napi::Env env, Napi::Function callback, AddonThreadSafeLogCallbackFunctionContext* context, addon_logger_log* data
+) {
+    bool called = false;
+
+    if (env != nullptr && callback != nullptr && addonJsLoggerCallbackSet) {
+        try {
+            callback.Call({
+                Napi::Number::New(env, data->logLevelNumber),
+                Napi::String::New(env, data->stringStream->str()),
+            });
+            called = true;
+        } catch (const Napi::Error& e) {
+            called = false;
+        }
+    }
+
+    if (!called && data != nullptr) {
+        if (data->logLevelNumber == 2) {
+            fputs(data->stringStream->str().c_str(), stderr);
+            fflush(stderr);
+        } else {
+            fputs(data->stringStream->str().c_str(), stdout);
+            fflush(stdout);
+        }
+    }
+
+    if (data != nullptr) {
+        delete data->stringStream;
+        delete data;
+    }
+}
+
+void addonLlamaCppLogCallback(ggml_log_level level, const char* text, void* user_data) {
+    int logLevelNumber = addonGetGgmlLogLevelNumber(level);
+
+    if (logLevelNumber > addonLoggerLogLevel) {
+        return;
+    }
+
+    if (addonJsLoggerCallbackSet) {
+        std::stringstream* stringStream = new std::stringstream();
+        if (text != nullptr) {
+            *stringStream << text;
+        }
+
+        addon_logger_log* data = new addon_logger_log {
+            logLevelNumber,
+            stringStream,
+        };
+
+        auto status = addonThreadSafeLoggerCallback.NonBlockingCall(data);
+
+        if (status == napi_ok) {
+            return;
+        } else {
+            delete stringStream;
+            delete data;
+        }
+    }
+
+    if (text != nullptr) {
+        if (level == 2) {
+            fputs(text, stderr);
+            fflush(stderr);
+        } else {
+            fputs(text, stdout);
+            fflush(stdout);
+        }
+    }
+}
+
+Napi::Value setLogger(const Napi::CallbackInfo& info) {
+    if (info.Length() < 1 || !info[0].IsFunction()) {
+        if (addonJsLoggerCallbackSet) {
+            addonJsLoggerCallbackSet = false;
+            addonThreadSafeLoggerCallback.Release();
+        }
+
+        return info.Env().Undefined();
+    }
+
+    auto addonLoggerJSCallback = info[0].As<Napi::Function>();
+    AddonThreadSafeLogCallbackFunctionContext* context = new Napi::Reference<Napi::Value>(Napi::Persistent(info.This()));
+    addonThreadSafeLoggerCallback = AddonThreadSafeLogCallbackFunction::New(
+        info.Env(),
+        addonLoggerJSCallback,
+        "loggerCallback",
+        0,
+        1,
+        context,
+        [](Napi::Env, void*, AddonThreadSafeLogCallbackFunctionContext* ctx) {
+            addonJsLoggerCallbackSet = false;
+
+            delete ctx;
+        }
+    );
+    addonJsLoggerCallbackSet = true;
+
+    // prevent blocking the main node process from exiting due to active resources
+    addonThreadSafeLoggerCallback.Unref(info.Env());
+
+    return info.Env().Undefined();
+}
+
+Napi::Value setLoggerLogLevel(const Napi::CallbackInfo& info) {
+    if (info.Length() < 1 || !info[0].IsNumber()) {
+        addonLoggerLogLevel = 5;
+
+        return info.Env().Undefined();
+    }
+
+    addonLoggerLogLevel = info[0].As<Napi::Number>().Int32Value();
+
+    return info.Env().Undefined();
+}

package/llama/addon/globals/addonLog.h

@@ -0,0 +1,21 @@
+#pragma once
+#include "llama.h"
+#include "napi.h"
+
+struct addon_logger_log {
+    public:
+        const int logLevelNumber;
+        const std::stringstream* stringStream;
+};
+
+void addonLlamaCppLogCallback(ggml_log_level level, const char* text, void* user_data);
+
+using AddonThreadSafeLogCallbackFunctionContext = Napi::Reference<Napi::Value>;
+void addonCallJsLogCallback(
+    Napi::Env env, Napi::Function callback, AddonThreadSafeLogCallbackFunctionContext* context, addon_logger_log* data
+);
+using AddonThreadSafeLogCallbackFunction =
+    Napi::TypedThreadSafeFunction<AddonThreadSafeLogCallbackFunctionContext, addon_logger_log, addonCallJsLogCallback>;
+
+Napi::Value setLogger(const Napi::CallbackInfo& info);
+Napi::Value setLoggerLogLevel(const Napi::CallbackInfo& info);

package/llama/addon/globals/addonProgress.cpp

@@ -0,0 +1,15 @@
+#include "addonProgress.h"
+
+void addonCallJsProgressCallback(
+    Napi::Env env, Napi::Function callback, AddonThreadSafeProgressCallbackFunctionContext* context, addon_progress_event* data
+) {
+    if (env != nullptr && callback != nullptr) {
+        try {
+            callback.Call({Napi::Number::New(env, data->progress)});
+        } catch (const Napi::Error& e) {}
+    }
+
+    if (data != nullptr) {
+        delete data;
+    }
+}

package/llama/addon/globals/addonProgress.h

@@ -0,0 +1,15 @@
+#pragma once
+#include "napi.h"
+
+struct addon_progress_event {
+    public:
+        const float progress;
+};
+
+using AddonThreadSafeProgressCallbackFunctionContext = Napi::Reference<Napi::Value>;
+void addonCallJsProgressCallback(
+    Napi::Env env, Napi::Function callback, AddonThreadSafeProgressCallbackFunctionContext* context, addon_progress_event* data
+);
+using AddonThreadSafeProgressEventCallbackFunction =
+    Napi::TypedThreadSafeFunction<AddonThreadSafeProgressCallbackFunctionContext, addon_progress_event, addonCallJsProgressCallback>;
+