cactus-react-native 0.2.3 → 0.2.4

package/README.md

@@ -7,7 +7,7 @@ Run LLMs, VLMs, and TTS models directly on mobile devices.
 ```json
 {
   "dependencies": {
-    "cactus-react-native": "^0.2.1",
+    "cactus-react-native": "^0.2.4",
     "react-native-fs": "^2.20.0"
   }
 }

package/android/src/main/java/com/cactus/Cactus.java

@@ -598,6 +598,41 @@ public class Cactus implements LifecycleEventListener {
     tasks.put(task, "getLoadedLoraAdapters-" + contextId);
   }
 
+  public void rewind(double id, Promise promise) {
+    final int contextId = (int) id;
+    AsyncTask task = new AsyncTask<Void, Void, Void>() {
+      private Exception exception;
+
+      @Override
+      protected Void doInBackground(Void... voids) {
+        try {
+          LlamaContext context = contexts.get(contextId);
+          if (context == null) {
+            throw new Exception("Context not found");
+          }
+          if (context.isPredicting()) {
+            throw new Exception("Context is busy");
+          }
+          context.rewind();
+        } catch (Exception e) {
+          exception = e;
+        }
+        return null;
+      }
+
+      @Override
+      protected void onPostExecute(Void result) {
+        if (exception != null) {
+          promise.reject(exception);
+          return;
+        }
+        promise.resolve(null);
+        tasks.remove(this);
+      }
+    }.executeOnExecutor(singleThreadExecutor);
+    tasks.put(task, "rewind-" + contextId);
+  }
+
   public void releaseContext(double id, Promise promise) {
     final int contextId = (int) id;
     AsyncTask task = new AsyncTask<Void, Void, Void>() {

package/android/src/main/java/com/cactus/LlamaContext.java

@@ -398,6 +398,10 @@ public class LlamaContext {
     return getLoadedLoraAdapters(this.context);
   }
 
+  public void rewind() {
+    rewind(this.context);
+  }
+
   public void release() {
     freeContext(context);
   }
@@ -747,6 +751,7 @@ public class LlamaContext {
   protected static native int applyLoraAdapters(long contextPtr, ReadableArray loraAdapters);
   protected static native void removeLoraAdapters(long contextPtr);
   protected static native WritableArray getLoadedLoraAdapters(long contextPtr);
+  protected static native void rewind(long contextPtr);
   protected static native void freeContext(long contextPtr);
   protected static native void setupLog(NativeLogCallback logCallback);
   protected static native void unsetLog();

package/android/src/main/jni.cpp

@@ -307,6 +307,7 @@ Java_com_cactus_LlamaContext_initContext(
     llama->is_load_interrupted = false;
     llama->loading_progress = 0;
 
+    callback_context *cb_ctx = nullptr;
     if (load_progress_callback != nullptr) {
         defaultParams.progress_callback = [](float progress, void * user_data) {
             callback_context *cb_ctx = (callback_context *)user_data;
@@ -323,7 +324,7 @@ Java_com_cactus_LlamaContext_initContext(
             return !llama->is_load_interrupted;
         };
 
-        callback_context *cb_ctx = new callback_context;
+        cb_ctx = new callback_context;
         cb_ctx->env = env;
         cb_ctx->llama = llama;
         cb_ctx->callback = env->NewGlobalRef(load_progress_callback);
@@ -339,20 +340,30 @@ Java_com_cactus_LlamaContext_initContext(
     env->ReleaseStringUTFChars(cache_type_v, cache_type_v_chars);
 
     LOGI("[CACTUS] is_model_loaded %s", (is_model_loaded ? "true" : "false"));
-    if (is_model_loaded) {
-        if (embedding && llama_model_has_encoder(llama->model) && llama_model_has_decoder(llama->model)) {
-            LOGI("[CACTUS] computing embeddings in encoder-decoder models is not supported");
-            llama_free(llama->ctx);
-            return -1;
+
+    if (!is_model_loaded) {
+        llama_free(llama->ctx);
+        if (cb_ctx) {
+            env->DeleteGlobalRef(cb_ctx->callback);
+            delete cb_ctx;
         }
-        context_map[(long) llama->ctx] = llama;
-    } else {
-        LOGE("[CACTUS] Failed to load model from path: %s", model_path_chars);
+        delete llama;
+        return -1;
+    }
+
+    if (embedding && llama_model_has_encoder(llama->model) && llama_model_has_decoder(llama->model)) {
+        LOGI("[CACTUS] computing embeddings in encoder-decoder models is not supported");
         llama_free(llama->ctx);
+        if (cb_ctx) {
+            env->DeleteGlobalRef(cb_ctx->callback);
+            delete cb_ctx;
+        }
         delete llama;
         return -1;
     }
 
+    context_map[(long) llama->ctx] = llama;
+
     std::vector<common_adapter_lora_info> lora;
     const char *lora_chars = env->GetStringUTFChars(lora_str, nullptr);
     if (lora_chars != nullptr && lora_chars[0] != '\0') {
@@ -381,11 +392,20 @@ Java_com_cactus_LlamaContext_initContext(
     env->ReleaseStringUTFChars(lora_str, lora_chars);
     int result = llama->applyLoraAdapters(lora);
     if (result != 0) {
-      LOGE("[CACTUS] Failed to apply lora adapters");
-      llama_free(llama->ctx);
-      context_map.erase((long) llama->ctx);
-      delete llama;
-      return -1;
+        LOGI("[Cactus] Failed to apply lora adapters");
+        context_map.erase((long) llama->ctx);
+        llama_free(llama->ctx);
+        if (cb_ctx) {
+            env->DeleteGlobalRef(cb_ctx->callback);
+            delete cb_ctx;
+        }
+        delete llama;
+        return -1;
+    }
+
+    if (cb_ctx) {
+        env->DeleteGlobalRef(cb_ctx->callback);
+        delete cb_ctx;
     }
 
     return reinterpret_cast<jlong>(llama->ctx);
@@ -1339,6 +1359,17 @@ Java_com_cactus_LlamaContext_getLoadedLoraAdapters(
     return result;
 }
 
+JNIEXPORT void JNICALL
+Java_com_cactus_LlamaContext_rewind(
+        JNIEnv *env, jobject thiz, jlong context_ptr) {
+    UNUSED(env);
+    UNUSED(thiz);
+    auto llama = context_map[(long) context_ptr];
+    if (llama != nullptr) {
+        llama->rewind();
+    }
+}
+
 JNIEXPORT void JNICALL
 Java_com_cactus_LlamaContext_freeContext(
         JNIEnv *env, jobject thiz, jlong context_ptr) {

package/android/src/newarch/java/com/cactus/CactusModule.java

@@ -189,6 +189,11 @@ public class CactusModule extends NativeCactusSpec {
     cactus.releaseVocoder(id, promise);
   }
 
+  @ReactMethod
+  public void rewind(double id, Promise promise) {
+    cactus.rewind(id, promise);
+  }
+
   @ReactMethod
   public void releaseContext(double id, Promise promise) {
     cactus.releaseContext(id, promise);

package/android/src/oldarch/java/com/cactus/CactusModule.java

@@ -188,6 +188,11 @@ public class CactusModule extends ReactContextBaseJavaModule {
     cactus.releaseVocoder(id, promise);
   }
 
+  @ReactMethod
+  public void rewind(double id, Promise promise) {
+    cactus.rewind(id, promise);
+  }
+
   @ReactMethod
   public void releaseContext(double id, Promise promise) {
     cactus.releaseContext(id, promise);

package/ios/Cactus.mm

@@ -473,6 +473,27 @@ RCT_EXPORT_METHOD(multimodalCompletion:(double)contextId
     });
 }
 
+RCT_EXPORT_METHOD(rewind:(double)contextId
+                 withResolver:(RCTPromiseResolveBlock)resolve
+                 withRejecter:(RCTPromiseRejectBlock)reject)
+{
+    CactusContext *context = llamaContexts[[NSNumber numberWithDouble:contextId]];
+    if (context == nil) {
+        reject(@"llama_error", @"Context not found", nil);
+        return;
+    }
+    if ([context isPredicting]) {
+        reject(@"llama_error", @"Context is busy", nil);
+        return;
+    }
+    @try {
+        [context rewind];
+        resolve(nil);
+    } @catch (NSException *exception) {
+        reject(@"llama_cpp_error", exception.reason, nil);
+    }
+}
+
 RCT_EXPORT_METHOD(releaseContext:(double)contextId
                  withResolver:(RCTPromiseResolveBlock)resolve
                  withRejecter:(RCTPromiseRejectBlock)reject)

package/ios/CactusContext.h

@@ -74,6 +74,7 @@
 - (NSArray *)decodeAudioTokens:(NSArray *)tokens;
 - (NSDictionary *)getDeviceInfo;
 - (void)releaseVocoder;
+- (void)rewind;
 
 - (void)invalidate;
 

package/ios/CactusContext.mm

@@ -962,6 +962,10 @@
     llama->releaseVocoder();
 }
 
+- (void)rewind {
+    llama->rewind();
+}
+
 - (void)invalidate {
     delete llama;
     // llama_backend_free();

package/ios/cactus.xcframework/ios-arm64/cactus.framework/Headers/cactus_ffi.h

@@ -182,7 +182,6 @@ CACTUS_FFI_EXPORT cactus_float_array_c_t cactus_decode_audio_tokens_c(cactus_con
 
 CACTUS_FFI_EXPORT void cactus_release_vocoder_c(cactus_context_handle_t handle);
 
-// **HIGH PRIORITY ADDITIONS**
 
 typedef struct {
     const char* path;
@@ -204,19 +203,13 @@ typedef struct {
     double tg_std;
 } cactus_bench_result_c_t;
 
-// **HIGH PRIORITY: Benchmarking**
 CACTUS_FFI_EXPORT cactus_bench_result_c_t cactus_bench_c(cactus_context_handle_t handle, int pp, int tg, int pl, int nr);
-
-// **HIGH PRIORITY: LoRA Adapter Support**
 CACTUS_FFI_EXPORT int cactus_apply_lora_adapters_c(cactus_context_handle_t handle, const cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_remove_lora_adapters_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cactus_context_handle_t handle);
-
-// **HIGH PRIORITY: Chat Template Support**
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
-// **ADVANCED: Chat with Jinja and Tools Support**
 typedef struct {
     char* prompt;
     char* json_schema;
@@ -235,28 +228,23 @@ CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
     const char* tool_choice
 );
 
-// **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
-// **COMPLETION CONTROL**
 CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
 
-// **TOKEN PROCESSING**
 CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
 CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
 
-// **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT char* cactus_get_model_desc_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_size_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t handle);
 
-// Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);

package/ios/cactus.xcframework/ios-arm64_x86_64-simulator/cactus.framework/Headers/cactus_ffi.h

@@ -182,7 +182,6 @@ CACTUS_FFI_EXPORT cactus_float_array_c_t cactus_decode_audio_tokens_c(cactus_con
 
 CACTUS_FFI_EXPORT void cactus_release_vocoder_c(cactus_context_handle_t handle);
 
-// **HIGH PRIORITY ADDITIONS**
 
 typedef struct {
     const char* path;
@@ -204,19 +203,13 @@ typedef struct {
     double tg_std;
 } cactus_bench_result_c_t;
 
-// **HIGH PRIORITY: Benchmarking**
 CACTUS_FFI_EXPORT cactus_bench_result_c_t cactus_bench_c(cactus_context_handle_t handle, int pp, int tg, int pl, int nr);
-
-// **HIGH PRIORITY: LoRA Adapter Support**
 CACTUS_FFI_EXPORT int cactus_apply_lora_adapters_c(cactus_context_handle_t handle, const cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_remove_lora_adapters_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cactus_context_handle_t handle);
-
-// **HIGH PRIORITY: Chat Template Support**
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
-// **ADVANCED: Chat with Jinja and Tools Support**
 typedef struct {
     char* prompt;
     char* json_schema;
@@ -235,28 +228,23 @@ CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
     const char* tool_choice
 );
 
-// **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
-// **COMPLETION CONTROL**
 CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
 
-// **TOKEN PROCESSING**
 CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
 CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
 
-// **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT char* cactus_get_model_desc_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_size_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t handle);
 
-// Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);

package/ios/cactus.xcframework/tvos-arm64/cactus.framework/Headers/cactus_ffi.h

@@ -182,7 +182,6 @@ CACTUS_FFI_EXPORT cactus_float_array_c_t cactus_decode_audio_tokens_c(cactus_con
 
 CACTUS_FFI_EXPORT void cactus_release_vocoder_c(cactus_context_handle_t handle);
 
-// **HIGH PRIORITY ADDITIONS**
 
 typedef struct {
     const char* path;
@@ -204,19 +203,13 @@ typedef struct {
     double tg_std;
 } cactus_bench_result_c_t;
 
-// **HIGH PRIORITY: Benchmarking**
 CACTUS_FFI_EXPORT cactus_bench_result_c_t cactus_bench_c(cactus_context_handle_t handle, int pp, int tg, int pl, int nr);
-
-// **HIGH PRIORITY: LoRA Adapter Support**
 CACTUS_FFI_EXPORT int cactus_apply_lora_adapters_c(cactus_context_handle_t handle, const cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_remove_lora_adapters_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cactus_context_handle_t handle);
-
-// **HIGH PRIORITY: Chat Template Support**
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
-// **ADVANCED: Chat with Jinja and Tools Support**
 typedef struct {
     char* prompt;
     char* json_schema;
@@ -235,28 +228,23 @@ CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
     const char* tool_choice
 );
 
-// **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
-// **COMPLETION CONTROL**
 CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
 
-// **TOKEN PROCESSING**
 CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
 CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
 
-// **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT char* cactus_get_model_desc_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_size_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t handle);
 
-// Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);

package/ios/cactus.xcframework/tvos-arm64_x86_64-simulator/cactus.framework/Headers/cactus_ffi.h

@@ -182,7 +182,6 @@ CACTUS_FFI_EXPORT cactus_float_array_c_t cactus_decode_audio_tokens_c(cactus_con
 
 CACTUS_FFI_EXPORT void cactus_release_vocoder_c(cactus_context_handle_t handle);
 
-// **HIGH PRIORITY ADDITIONS**
 
 typedef struct {
     const char* path;
@@ -204,19 +203,13 @@ typedef struct {
     double tg_std;
 } cactus_bench_result_c_t;
 
-// **HIGH PRIORITY: Benchmarking**
 CACTUS_FFI_EXPORT cactus_bench_result_c_t cactus_bench_c(cactus_context_handle_t handle, int pp, int tg, int pl, int nr);
-
-// **HIGH PRIORITY: LoRA Adapter Support**
 CACTUS_FFI_EXPORT int cactus_apply_lora_adapters_c(cactus_context_handle_t handle, const cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_remove_lora_adapters_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT cactus_lora_adapters_c_t cactus_get_loaded_lora_adapters_c(cactus_context_handle_t handle);
-
-// **HIGH PRIORITY: Chat Template Support**
 CACTUS_FFI_EXPORT bool cactus_validate_chat_template_c(cactus_context_handle_t handle, bool use_jinja, const char* name);
 CACTUS_FFI_EXPORT char* cactus_get_formatted_chat_c(cactus_context_handle_t handle, const char* messages, const char* chat_template);
 
-// **ADVANCED: Chat with Jinja and Tools Support**
 typedef struct {
     char* prompt;
     char* json_schema;
@@ -235,28 +228,23 @@ CACTUS_FFI_EXPORT cactus_chat_result_c_t cactus_get_formatted_chat_with_jinja_c(
     const char* tool_choice
 );
 
-// **HIGH PRIORITY: Context Management**
 CACTUS_FFI_EXPORT void cactus_rewind_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT bool cactus_init_sampling_c(cactus_context_handle_t handle);
 
-// **COMPLETION CONTROL**
 CACTUS_FFI_EXPORT void cactus_begin_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_end_completion_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT void cactus_load_prompt_with_media_c(cactus_context_handle_t handle, const char** media_paths, int media_count);
 
-// **TOKEN PROCESSING**
 CACTUS_FFI_EXPORT int cactus_do_completion_step_c(cactus_context_handle_t handle, char** token_text);
 CACTUS_FFI_EXPORT size_t cactus_find_stopping_strings_c(cactus_context_handle_t handle, const char* text, size_t last_token_size, int stop_type);
 
-// **HIGH PRIORITY: Model Information**
 CACTUS_FFI_EXPORT int32_t cactus_get_n_ctx_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int32_t cactus_get_n_embd_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT char* cactus_get_model_desc_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_size_c(cactus_context_handle_t handle);
 CACTUS_FFI_EXPORT int64_t cactus_get_model_params_c(cactus_context_handle_t handle);
 
-// Memory management functions
 CACTUS_FFI_EXPORT void cactus_free_bench_result_members_c(cactus_bench_result_c_t* result);
 CACTUS_FFI_EXPORT void cactus_free_lora_adapters_c(cactus_lora_adapters_c_t* adapters);
 CACTUS_FFI_EXPORT void cactus_free_chat_result_members_c(cactus_chat_result_c_t* result);

package/lib/commonjs/NativeCactus.js

@@ -5,6 +5,5 @@ Object.defineProperty(exports, "__esModule", {
 });
 exports.default = void 0;
 var _reactNative = require("react-native");
-// New TTS/Audio types
 var _default = exports.default = _reactNative.TurboModuleRegistry.get('Cactus');
 //# sourceMappingURL=NativeCactus.js.map

package/lib/commonjs/NativeCactus.js.map

@@ -1 +1 @@
-{"version":3,"names":["_reactNative","require","_default","exports","default","TurboModuleRegistry","get"],"sourceRoot":"../../src","sources":["NativeCactus.ts"],"mappings":";;;;;;AACA,IAAAA,YAAA,GAAAC,OAAA;AAkSA;AAAA,IAAAC,QAAA,GAAAC,OAAA,CAAAC,OAAA,GAkMeC,gCAAmB,CAACC,GAAG,CAAO,QAAQ,CAAC","ignoreList":[]}
+{"version":3,"names":["_reactNative","require","_default","exports","default","TurboModuleRegistry","get"],"sourceRoot":"../../src","sources":["NativeCactus.ts"],"mappings":";;;;;;AACA,IAAAA,YAAA,GAAAC,OAAA;AAAkD,IAAAC,QAAA,GAAAC,OAAA,CAAAC,OAAA,GA2TnCC,gCAAmB,CAACC,GAAG,CAAO,QAAQ,CAAC","ignoreList":[]}

package/lib/commonjs/chat.js

@@ -3,6 +3,7 @@
 Object.defineProperty(exports, "__esModule", {
   value: true
 });
+exports.ConversationHistoryManager = void 0;
 exports.formatChat = formatChat;
 function formatChat(messages) {
   const chat = [];
@@ -34,4 +35,36 @@ function formatChat(messages) {
   });
   return chat;
 }
+class ConversationHistoryManager {
+  history = [];
+  processNewMessages(fullMessageHistory) {
+    let divergent = fullMessageHistory.length < this.history.length;
+    if (!divergent) {
+      for (let i = 0; i < this.history.length; i++) {
+        if (JSON.stringify(this.history[i]) !== JSON.stringify(fullMessageHistory[i])) {
+          divergent = true;
+          break;
+        }
+      }
+    }
+    if (divergent) {
+      return {
+        newMessages: fullMessageHistory,
+        requiresReset: true
+      };
+    }
+    const newMessages = fullMessageHistory.slice(this.history.length);
+    return {
+      newMessages,
+      requiresReset: false
+    };
+  }
+  update(newMessages, assistantResponse) {
+    this.history.push(...newMessages, assistantResponse);
+  }
+  reset() {
+    this.history = [];
+  }
+}
+exports.ConversationHistoryManager = ConversationHistoryManager;
 //# sourceMappingURL=chat.js.map

package/lib/commonjs/chat.js.map

@@ -1 +1 @@
-{"version":3,"names":["formatChat","messages","chat","forEach","currMsg","role","content","Array","isArray","part","text","TypeError","Error","push"],"sourceRoot":"../../src","sources":["chat.ts"],"mappings":";;;;;;AAWO,SAASA,UAAUA,CACxBC,QAAsC,EACZ;EAC1B,MAAMC,IAA8B,GAAG,EAAE;EAEzCD,QAAQ,CAACE,OAAO,CAAEC,OAAO,IAAK;IAC5B,MAAMC,IAAY,GAAGD,OAAO,CAACC,IAAI,IAAI,EAAE;IAEvC,IAAIC,OAAe,GAAG,EAAE;IACxB,IAAI,SAAS,IAAIF,OAAO,EAAE;MACxB,IAAI,OAAOA,OAAO,CAACE,OAAO,KAAK,QAAQ,EAAE;QACvC;QAAC,CAAC;UAAEA;QAAQ,CAAC,GAAGF,OAAO;MACzB,CAAC,MAAM,IAAIG,KAAK,CAACC,OAAO,CAACJ,OAAO,CAACE,OAAO,CAAC,EAAE;QACzCF,OAAO,CAACE,OAAO,CAACH,OAAO,CAAEM,IAAI,IAAK;UAChC,IAAI,MAAM,IAAIA,IAAI,EAAE;YAClBH,OAAO,IAAI,GAAGA,OAAO,GAAG,IAAI,GAAG,EAAE,GAAGG,IAAI,CAACC,IAAI,EAAE;UACjD;QACF,CAAC,CAAC;MACJ,CAAC,MAAM;QACL,MAAM,IAAIC,SAAS,CACjB,kFACF,CAAC;MACH;IACF,CAAC,MAAM;MACL,MAAM,IAAIC,KAAK,CACb,6EACF,CAAC;IACH;IAEAV,IAAI,CAACW,IAAI,CAAC;MAAER,IAAI;MAAEC;IAAQ,CAAC,CAAC;EAC9B,CAAC,CAAC;EACF,OAAOJ,IAAI;AACb","ignoreList":[]}
+{"version":3,"names":["formatChat","messages","chat","forEach","currMsg","role","content","Array","isArray","part","text","TypeError","Error","push","ConversationHistoryManager","history","processNewMessages","fullMessageHistory","divergent","length","i","JSON","stringify","newMessages","requiresReset","slice","update","assistantResponse","reset","exports"],"sourceRoot":"../../src","sources":["chat.ts"],"mappings":";;;;;;;AAWO,SAASA,UAAUA,CACxBC,QAAsC,EACZ;EAC1B,MAAMC,IAA8B,GAAG,EAAE;EAEzCD,QAAQ,CAACE,OAAO,CAAEC,OAAO,IAAK;IAC5B,MAAMC,IAAY,GAAGD,OAAO,CAACC,IAAI,IAAI,EAAE;IAEvC,IAAIC,OAAe,GAAG,EAAE;IACxB,IAAI,SAAS,IAAIF,OAAO,EAAE;MACxB,IAAI,OAAOA,OAAO,CAACE,OAAO,KAAK,QAAQ,EAAE;QACvC;QAAC,CAAC;UAAEA;QAAQ,CAAC,GAAGF,OAAO;MACzB,CAAC,MAAM,IAAIG,KAAK,CAACC,OAAO,CAACJ,OAAO,CAACE,OAAO,CAAC,EAAE;QACzCF,OAAO,CAACE,OAAO,CAACH,OAAO,CAAEM,IAAI,IAAK;UAChC,IAAI,MAAM,IAAIA,IAAI,EAAE;YAClBH,OAAO,IAAI,GAAGA,OAAO,GAAG,IAAI,GAAG,EAAE,GAAGG,IAAI,CAACC,IAAI,EAAE;UACjD;QACF,CAAC,CAAC;MACJ,CAAC,MAAM;QACL,MAAM,IAAIC,SAAS,CACjB,kFACF,CAAC;MACH;IACF,CAAC,MAAM;MACL,MAAM,IAAIC,KAAK,CACb,6EACF,CAAC;IACH;IAEAV,IAAI,CAACW,IAAI,CAAC;MAAER,IAAI;MAAEC;IAAQ,CAAC,CAAC;EAC9B,CAAC,CAAC;EACF,OAAOJ,IAAI;AACb;AAOO,MAAMY,0BAA0B,CAAC;EAC9BC,OAAO,GAAiC,EAAE;EAE3CC,kBAAkBA,CACvBC,kBAAgD,EAC7B;IACnB,IAAIC,SAAS,GAAGD,kBAAkB,CAACE,MAAM,GAAG,IAAI,CAACJ,OAAO,CAACI,MAAM;IAC/D,IAAI,CAACD,SAAS,EAAE;MACd,KAAK,IAAIE,CAAC,GAAG,CAAC,EAAEA,CAAC,GAAG,IAAI,CAACL,OAAO,CAACI,MAAM,EAAEC,CAAC,EAAE,EAAE;QAC5C,IAAIC,IAAI,CAACC,SAAS,CAAC,IAAI,CAACP,OAAO,CAACK,CAAC,CAAC,CAAC,KAAKC,IAAI,CAACC,SAAS,CAACL,kBAAkB,CAACG,CAAC,CAAC,CAAC,EAAE;UAC7EF,SAAS,GAAG,IAAI;UAChB;QACF;MACF;IACF;IAEA,IAAIA,SAAS,EAAE;MACb,OAAO;QAAEK,WAAW,EAAEN,kBAAkB;QAAEO,aAAa,EAAE;MAAK,CAAC;IACjE;IAEA,MAAMD,WAAW,GAAGN,kBAAkB,CAACQ,KAAK,CAAC,IAAI,CAACV,OAAO,CAACI,MAAM,CAAC;IACjE,OAAO;MAAEI,WAAW;MAAEC,aAAa,EAAE;IAAM,CAAC;EAC9C;EAEOE,MAAMA,CACXH,WAAyC,EACzCI,iBAA6C,EAC7C;IACA,IAAI,CAACZ,OAAO,CAACF,IAAI,CAAC,GAAGU,WAAW,EAAEI,iBAAiB,CAAC;EACtD;EAEOC,KAAKA,CAAA,EAAG;IACb,IAAI,CAACb,OAAO,GAAG,EAAE;EACnB;AACF;AAACc,OAAA,CAAAf,0BAAA,GAAAA,0BAAA","ignoreList":[]}

package/lib/commonjs/index.js

@@ -27,8 +27,6 @@ var _exportNames = {
   releaseVocoder: true,
   tokenize: true,
   getDeviceInfo: true,
-  SchemaGrammarConverter: true,
-  convertJsonSchemaToGrammar: true,
   Tools: true,
   CactusLM: true,
   CactusVLM: true,
@@ -53,12 +51,6 @@ Object.defineProperty(exports, "CactusVLM", {
   }
 });
 exports.LlamaContext = void 0;
-Object.defineProperty(exports, "SchemaGrammarConverter", {
-  enumerable: true,
-  get: function () {
-    return _grammar.SchemaGrammarConverter;
-  }
-});
 Object.defineProperty(exports, "Tools", {
   enumerable: true,
   get: function () {
@@ -66,12 +58,6 @@ Object.defineProperty(exports, "Tools", {
   }
 });
 exports.addNativeLogListener = addNativeLogListener;
-Object.defineProperty(exports, "convertJsonSchemaToGrammar", {
-  enumerable: true,
-  get: function () {
-    return _grammar.convertJsonSchemaToGrammar;
-  }
-});
 exports.initContext = exports.getTTSType = exports.getFormattedAudioCompletion = exports.getDeviceInfo = exports.getAudioCompletionGuideTokens = exports.decodeAudioTokens = void 0;
 exports.initLlama = initLlama;
 exports.isVocoderEnabled = exports.isMultimodalSupportVision = exports.isMultimodalSupportAudio = exports.isMultimodalEnabled = exports.initVocoder = exports.initMultimodal = void 0;
@@ -84,7 +70,6 @@ exports.toggleNativeLog = toggleNativeLog;
 exports.tokenize = void 0;
 var _reactNative = require("react-native");
 var _NativeCactus = _interopRequireDefault(require("./NativeCactus"));
-var _grammar = require("./grammar");
 var _chat = require("./chat");
 var _tools = require("./tools");
 var _telemetry = require("./telemetry");
@@ -202,7 +187,6 @@ class LlamaContext {
       return this.completion(params, callback);
     }
     if (recursionCount >= recursionLimit) {
-      // console.log(`Recursion limit reached (${recursionCount}/${recursionLimit}), returning default completion`)
       return this.completion({
         ...params,
         jinja: true,
@@ -211,15 +195,12 @@ class LlamaContext {
     }
     const messages = [...params.messages]; // avoid mutating the original messages
 
-    // console.log('Calling completion...')
     const result = await this.completion({
       ...params,
       messages: messages,
       jinja: true,
       tools: params.tools.getSchemas()
     }, callback);
-    // console.log('Completion result:', result);
-
     const {
       toolCalled,
       toolName,
@@ -240,9 +221,6 @@ class LlamaContext {
         tool_call_id: toolCallId
       };
       messages.push(toolMessage);
-
-      // console.log('Messages being sent to next completion:', JSON.stringify(messages, null, 2));
-
       return await this.completionWithTools({
         ...params,
         messages: messages
@@ -370,7 +348,6 @@ class LlamaContext {
     return _NativeCactus.default.releaseContext(this.id);
   }
   async rewind() {
-    // @ts-ignore
     return _NativeCactus.default.rewind(this.id);
   }
 }