feat: add chat_template for override default on init

android/src/main/java/com/rnllama/LlamaContext.java

@@ -44,6 +44,8 @@ public LlamaContext(int id, ReactApplicationContext reactContext, ReadableMap pa
     this.context = initContext(
       // String model,
       params.getString("model"),
+      // String chat_template,
+      params.hasKey("chat_template") ? params.getString("chat_template") : "",
       // boolean embedding,
       params.hasKey("embedding") ? params.getBoolean("embedding") : false,
       // int embd_normalize,
@@ -437,6 +439,7 @@ protected static native WritableMap modelInfo(
   );
   protected static native long initContext(
     String model,
+    String chat_template,
     boolean embedding,
     int embd_normalize,
     int n_ctx,

android/src/main/jni.cpp

@@ -222,6 +222,7 @@ Java_com_rnllama_LlamaContext_initContext(
     JNIEnv *env,
     jobject thiz,
     jstring model_path_str,
+    jstring chat_template,
     jboolean embedding,
     jint embd_normalize,
     jint n_ctx,
@@ -255,6 +256,9 @@ Java_com_rnllama_LlamaContext_initContext(
     const char *model_path_chars = env->GetStringUTFChars(model_path_str, nullptr);
     defaultParams.model = model_path_chars;
 
+    const char *chat_template_chars = env->GetStringUTFChars(chat_template, nullptr);
+    defaultParams.chat_template = chat_template_chars;
+
     defaultParams.n_ctx = n_ctx;
     defaultParams.n_batch = n_batch;
     defaultParams.n_ubatch = n_ubatch;
@@ -321,6 +325,7 @@ Java_com_rnllama_LlamaContext_initContext(
     bool is_model_loaded = llama->loadModel(defaultParams);
 
     env->ReleaseStringUTFChars(model_path_str, model_path_chars);
+    env->ReleaseStringUTFChars(chat_template, chat_template_chars);
     env->ReleaseStringUTFChars(cache_type_k, cache_type_k_chars);
     env->ReleaseStringUTFChars(cache_type_v, cache_type_v_chars);
 

cpp/rn-llama.cpp

@@ -194,7 +194,7 @@ bool llama_rn_context::loadModel(common_params &params_)
         LOG_ERROR("unable to load model: %s", params_.model.c_str());
         return false;
     }
-    templates = common_chat_templates_from_model(model, "");
+    templates = common_chat_templates_from_model(model, params.chat_template);
     n_ctx = llama_n_ctx(ctx);
 
     // We can uncomment for debugging or after this fix: https://github.com/ggerganov/llama.cpp/pull/11101

ios/RNLlamaContext.mm

@@ -57,6 +57,12 @@ + (instancetype)initWithParams:(NSDictionary *)params onProgress:(void (^)(unsig
     if (isAsset) path = [[NSBundle mainBundle] pathForResource:modelPath ofType:nil];
     defaultParams.model = [path UTF8String];
 
+    NSString *chatTemplate = params[@"chat_template"];
+    if (chatTemplate) {
+        defaultParams.chat_template = [chatTemplate UTF8String];
+        NSLog(@"chatTemplate: %@", chatTemplate);
+    }
+
     if (params[@"n_ctx"]) defaultParams.n_ctx = [params[@"n_ctx"] intValue];
     if (params[@"use_mlock"]) defaultParams.use_mlock = [params[@"use_mlock"]boolValue];
 

src/NativeRNLlama.ts

@@ -7,6 +7,11 @@ export type NativeEmbeddingParams = {
 
 export type NativeContextParams = {
   model: string
+  /**
+   * Chat template to override the default one from the model.
+   */
+  chat_template?: string
+
   is_model_asset?: boolean
   use_progress_callback?: boolean
 

src/index.ts

@@ -103,7 +103,8 @@ export type CompletionResponseFormat = {
 export type CompletionBaseParams = {
   prompt?: string
   messages?: RNLlamaOAICompatibleMessage[]
-  chatTemplate?: string
+  chatTemplate?: string // deprecated
+  chat_template?: string
   jinja?: boolean
   tools?: object
   parallel_tool_calls?: object
@@ -232,7 +233,7 @@ export class LlamaContext {
       // messages always win
       const formattedResult = await this.getFormattedChat(
         params.messages,
-        params.chatTemplate,
+        params.chat_template || params.chatTemplate,
         {
           jinja: params.jinja,
           tools: params.tools,