cui-llama.rn 1.3.6 → 1.4.0

package/lib/typescript/index.d.ts

@@ -8,43 +8,19 @@ export type TokenData = {
     token: string;
     completion_probabilities?: Array<NativeCompletionTokenProb>;
 };
-export declare enum GGML_TYPE {
-    LM_GGML_TYPE_F32 = 0,
-    LM_GGML_TYPE_F16 = 1,
-    LM_GGML_TYPE_Q4_0 = 2,
-    LM_GGML_TYPE_Q4_1 = 3,
-    LM_GGML_TYPE_Q5_0 = 6,
-    LM_GGML_TYPE_Q5_1 = 7,
-    LM_GGML_TYPE_Q8_0 = 8,
-    LM_GGML_TYPE_Q8_1 = 9,
-    LM_GGML_TYPE_Q2_K = 10,
-    LM_GGML_TYPE_Q3_K = 11,
-    LM_GGML_TYPE_Q4_K = 12,
-    LM_GGML_TYPE_Q5_K = 13,
-    LM_GGML_TYPE_Q6_K = 14,
-    LM_GGML_TYPE_Q8_K = 15,
-    LM_GGML_TYPE_IQ2_XXS = 16,
-    LM_GGML_TYPE_IQ2_XS = 17,
-    LM_GGML_TYPE_IQ3_XXS = 18,
-    LM_GGML_TYPE_IQ1_S = 19,
-    LM_GGML_TYPE_IQ4_NL = 20,
-    LM_GGML_TYPE_IQ3_S = 21,
-    LM_GGML_TYPE_IQ2_S = 22,
-    LM_GGML_TYPE_IQ4_XS = 23,
-    LM_GGML_TYPE_I8 = 24,
-    LM_GGML_TYPE_I16 = 25,
-    LM_GGML_TYPE_I32 = 26,
-    LM_GGML_TYPE_I64 = 27,
-    LM_GGML_TYPE_F64 = 28,
-    LM_GGML_TYPE_IQ1_M = 29,
-    LM_GGML_TYPE_BF16 = 30,
-    LM_GGML_TYPE_TQ1_0 = 34,
-    LM_GGML_TYPE_TQ2_0 = 35,
-    LM_GGML_TYPE_COUNT = 39
+export declare enum CACHE_TYPE {
+    F16 = "f16",
+    F32 = "f32",
+    Q8_0 = "q8_0",
+    Q4_0 = "q4_0",
+    Q4_1 = "q4_1",
+    IQ4_NL = "iq4_nl",
+    Q5_0 = "q5_0",
+    Q5_1 = "q5_1"
 }
 export type ContextParams = Omit<NativeContextParams, 'cache_type_k' | 'cache_type_v' | 'pooling_type'> & {
-    cache_type_k?: GGML_TYPE;
-    cache_type_v?: GGML_TYPE;
+    cache_type_k?: CACHE_TYPE;
+    cache_type_v?: CACHE_TYPE;
     pooling_type?: 'none' | 'mean' | 'cls' | 'last' | 'rank';
 };
 export type EmbeddingParams = NativeEmbeddingParams;
@@ -88,11 +64,20 @@ export declare class LlamaContext {
     detokenize(tokens: number[]): Promise<string>;
     embedding(text: string, params?: EmbeddingParams): Promise<NativeEmbeddingResult>;
     bench(pp: number, tg: number, pl: number, nr: number): Promise<BenchResult>;
+    applyLoraAdapters(loraList: Array<{
+        path: string;
+        scaled?: number;
+    }>): Promise<void>;
+    removeLoraAdapters(): Promise<void>;
+    getLoadedLoraAdapters(): Promise<Array<{
+        path: string;
+        scaled?: number;
+    }>>;
     release(): Promise<void>;
 }
 export declare function getCpuFeatures(): Promise<NativeCPUFeatures>;
 export declare function setContextLimit(limit: number): Promise<void>;
 export declare function loadLlamaModelInfo(model: string): Promise<Object>;
-export declare function initLlama({ model, is_model_asset: isModelAsset, pooling_type: poolingType, lora, ...rest }: ContextParams, onProgress?: (progress: number) => void): Promise<LlamaContext>;
+export declare function initLlama({ model, is_model_asset: isModelAsset, pooling_type: poolingType, lora, lora_list: loraList, ...rest }: ContextParams, onProgress?: (progress: number) => void): Promise<LlamaContext>;
 export declare function releaseAllLlama(): Promise<void>;
 //# sourceMappingURL=index.d.ts.map

package/lib/typescript/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EACV,mBAAmB,EACnB,kBAAkB,EAClB,sBAAsB,EACtB,yBAAyB,EACzB,sBAAsB,EACtB,oBAAoB,EACpB,qBAAqB,EACrB,uBAAuB,EACvB,iBAAiB,EACjB,qBAAqB,EACrB,6BAA6B,EAC7B,6BAA6B,EAC9B,MAAM,iBAAiB,CAAA;AACxB,OAAO,KAAK,EAAE,+BAA+B,EAAE,iCAAiC,EAAE,MAAM,WAAW,CAAA;AACnG,OAAO,EAAE,sBAAsB,EAAE,0BAA0B,EAAE,MAAM,WAAW,CAAA;AAC9E,OAAO,KAAK,EAAE,kBAAkB,EAAE,2BAA2B,EAAE,MAAM,QAAQ,CAAA;AAG7E,YAAY,EACV,mBAAmB,EACnB,kBAAkB,EAClB,sBAAsB,EACtB,yBAAyB,EACzB,sBAAsB,EACtB,oBAAoB,EACpB,qBAAqB,EACrB,uBAAuB,EACvB,qBAAqB,EACrB,6BAA6B,EAC7B,6BAA6B,EAC7B,kBAAkB,EAClB,2BAA2B,EAC3B,+BAA+B,EAC/B,iCAAiC,GAClC,CAAA;AAED,OAAO,EAAE,sBAAsB,EAAE,0BAA0B,EAAE,CAAA;AAc7D,MAAM,MAAM,SAAS,GAAG;IACtB,KAAK,EAAE,MAAM,CAAA;IACb,wBAAwB,CAAC,EAAE,KAAK,CAAC,yBAAyB,CAAC,CAAA;CAC5D,CAAA;AAOD,oBAAY,SAAS;IACnB,gBAAgB,IAAQ;IACxB,gBAAgB,IAAQ;IACxB,iBAAiB,IAAO;IACxB,iBAAiB,IAAO;IAGxB,iBAAiB,IAAO;IACxB,iBAAiB,IAAO;IACxB,iBAAiB,IAAO;IACxB,iBAAiB,IAAO;IACxB,iBAAiB,KAAQ;IACzB,iBAAiB,KAAQ;IACzB,iBAAiB,KAAQ;IACzB,iBAAiB,KAAQ;IACzB,iBAAiB,KAAQ;IACzB,iBAAiB,KAAQ;IACzB,oBAAoB,KAAK;IACzB,mBAAmB,KAAM;IACzB,oBAAoB,KAAK;IACzB,kBAAkB,KAAO;IACzB,mBAAmB,KAAM;IACzB,kBAAkB,KAAO;IACzB,kBAAkB,KAAO;IACzB,mBAAmB,KAAM;IACzB,eAAe,KAAU;IACzB,gBAAgB,KAAS;IACzB,gBAAgB,KAAS;IACzB,gBAAgB,KAAS;IACzB,gBAAgB,KAAS;IACzB,kBAAkB,KAAO;IACzB,iBAAiB,KAAQ;IAIzB,kBAAkB,KAAO;IACzB,kBAAkB,KAAO;IAIzB,kBAAkB,KAAO;CAC1B;AAGD,MAAM,MAAM,aAAa,GAAG,IAAI,CAC9B,mBAAmB,EACnB,cAAc,GAAG,cAAc,GAAI,cAAc,CAClD,GAAG;IACF,YAAY,CAAC,EAAE,SAAS,CAAA;IACxB,YAAY,CAAC,EAAE,SAAS,CAAA;IACxB,YAAY,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,MAAM,CAAA;CACzD,CAAA;AAED,MAAM,MAAM,eAAe,GAAG,qBAAqB,CAAA;AAEnD,MAAM,MAAM,gBAAgB,GAAG,IAAI,CACjC,sBAAsB,EACtB,yBAAyB,GAAG,QAAQ,CACrC,GAAG;IACF,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,2BAA2B,EAAE,CAAA;IACxC,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB,CAAA;AAED,MAAM,MAAM,WAAW,GAAG;IACxB,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE,MAAM,CAAA;IACpB,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;CACd,CAAA;AAED,qBAAa,YAAY;IACvB,EAAE,EAAE,MAAM,CAAA;IAEV,GAAG,EAAE,OAAO,CAAQ;IAEpB,WAAW,EAAE,MAAM,CAAK;IAExB,KAAK,EAAE;QACL,uBAAuB,CAAC,EAAE,OAAO,CAAA;KAClC,CAAK;gBAEM,EAAE,SAAS,EAAE,GAAG,EAAE,WAAW,EAAE,KAAK,EAAE,EAAE,kBAAkB;IAOtE;;OAEG;IACG,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,uBAAuB,CAAC;IAMrE;;OAEG;IACG,WAAW,CACf,QAAQ,EAAE,MAAM,EAChB,OAAO,CAAC,EAAE;QAAE,SAAS,EAAE,MAAM,CAAA;KAAE,GAC9B,OAAO,CAAC,MAAM,CAAC;IAIZ,gBAAgB,CACpB,QAAQ,EAAE,2BAA2B,EAAE,EACvC,QAAQ,CAAC,EAAE,MAAM,GAChB,OAAO,CAAC,MAAM,CAAC;IAOZ,UAAU,CACd,MAAM,EAAE,gBAAgB,EACxB,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,SAAS,KAAK,IAAI,GACnC,OAAO,CAAC,sBAAsB,CAAC;IAkClC,cAAc,IAAI,OAAO,CAAC,IAAI,CAAC;IAI/B,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,oBAAoB,CAAC;IAI1D,YAAY,CAAC,IAAI,EAAE,MAAM,GAAG,oBAAoB;IAIhD,UAAU,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC;IAI7C,SAAS,CACP,IAAI,EAAE,MAAM,EACZ,MAAM,CAAC,EAAE,eAAe,GACvB,OAAO,CAAC,qBAAqB,CAAC;IAI3B,KAAK,CACT,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,GACT,OAAO,CAAC,WAAW,CAAC;IAejB,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;CAG/B;AAED,wBAAsB,cAAc,IAAK,OAAO,CAAC,iBAAiB,CAAC,CAElE;AAED,wBAAsB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAElE;AAYD,wBAAsB,kBAAkB,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAIvE;AAWD,wBAAsB,SAAS,CAC7B,EACE,KAAK,EACL,cAAc,EAAE,YAAY,EAC5B,YAAY,EAAE,WAAW,EACzB,IAAI,EACJ,GAAG,IAAI,EACR,EAAE,aAAa,EAChB,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,GACtC,OAAO,CAAC,YAAY,CAAC,CAuCvB;AAED,wBAAsB,eAAe,IAAI,OAAO,CAAC,IAAI,CAAC,CAErD"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/index.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EACV,mBAAmB,EACnB,kBAAkB,EAClB,sBAAsB,EACtB,yBAAyB,EACzB,sBAAsB,EACtB,oBAAoB,EACpB,qBAAqB,EACrB,uBAAuB,EACvB,iBAAiB,EACjB,qBAAqB,EACrB,6BAA6B,EAC7B,6BAA6B,EAC9B,MAAM,iBAAiB,CAAA;AACxB,OAAO,KAAK,EACV,+BAA+B,EAC/B,iCAAiC,EAClC,MAAM,WAAW,CAAA;AAClB,OAAO,EAAE,sBAAsB,EAAE,0BAA0B,EAAE,MAAM,WAAW,CAAA;AAC9E,OAAO,KAAK,EAAE,kBAAkB,EAAE,2BAA2B,EAAE,MAAM,QAAQ,CAAA;AAG7E,YAAY,EACV,mBAAmB,EACnB,kBAAkB,EAClB,sBAAsB,EACtB,yBAAyB,EACzB,sBAAsB,EACtB,oBAAoB,EACpB,qBAAqB,EACrB,uBAAuB,EACvB,qBAAqB,EACrB,6BAA6B,EAC7B,6BAA6B,EAC7B,kBAAkB,EAClB,2BAA2B,EAC3B,+BAA+B,EAC/B,iCAAiC,GAClC,CAAA;AAED,OAAO,EAAE,sBAAsB,EAAE,0BAA0B,EAAE,CAAA;AAc7D,MAAM,MAAM,SAAS,GAAG;IACtB,KAAK,EAAE,MAAM,CAAA;IACb,wBAAwB,CAAC,EAAE,KAAK,CAAC,yBAAyB,CAAC,CAAA;CAC5D,CAAA;AAOD,oBAAY,UAAU;IACpB,GAAG,QAAQ;IACX,GAAG,QAAQ;IACX,IAAI,SAAS;IACb,IAAI,SAAS;IACb,IAAI,SAAS;IACb,MAAM,WAAW;IACjB,IAAI,SAAS;IACb,IAAI,SAAS;CACd;AAGD,MAAM,MAAM,aAAa,GAAG,IAAI,CAC9B,mBAAmB,EACnB,cAAc,GAAG,cAAc,GAAG,cAAc,CACjD,GAAG;IACF,YAAY,CAAC,EAAE,UAAU,CAAA;IACzB,YAAY,CAAC,EAAE,UAAU,CAAA;IACzB,YAAY,CAAC,EAAE,MAAM,GAAG,MAAM,GAAG,KAAK,GAAG,MAAM,GAAG,MAAM,CAAA;CACzD,CAAA;AAED,MAAM,MAAM,eAAe,GAAG,qBAAqB,CAAA;AAEnD,MAAM,MAAM,gBAAgB,GAAG,IAAI,CACjC,sBAAsB,EACtB,yBAAyB,GAAG,QAAQ,CACrC,GAAG;IACF,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,QAAQ,CAAC,EAAE,2BAA2B,EAAE,CAAA;IACxC,YAAY,CAAC,EAAE,MAAM,CAAA;CACtB,CAAA;AAED,MAAM,MAAM,WAAW,GAAG;IACxB,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE,MAAM,CAAA;IACpB,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;IACb,KAAK,EAAE,MAAM,CAAA;CACd,CAAA;AAED,qBAAa,YAAY;IACvB,EAAE,EAAE,MAAM,CAAA;IAEV,GAAG,EAAE,OAAO,CAAQ;IAEpB,WAAW,EAAE,MAAM,CAAK;IAExB,KAAK,EAAE;QACL,uBAAuB,CAAC,EAAE,OAAO,CAAA;KAClC,CAAK;gBAEM,EAAE,SAAS,EAAE,GAAG,EAAE,WAAW,EAAE,KAAK,EAAE,EAAE,kBAAkB;IAOtE;;OAEG;IACG,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,uBAAuB,CAAC;IAMrE;;OAEG;IACG,WAAW,CACf,QAAQ,EAAE,MAAM,EAChB,OAAO,CAAC,EAAE;QAAE,SAAS,EAAE,MAAM,CAAA;KAAE,GAC9B,OAAO,CAAC,MAAM,CAAC;IAIZ,gBAAgB,CACpB,QAAQ,EAAE,2BAA2B,EAAE,EACvC,QAAQ,CAAC,EAAE,MAAM,GAChB,OAAO,CAAC,MAAM,CAAC;IAOZ,UAAU,CACd,MAAM,EAAE,gBAAgB,EACxB,QAAQ,CAAC,EAAE,CAAC,IAAI,EAAE,SAAS,KAAK,IAAI,GACnC,OAAO,CAAC,sBAAsB,CAAC;IAqClC,cAAc,IAAI,OAAO,CAAC,IAAI,CAAC;IAI/B,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,oBAAoB,CAAC;IAI1D,YAAY,CAAC,IAAI,EAAE,MAAM,GAAG,oBAAoB;IAIhD,UAAU,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC;IAI7C,SAAS,CACP,IAAI,EAAE,MAAM,EACZ,MAAM,CAAC,EAAE,eAAe,GACvB,OAAO,CAAC,qBAAqB,CAAC;IAI3B,KAAK,CACT,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,EACV,EAAE,EAAE,MAAM,GACT,OAAO,CAAC,WAAW,CAAC;IAejB,iBAAiB,CACrB,QAAQ,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,GACjD,OAAO,CAAC,IAAI,CAAC;IAUV,kBAAkB,IAAI,OAAO,CAAC,IAAI,CAAC;IAInC,qBAAqB,IAAI,OAAO,CACpC,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAC,CACzC;IAIK,OAAO,IAAI,OAAO,CAAC,IAAI,CAAC;CAG/B;AAED,wBAAsB,cAAc,IAAK,OAAO,CAAC,iBAAiB,CAAC,CAElE;AAED,wBAAsB,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAElE;AAYD,wBAAsB,kBAAkB,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAIvE;AAWD,wBAAsB,SAAS,CAC7B,EACE,KAAK,EACL,cAAc,EAAE,YAAY,EAC5B,YAAY,EAAE,WAAW,EACzB,IAAI,EACJ,SAAS,EAAE,QAAQ,EACnB,GAAG,IAAI,EACR,EAAE,aAAa,EAChB,UAAU,CAAC,EAAE,CAAC,QAAQ,EAAE,MAAM,KAAK,IAAI,GACtC,OAAO,CAAC,YAAY,CAAC,CA+CvB;AAED,wBAAsB,eAAe,IAAI,OAAO,CAAC,IAAI,CAAC,CAErD"}

package/llama-rn.podspec

@@ -2,8 +2,7 @@ require "json"
 
 package = JSON.parse(File.read(File.join(__dir__, "package.json")))
 base_ld_flags = "-framework Accelerate -framework Foundation -framework Metal -framework MetalKit"
-base_compiler_flags = "-fno-objc-arc -DLM_GGML_USE_ACCELERATE -Wno-shorten-64-to-32"
-folly_compiler_flags = "-DFOLLY_NO_CONFIG -DFOLLY_MOBILE=1 -DFOLLY_USE_LIBCPP=1 -Wno-comma"
+base_compiler_flags = "-fno-objc-arc -DLM_GGML_USE_CPU -DLM_GGML_USE_ACCELERATE -Wno-shorten-64-to-32"
 
 if ENV["RNLLAMA_DISABLE_METAL"] != "1" then
   base_compiler_flags += " -DLM_GGML_USE_METAL" # -DLM_GGML_METAL_NDEBUG
@@ -21,7 +20,7 @@ Pod::Spec.new do |s|
   s.license      = package["license"]
   s.authors      = package["author"]
 
-  s.platforms    = { :ios => "11.0", :tvos => "11.0" }
+  s.platforms    = { :ios => "13.0", :tvos => "13.0" }
   s.source       = { :git => "https://github.com/mybigday/llama.rn.git", :tag => "#{s.version}" }
 
   s.source_files = "ios/**/*.{h,m,mm}", "cpp/**/*.{h,cpp,hpp,c,m,mm}"
@@ -33,24 +32,11 @@ Pod::Spec.new do |s|
   s.pod_target_xcconfig = {
     "OTHER_LDFLAGS" => base_ld_flags,
     "OTHER_CFLAGS" => base_optimizer_flags,
-    "OTHER_CPLUSPLUSFLAGS" => base_optimizer_flags
+    "OTHER_CPLUSPLUSFLAGS" => base_optimizer_flags + " -std=c++17"
   }
 
   # Don't install the dependencies when we run `pod install` in the old architecture.
   if ENV['RCT_NEW_ARCH_ENABLED'] == '1' then
-    s.compiler_flags = base_compiler_flags + " " + folly_compiler_flags + " -DRCT_NEW_ARCH_ENABLED=1"
-    new_arch_cpp_flags = "-DFOLLY_NO_CONFIG -DFOLLY_MOBILE=1 -DFOLLY_USE_LIBCPP=1"
-    s.pod_target_xcconfig = {
-      "CLANG_CXX_LANGUAGE_STANDARD" => "c++17",
-      "HEADER_SEARCH_PATHS" => "\"$(PODS_ROOT)/boost\"",
-      "OTHER_LDFLAGS" => "-framework Accelerate",
-      "OTHER_CFLAGS" => base_optimizer_flags,
-      "OTHER_CPLUSPLUSFLAGS" => new_arch_cpp_flags + " " + base_optimizer_flags
-    }
-    s.dependency "React-Codegen"
-    s.dependency "RCT-Folly"
-    s.dependency "RCTRequired"
-    s.dependency "RCTTypeSafety"
-    s.dependency "ReactCommon/turbomodule/core"
+    install_modules_dependencies(s)
   end
 end

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cui-llama.rn",
-  "version": "1.3.6",
+  "version": "1.4.0",
   "description": "Fork of llama.rn for ChatterUI",
   "main": "lib/commonjs/index",
   "module": "lib/module/index",
@@ -13,8 +13,7 @@
     "jest",
     "ios",
     "android",
-    "cpp/*.*",
-    "cpp/amx/*.*",
+    "cpp",
     "*.podspec",
     "!lib/typescript/example",
     "!ios/build",

package/src/NativeRNLlama.ts

@@ -12,6 +12,7 @@ export type NativeContextParams = {
 
   n_ctx?: number
   n_batch?: number
+  n_ubatch?: number
 
   n_threads?: number
   n_gpu_layers?: number
@@ -24,18 +25,28 @@ export type NativeContextParams = {
   /**
    * KV cache data type for the K (Experimental in llama.cpp)
    */
-  cache_type_k?: number
+  cache_type_k?: string
   /**
    * KV cache data type for the V (Experimental in llama.cpp)
    */
-  cache_type_v?: number
+  cache_type_v?: string
 
   use_mlock?: boolean
   use_mmap?: boolean
   vocab_only?: boolean
 
-  lora?: string // lora_adaptor
+  /**
+   * Single LoRA adapter path
+   */
+  lora?: string
+  /**
+   * Single LoRA adapter scale
+   */
   lora_scaled?: number
+  /**
+   * LoRA adapter list
+   */
+  lora_list?: Array<{ path: string; scaled?: number }>
 
   rope_freq_base?: number
   rope_freq_scale?: number
@@ -115,10 +126,6 @@ export type NativeCompletionParams = {
    * Repeat alpha presence penalty. Default: `0.0`, which is disabled.
    */
   penalty_present?: number
-  /**
-   * Penalize newline tokens when applying the repeat penalty. Default: `false`
-   */
-  // penalize_nl?: boolean
   /**
    * Enable Mirostat sampling, controlling perplexity during text generation. Default: `0`, where `0` is disabled, `1` is Mirostat, and `2` is Mirostat 2.0.
    */
@@ -243,8 +250,16 @@ export interface Spec extends TurboModule {
   setContextLimit(limit: number): Promise<void>
 
   modelInfo(path: string, skip?: string[]): Promise<Object>
-  initContext(contextId: number, params: NativeContextParams): Promise<NativeLlamaContext>
+  initContext(
+    contextId: number,
+    params: NativeContextParams,
+  ): Promise<NativeLlamaContext>
 
+  getFormattedChat(
+    contextId: number,
+    messages: NativeLlamaChatMessage[],
+    chatTemplate?: string,
+  ): Promise<string>
   loadSession(
     contextId: number,
     filepath: string,
@@ -262,11 +277,6 @@ export interface Spec extends TurboModule {
   tokenizeAsync(contextId: number, text: string): Promise<NativeTokenizeResult>
   tokenizeSync(contextId: number, text: string): NativeTokenizeResult
   getCpuFeatures() : Promise<NativeCPUFeatures>
-  getFormattedChat(
-    contextId: number,
-    messages: NativeLlamaChatMessage[],
-    chatTemplate?: string,
-  ): Promise<string>
   detokenize(contextId: number, tokens: number[]): Promise<string>
   embedding(
     contextId: number,
@@ -281,6 +291,15 @@ export interface Spec extends TurboModule {
     nr: number,
   ): Promise<string>
 
+  applyLoraAdapters(
+    contextId: number,
+    loraAdapters: Array<{ path: string; scaled?: number }>,
+  ): Promise<void>
+  removeLoraAdapters(contextId: number): Promise<void>
+  getLoadedLoraAdapters(
+    contextId: number,
+  ): Promise<Array<{ path: string; scaled?: number }>>
+
   releaseContext(contextId: number): Promise<void>
 
   releaseAllContexts(): Promise<void>

package/src/index.ts

@@ -15,7 +15,10 @@ import type {
   NativeCompletionTokenProbItem,
   NativeCompletionResultTimings,
 } from './NativeRNLlama'
-import type { SchemaGrammarConverterPropOrder, SchemaGrammarConverterBuiltinRule } from './grammar'
+import type {
+  SchemaGrammarConverterPropOrder,
+  SchemaGrammarConverterBuiltinRule,
+} from './grammar'
 import { SchemaGrammarConverter, convertJsonSchemaToGrammar } from './grammar'
 import type { RNLlamaMessagePart, RNLlamaOAICompatibleMessage } from './chat'
 import { formatChat } from './chat'
@@ -62,56 +65,24 @@ type TokenNativeEvent = {
   tokenResult: TokenData
 }
 
-export enum GGML_TYPE {
-  LM_GGML_TYPE_F32     = 0,
-  LM_GGML_TYPE_F16     = 1,
-  LM_GGML_TYPE_Q4_0    = 2,
-  LM_GGML_TYPE_Q4_1    = 3,
-  // LM_GGML_TYPE_Q4_2 = 4, support has been removed
-  // LM_GGML_TYPE_Q4_3 = 5, support has been removed
-  LM_GGML_TYPE_Q5_0    = 6,
-  LM_GGML_TYPE_Q5_1    = 7,
-  LM_GGML_TYPE_Q8_0    = 8,
-  LM_GGML_TYPE_Q8_1    = 9,
-  LM_GGML_TYPE_Q2_K    = 10,
-  LM_GGML_TYPE_Q3_K    = 11,
-  LM_GGML_TYPE_Q4_K    = 12,
-  LM_GGML_TYPE_Q5_K    = 13,
-  LM_GGML_TYPE_Q6_K    = 14,
-  LM_GGML_TYPE_Q8_K    = 15,
-  LM_GGML_TYPE_IQ2_XXS = 16,
-  LM_GGML_TYPE_IQ2_XS  = 17,
-  LM_GGML_TYPE_IQ3_XXS = 18,
-  LM_GGML_TYPE_IQ1_S   = 19,
-  LM_GGML_TYPE_IQ4_NL  = 20,
-  LM_GGML_TYPE_IQ3_S   = 21,
-  LM_GGML_TYPE_IQ2_S   = 22,
-  LM_GGML_TYPE_IQ4_XS  = 23,
-  LM_GGML_TYPE_I8      = 24,
-  LM_GGML_TYPE_I16     = 25,
-  LM_GGML_TYPE_I32     = 26,
-  LM_GGML_TYPE_I64     = 27,
-  LM_GGML_TYPE_F64     = 28,
-  LM_GGML_TYPE_IQ1_M   = 29,
-  LM_GGML_TYPE_BF16    = 30,
-  // LM_GGML_TYPE_Q4_0_4_4 = 31, support has been removed from gguf files
-  // LM_GGML_TYPE_Q4_0_4_8 = 32,
-  // LM_GGML_TYPE_Q4_0_8_8 = 33,
-  LM_GGML_TYPE_TQ1_0   = 34,
-  LM_GGML_TYPE_TQ2_0   = 35,
-  // LM_GGML_TYPE_IQ4_NL_4_4 = 36,
-  // LM_GGML_TYPE_IQ4_NL_4_8 = 37,
-  // LM_GGML_TYPE_IQ4_NL_8_8 = 38,
-  LM_GGML_TYPE_COUNT   = 39,
-};
+export enum CACHE_TYPE  {
+  F16 = 'f16',
+  F32 = 'f32',
+  Q8_0 = 'q8_0',
+  Q4_0 = 'q4_0',
+  Q4_1 = 'q4_1',
+  IQ4_NL = 'iq4_nl',
+  Q5_0 = 'q5_0',
+  Q5_1 = 'q5_1'
+}
 
 
 export type ContextParams = Omit<
   NativeContextParams,
-  'cache_type_k' | 'cache_type_v' |  'pooling_type'
+  'cache_type_k' | 'cache_type_v' | 'pooling_type'
 > & {
-  cache_type_k?: GGML_TYPE
-  cache_type_v?: GGML_TYPE
+  cache_type_k?: CACHE_TYPE
+  cache_type_v?: CACHE_TYPE
   pooling_type?: 'none' | 'mean' | 'cls' | 'last' | 'rank'
 }
 
@@ -190,7 +161,10 @@ export class LlamaContext {
     let finalPrompt = params.prompt
     if (params.messages) {
       // messages always win
-      finalPrompt = await this.getFormattedChat(params.messages, params.chatTemplate)
+      finalPrompt = await this.getFormattedChat(
+        params.messages,
+        params.chatTemplate,
+      )
     }
 
     let tokenListener: any =
@@ -263,6 +237,28 @@ export class LlamaContext {
     }
   }
 
+  async applyLoraAdapters(
+    loraList: Array<{ path: string; scaled?: number }>
+  ): Promise<void> {
+    let loraAdapters: Array<{ path: string; scaled?: number }> = []
+    if (loraList)
+      loraAdapters = loraList.map((l) => ({
+        path: l.path.replace(/file:\/\//, ''),
+        scaled: l.scaled,
+      }))
+    return RNLlama.applyLoraAdapters(this.id, loraAdapters)
+  }
+
+  async removeLoraAdapters(): Promise<void> {
+    return RNLlama.removeLoraAdapters(this.id)
+  }
+
+  async getLoadedLoraAdapters(): Promise<
+    Array<{ path: string; scaled?: number }>
+  > {
+    return RNLlama.getLoadedLoraAdapters(this.id)
+  }
+
   async release(): Promise<void> {
     return RNLlama.releaseContext(this.id)
   }
@@ -307,6 +303,7 @@ export async function initLlama(
     is_model_asset: isModelAsset,
     pooling_type: poolingType,
     lora,
+    lora_list: loraList,
     ...rest
   }: ContextParams,
   onProgress?: (progress: number) => void,
@@ -317,6 +314,13 @@ export async function initLlama(
   let loraPath = lora
   if (loraPath?.startsWith('file://')) loraPath = loraPath.slice(7)
 
+  let loraAdapters: Array<{ path: string; scaled?: number }> = []
+  if (loraList)
+    loraAdapters = loraList.map((l) => ({
+      path: l.path.replace(/file:\/\//, ''),
+      scaled: l.scaled,
+    }))
+
   const contextId = contextIdCounter + contextIdRandom()
   contextIdCounter += 1
 
@@ -342,6 +346,7 @@ export async function initLlama(
     use_progress_callback: !!onProgress,
     pooling_type: poolType,
     lora: loraPath,
+    lora_list: loraAdapters,
     ...rest,
   }).catch((err: any) => {
     removeProgressListener?.remove()

package/cpp/llama.cpp.rej

@@ -1,23 +0,0 @@
---- llama.cpp.orig	2024-11-02 12:42:13
-+++ llama.cpp	2024-11-02 13:00:37
-@@ -1941,16 +1952,16 @@
- 
-         if (prefetch > 0) {
-             // advise the kernel to preload the mapped memory
--            if (posix_madvise(addr, std::min(file->size, prefetch), POSIX_MADV_WILLNEED)) {
--                LLAMA_LOG_WARN("warning: posix_madvise(.., POSIX_MADV_WILLNEED) failed: %s\n",
-+            if (madvise(addr, std::min(file->size, prefetch), MADV_WILLNEED)) {
-+                fprintf(stderr, "warning: madvise(.., MADV_WILLNEED) failed: %s\n",
-                         strerror(errno));
-             }
-         }
-         if (numa) {
-             // advise the kernel not to use readahead
-             // (because the next page might not belong on the same node)
--            if (posix_madvise(addr, file->size, POSIX_MADV_RANDOM)) {
--                LLAMA_LOG_WARN("warning: posix_madvise(.., POSIX_MADV_RANDOM) failed: %s\n",
-+            if (madvise(addr, file->size, MADV_RANDOM)) {
-+                fprintf(stderr, "warning: madvise(.., MADV_RANDOM) failed: %s\n",
-                         strerror(errno));
-             }
-         }