node-llama-cpp 1.3.0 → 1.3.1

package/README.md

@@ -25,7 +25,7 @@ To disable this behavior set the environment variable `NODE_LLAMA_CPP_SKIP_DOWNL
 ```typescript
 import {fileURLToPath} from "url";
 import path from "path";
-import {LlamaModel, LlamaChatSession} from "node-llama-cpp";
+import {LlamaModel, LlamaContext, LlamaChatSession} from "node-llama-cpp";
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 
@@ -54,7 +54,7 @@ console.log("AI: " + a2);
 ```typescript
 import {fileURLToPath} from "url";
 import path from "path";
-import {LlamaModel, LlamaChatSession, ChatPromptWrapper} from "node-llama-cpp";
+import {LlamaModel, LlamaContext, LlamaChatSession, ChatPromptWrapper} from "node-llama-cpp";
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 
@@ -98,7 +98,7 @@ console.log("AI: " + a2);
 ```typescript
 import {fileURLToPath} from "url";
 import path from "path";
-import {LlamaModel, LlamaChatSession} from "node-llama-cpp";
+import {LlamaModel, LlamaContext, LlamaChatSession} from "node-llama-cpp";
 
 const __dirname = path.dirname(fileURLToPath(import.meta.url));
 

package/dist/llamaEvaluator/LlamaModel.d.ts

@@ -7,7 +7,7 @@ export declare class LlamaModel {
      * @param {number | null} [options.seed] - If null, a random seed will be used
      * @param {number} [options.contextSize] - text context size
      * @param {number} [options.batchSize] - prompt processing batch size
-     * @param {number} [options.gpuCores] - number of layers to store in VRAM
+     * @param {number} [options.gpuLayers] - number of layers to store in VRAM
      * @param {boolean} [options.lowVram] - if true, reduce VRAM usage at the cost of performance
      * @param {boolean} [options.f16Kv] - use fp16 for KV cache
      * @param {boolean} [options.logitsAll] - the llama_eval() call computes all logits, not just the last one
@@ -16,18 +16,30 @@ export declare class LlamaModel {
      * @param {boolean} [options.useMlock] - force system to keep model in RAM
      * @param {boolean} [options.embedding] - embedding mode only
      */
-    constructor({ modelPath, seed, contextSize, batchSize, gpuCores, lowVram, f16Kv, logitsAll, vocabOnly, useMmap, useMlock, embedding }: {
+    constructor({ modelPath, seed, contextSize, batchSize, gpuLayers, lowVram, f16Kv, logitsAll, vocabOnly, useMmap, useMlock, embedding }: {
+        /** path to the model on the filesystem */
         modelPath: string;
+        /** If null, a random seed will be used */
         seed?: number | null;
+        /** text context size */
         contextSize?: number;
+        /** prompt processing batch size */
         batchSize?: number;
-        gpuCores?: number;
+        /** number of layers to store in VRAM */
+        gpuLayers?: number;
+        /** if true, reduce VRAM usage at the cost of performance */
         lowVram?: boolean;
+        /** use fp16 for KV cache */
         f16Kv?: boolean;
+        /** the llama_eval() call computes all logits, not just the last one */
         logitsAll?: boolean;
+        /** only load the vocabulary, no weights */
         vocabOnly?: boolean;
+        /** use mmap if possible */
         useMmap?: boolean;
+        /** force system to keep model in RAM */
         useMlock?: boolean;
+        /** embedding mode only */
         embedding?: boolean;
     });
     static get systemInfo(): string;

package/dist/llamaEvaluator/LlamaModel.js

@@ -10,7 +10,7 @@ export class LlamaModel {
      * @param {number | null} [options.seed] - If null, a random seed will be used
      * @param {number} [options.contextSize] - text context size
      * @param {number} [options.batchSize] - prompt processing batch size
-     * @param {number} [options.gpuCores] - number of layers to store in VRAM
+     * @param {number} [options.gpuLayers] - number of layers to store in VRAM
      * @param {boolean} [options.lowVram] - if true, reduce VRAM usage at the cost of performance
      * @param {boolean} [options.f16Kv] - use fp16 for KV cache
      * @param {boolean} [options.logitsAll] - the llama_eval() call computes all logits, not just the last one
@@ -19,12 +19,12 @@ export class LlamaModel {
      * @param {boolean} [options.useMlock] - force system to keep model in RAM
      * @param {boolean} [options.embedding] - embedding mode only
      */
-    constructor({ modelPath, seed = null, contextSize = 1024 * 4, batchSize, gpuCores, lowVram, f16Kv, logitsAll, vocabOnly, useMmap, useMlock, embedding }) {
+    constructor({ modelPath, seed = null, contextSize = 1024 * 4, batchSize, gpuLayers, lowVram, f16Kv, logitsAll, vocabOnly, useMmap, useMlock, embedding }) {
         this._model = new LLAMAModel(modelPath, removeNullFields({
             seed: seed != null ? Math.max(-1, seed) : undefined,
             contextSize,
             batchSize,
-            gpuCores,
+            gpuLayers,
             lowVram,
             f16Kv,
             logitsAll,

package/dist/llamaEvaluator/LlamaModel.js.map

@@ -1 +1 @@
-{"version":3,"file":"LlamaModel.js","sourceRoot":"","sources":["../../src/llamaEvaluator/LlamaModel.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,YAAY,EAAE,UAAU,EAAC,MAAM,gBAAgB,CAAC;AAGxD,MAAM,OAAO,UAAU;IACnB,gBAAgB;IACA,MAAM,CAAa;IAEnC;;;;;;;;;;;;;;;;OAgBG;IACH,YAAmB,EACf,SAAS,EAAE,IAAI,GAAG,IAAI,EAAE,WAAW,GAAG,IAAI,GAAG,CAAC,EAAE,SAAS,EAAE,QAAQ,EACnE,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAKrE;QACG,IAAI,CAAC,MAAM,GAAG,IAAI,UAAU,CAAC,SAAS,EAAE,gBAAgB,CAAC;YACrD,IAAI,EAAE,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS;YACnD,WAAW;YACX,SAAS;YACT,QAAQ;YACR,OAAO;YACP,KAAK;YACL,SAAS;YACT,SAAS;YACT,OAAO;YACP,QAAQ;YACR,SAAS;SACZ,CAAC,CAAC,CAAC;IACR,CAAC;IAEM,MAAM,KAAK,UAAU;QACxB,OAAO,YAAY,CAAC,UAAU,EAAE,CAAC;IACrC,CAAC;CACJ;AAED,SAAS,gBAAgB,CAAmB,GAAM;IAC9C,MAAM,MAAM,GAAM,MAAM,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;IAEzC,KAAK,MAAM,GAAG,IAAI,GAAG,EAAE;QACnB,IAAI,MAAM,CAAC,GAAG,CAAC,IAAI,IAAI;YACnB,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC;KAC1B;IAED,OAAO,MAAM,CAAC;AAClB,CAAC"}
+{"version":3,"file":"LlamaModel.js","sourceRoot":"","sources":["../../src/llamaEvaluator/LlamaModel.ts"],"names":[],"mappings":"AAAA,OAAO,EAAC,YAAY,EAAE,UAAU,EAAC,MAAM,gBAAgB,CAAC;AAGxD,MAAM,OAAO,UAAU;IACnB,gBAAgB;IACA,MAAM,CAAa;IAEnC;;;;;;;;;;;;;;;;OAgBG;IACH,YAAmB,EACf,SAAS,EAAE,IAAI,GAAG,IAAI,EAAE,WAAW,GAAG,IAAI,GAAG,CAAC,EAAE,SAAS,EAAE,SAAS,EACpE,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,OAAO,EAAE,QAAQ,EAAE,SAAS,EAqCrE;QACG,IAAI,CAAC,MAAM,GAAG,IAAI,UAAU,CAAC,SAAS,EAAE,gBAAgB,CAAC;YACrD,IAAI,EAAE,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS;YACnD,WAAW;YACX,SAAS;YACT,SAAS;YACT,OAAO;YACP,KAAK;YACL,SAAS;YACT,SAAS;YACT,OAAO;YACP,QAAQ;YACR,SAAS;SACZ,CAAC,CAAC,CAAC;IACR,CAAC;IAEM,MAAM,KAAK,UAAU;QACxB,OAAO,YAAY,CAAC,UAAU,EAAE,CAAC;IACrC,CAAC;CACJ;AAED,SAAS,gBAAgB,CAAmB,GAAM;IAC9C,MAAM,MAAM,GAAM,MAAM,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,CAAC,CAAC;IAEzC,KAAK,MAAM,GAAG,IAAI,GAAG,EAAE;QACnB,IAAI,MAAM,CAAC,GAAG,CAAC,IAAI,IAAI;YACnB,OAAO,MAAM,CAAC,GAAG,CAAC,CAAC;KAC1B;IAED,OAAO,MAAM,CAAC;AAClB,CAAC"}

package/llama/addon.cpp

@@ -34,8 +34,8 @@ class LLAMAModel : public Napi::ObjectWrap<LLAMAModel> {
                 params.n_batch = options.Get("batchSize").As<Napi::Number>().Int32Value();
             }
 
-            if (options.Has("gpuCores")) {
-                params.n_gpu_layers = options.Get("gpuCores").As<Napi::Number>().Int32Value();
+            if (options.Has("gpuLayers")) {
+                params.n_gpu_layers = options.Get("gpuLayers").As<Napi::Number>().Int32Value();
             }
 
             if (options.Has("lowVram")) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "node-llama-cpp",
-  "version": "1.3.0",
+  "version": "1.3.1",
   "description": "node.js bindings for llama.cpp",
   "main": "dist/index.js",
   "type": "module",