node-llama-cpp - npm Package Compare versions

Comparing version 1.3.0 to 1.3.1

dist/llamaEvaluator/LlamaModel.d.ts

@@ -10,3 +10,3 @@ export declare class LlamaModel {
      * @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
@@ -20,14 +20,26 @@ * @param {boolean} [options.f16Kv] - use fp16 for KV cache
      */
-    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;
@@ -34,0 +46,0 @@ });

dist/llamaEvaluator/LlamaModel.js

@@ -13,3 +13,3 @@ import { llamaCppNode, LLAMAModel } from "./LlamaBins.js";
      * @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
@@ -23,3 +23,3 @@ * @param {boolean} [options.f16Kv] - use fp16 for KV cache
      */
-    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({
@@ -29,3 +29,3 @@ seed: seed != null ? Math.max(-1, seed) : undefined,
             batchSize,
-            gpuCores,
+            gpuLayers,
             lowVram,
@@ -32,0 +32,0 @@ f16Kv,

package.json

 {
   "name": "node-llama-cpp",
-  "version": "1.3.0",
+  "version": "1.3.1",
   "description": "node.js bindings for llama.cpp",
@@ -5,0 +5,0 @@ "main": "dist/index.js",

README.md

@@ -28,3 +28,3 @@ # Node Llama.cpp
 import path from "path";
-import {LlamaModel, LlamaChatSession} from "node-llama-cpp";
+import {LlamaModel, LlamaContext, LlamaChatSession} from "node-llama-cpp";
 
@@ -58,3 +58,3 @@ const __dirname = path.dirname(fileURLToPath(import.meta.url));
 import path from "path";
-import {LlamaModel, LlamaChatSession, ChatPromptWrapper} from "node-llama-cpp";
+import {LlamaModel, LlamaContext, LlamaChatSession, ChatPromptWrapper} from "node-llama-cpp";
 
@@ -103,3 +103,3 @@ const __dirname = path.dirname(fileURLToPath(import.meta.url));
 import path from "path";
-import {LlamaModel, LlamaChatSession} from "node-llama-cpp";
+import {LlamaModel, LlamaContext, LlamaChatSession} from "node-llama-cpp";
 
@@ -106,0 +106,0 @@ const __dirname = path.dirname(fileURLToPath(import.meta.url));

No alert changes

Improved metrics

Total package byte prevSize

25301898

0

Lines of code

1149

1.06%

No dependency changes