langchain 0.0.176 → 0.0.177

package/dist/llms/bedrock.d.ts

@@ -23,6 +23,7 @@ export declare class Bedrock extends LLM implements BaseBedrockInput {
     stopSequences?: string[];
     modelKwargs?: Record<string, unknown>;
     codec: EventStreamCodec;
+    streaming: boolean;
     get lc_secrets(): {
         [key: string]: string;
     } | undefined;
@@ -39,7 +40,7 @@ export declare class Bedrock extends LLM implements BaseBedrockInput {
       Example:
         response = model.call("Tell me a joke.")
     */
-    _call(prompt: string, options: this["ParsedCallOptions"]): Promise<string>;
+    _call(prompt: string, options: this["ParsedCallOptions"], runManager?: CallbackManagerForLLMRun): Promise<string>;
     _signedFetch(prompt: string, options: this["ParsedCallOptions"], fields: {
         bedrockMethod: "invoke" | "invoke-with-response-stream";
         endpointHost: string;

package/dist/llms/bedrock.js

@@ -89,6 +89,12 @@ export class Bedrock extends LLM {
             writable: true,
             value: new EventStreamCodec(toUtf8, fromUtf8)
         });
+        Object.defineProperty(this, "streaming", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: false
+        });
         this.model = fields?.model ?? this.model;
         const allowedModels = ["ai21", "anthropic", "amazon"];
         if (!allowedModels.includes(this.model.split(".")[0])) {
@@ -106,6 +112,7 @@ export class Bedrock extends LLM {
         this.endpointHost = fields?.endpointHost ?? fields?.endpointUrl;
         this.stopSequences = fields?.stopSequences;
         this.modelKwargs = fields?.modelKwargs;
+        this.streaming = fields?.streaming ?? this.streaming;
     }
     /** Call out to Bedrock service model.
       Arguments:
@@ -117,10 +124,23 @@ export class Bedrock extends LLM {
       Example:
         response = model.call("Tell me a joke.")
     */
-    async _call(prompt, options) {
+    async _call(prompt, options, runManager) {
         const service = "bedrock-runtime";
         const endpointHost = this.endpointHost ?? `${service}.${this.region}.amazonaws.com`;
         const provider = this.model.split(".")[0];
+        if (this.streaming) {
+            const stream = this._streamResponseChunks(prompt, options, runManager);
+            let finalResult;
+            for await (const chunk of stream) {
+                if (finalResult === undefined) {
+                    finalResult = chunk;
+                }
+                else {
+                    finalResult = finalResult.concat(chunk);
+                }
+            }
+            return finalResult?.text ?? "";
+        }
         const response = await this._signedFetch(prompt, options, {
             bedrockMethod: "invoke",
             endpointHost,
@@ -201,7 +221,8 @@ export class Bedrock extends LLM {
                         text,
                         generationInfo: {},
                     });
-                    await runManager?.handleLLMNewToken(text);
+                    // eslint-disable-next-line no-void
+                    void runManager?.handleLLMNewToken(text);
                 }
             }
         }
@@ -212,7 +233,8 @@ export class Bedrock extends LLM {
                 text,
                 generationInfo: {},
             });
-            await runManager?.handleLLMNewToken(text);
+            // eslint-disable-next-line no-void
+            void runManager?.handleLLMNewToken(text);
         }
     }
     // eslint-disable-next-line @typescript-eslint/no-explicit-any

package/dist/llms/hf.cjs

@@ -57,6 +57,12 @@ class HuggingFaceInference extends base_js_1.LLM {
             writable: true,
             value: undefined
         });
+        Object.defineProperty(this, "endpointUrl", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: undefined
+        });
         this.model = fields?.model ?? this.model;
         this.temperature = fields?.temperature ?? this.temperature;
         this.maxTokens = fields?.maxTokens ?? this.maxTokens;
@@ -65,6 +71,7 @@ class HuggingFaceInference extends base_js_1.LLM {
         this.frequencyPenalty = fields?.frequencyPenalty ?? this.frequencyPenalty;
         this.apiKey =
             fields?.apiKey ?? (0, env_js_1.getEnvironmentVariable)("HUGGINGFACEHUB_API_KEY");
+        this.endpointUrl = fields?.endpointUrl;
         if (!this.apiKey) {
             throw new Error("Please set an API key for HuggingFace Hub in the environment variable HUGGINGFACEHUB_API_KEY or in the apiKey field of the HuggingFaceInference constructor.");
         }
@@ -75,7 +82,9 @@ class HuggingFaceInference extends base_js_1.LLM {
     /** @ignore */
     async _call(prompt, options) {
         const { HfInference } = await HuggingFaceInference.imports();
-        const hf = new HfInference(this.apiKey);
+        const hf = this.endpointUrl
+            ? new HfInference(this.apiKey).endpoint(this.endpointUrl)
+            : new HfInference(this.apiKey);
         const res = await this.caller.callWithOptions({ signal: options.signal }, hf.textGeneration.bind(hf), {
             model: this.model,
             parameters: {

package/dist/llms/hf.d.ts

@@ -6,6 +6,8 @@ import { LLM, BaseLLMParams } from "./base.js";
 export interface HFInput {
     /** Model to use */
     model: string;
+    /** Custom inference endpoint URL to use */
+    endpointUrl?: string;
     /** Sampling temperature to use */
     temperature?: number;
     /**
@@ -36,6 +38,7 @@ export declare class HuggingFaceInference extends LLM implements HFInput {
     topK: number | undefined;
     frequencyPenalty: number | undefined;
     apiKey: string | undefined;
+    endpointUrl: string | undefined;
     constructor(fields?: Partial<HFInput> & BaseLLMParams);
     _llmType(): string;
     /** @ignore */

package/dist/llms/hf.js

@@ -54,6 +54,12 @@ export class HuggingFaceInference extends LLM {
             writable: true,
             value: undefined
         });
+        Object.defineProperty(this, "endpointUrl", {
+            enumerable: true,
+            configurable: true,
+            writable: true,
+            value: undefined
+        });
         this.model = fields?.model ?? this.model;
         this.temperature = fields?.temperature ?? this.temperature;
         this.maxTokens = fields?.maxTokens ?? this.maxTokens;
@@ -62,6 +68,7 @@ export class HuggingFaceInference extends LLM {
         this.frequencyPenalty = fields?.frequencyPenalty ?? this.frequencyPenalty;
         this.apiKey =
             fields?.apiKey ?? getEnvironmentVariable("HUGGINGFACEHUB_API_KEY");
+        this.endpointUrl = fields?.endpointUrl;
         if (!this.apiKey) {
             throw new Error("Please set an API key for HuggingFace Hub in the environment variable HUGGINGFACEHUB_API_KEY or in the apiKey field of the HuggingFaceInference constructor.");
         }
@@ -72,7 +79,9 @@ export class HuggingFaceInference extends LLM {
     /** @ignore */
     async _call(prompt, options) {
         const { HfInference } = await HuggingFaceInference.imports();
-        const hf = new HfInference(this.apiKey);
+        const hf = this.endpointUrl
+            ? new HfInference(this.apiKey).endpoint(this.endpointUrl)
+            : new HfInference(this.apiKey);
         const res = await this.caller.callWithOptions({ signal: options.signal }, hf.textGeneration.bind(hf), {
             model: this.model,
             parameters: {

package/dist/llms/llama_cpp.cjs

@@ -1,7 +1,7 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.LlamaCpp = void 0;
-const node_llama_cpp_1 = require("node-llama-cpp");
+const llama_cpp_js_1 = require("../util/llama_cpp.cjs");
 const base_js_1 = require("./base.cjs");
 /**
  *  To use this model you need to have the `node-llama-cpp` module installed.
@@ -15,73 +15,31 @@ class LlamaCpp extends base_js_1.LLM {
     }
     constructor(inputs) {
         super(inputs);
-        Object.defineProperty(this, "batchSize", {
+        Object.defineProperty(this, "maxTokens", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "contextSize", {
+        Object.defineProperty(this, "temperature", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "embedding", {
+        Object.defineProperty(this, "topK", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "f16Kv", {
+        Object.defineProperty(this, "topP", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "gpuLayers", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "logitsAll", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "lowVram", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "seed", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "useMlock", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "useMmap", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "vocabOnly", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "modelPath", {
+        Object.defineProperty(this, "trimWhitespaceSuffix", {
             enumerable: true,
             configurable: true,
             writable: true,
@@ -105,29 +63,31 @@ class LlamaCpp extends base_js_1.LLM {
             writable: true,
             value: void 0
         });
-        this.batchSize = inputs.batchSize;
-        this.contextSize = inputs.contextSize;
-        this.embedding = inputs.embedding;
-        this.f16Kv = inputs.f16Kv;
-        this.gpuLayers = inputs.gpuLayers;
-        this.logitsAll = inputs.logitsAll;
-        this.lowVram = inputs.lowVram;
-        this.modelPath = inputs.modelPath;
-        this.seed = inputs.seed;
-        this.useMlock = inputs.useMlock;
-        this.useMmap = inputs.useMmap;
-        this.vocabOnly = inputs.vocabOnly;
-        this._model = new node_llama_cpp_1.LlamaModel(inputs);
-        this._context = new node_llama_cpp_1.LlamaContext({ model: this._model });
-        this._session = new node_llama_cpp_1.LlamaChatSession({ context: this._context });
+        this.maxTokens = inputs?.maxTokens;
+        this.temperature = inputs?.temperature;
+        this.topK = inputs?.topK;
+        this.topP = inputs?.topP;
+        this.trimWhitespaceSuffix = inputs?.trimWhitespaceSuffix;
+        this._model = (0, llama_cpp_js_1.createLlamaModel)(inputs);
+        this._context = (0, llama_cpp_js_1.createLlamaContext)(this._model, inputs);
+        this._session = (0, llama_cpp_js_1.createLlamaSession)(this._context);
     }
     _llmType() {
         return "llama2_cpp";
     }
     /** @ignore */
-    async _call(prompt, options) {
+    async _call(prompt, 
+    // @ts-expect-error - TS6133: 'options' is declared but its value is never read.
+    options) {
         try {
-            const completion = await this._session.prompt(prompt, options);
+            const promptOptions = {
+                maxTokens: this?.maxTokens,
+                temperature: this?.temperature,
+                topK: this?.topK,
+                topP: this?.topP,
+                trimWhitespaceSuffix: this?.trimWhitespaceSuffix,
+            };
+            const completion = await this._session.prompt(prompt, promptOptions);
             return completion;
         }
         catch (e) {

package/dist/llms/llama_cpp.d.ts

@@ -1,40 +1,11 @@
 import { LlamaModel, LlamaContext, LlamaChatSession } from "node-llama-cpp";
+import { LlamaBaseCppInputs } from "../util/llama_cpp.js";
 import { LLM, BaseLLMCallOptions, BaseLLMParams } from "./base.js";
 /**
  * Note that the modelPath is the only required parameter. For testing you
  * can set this in the environment variable `LLAMA_PATH`.
  */
-export interface LlamaCppInputs extends BaseLLMParams {
-    /** Prompt processing batch size. */
-    batchSize?: number;
-    /** Text context size. */
-    contextSize?: number;
-    /** Embedding mode only. */
-    embedding?: boolean;
-    /** Use fp16 for KV cache. */
-    f16Kv?: boolean;
-    /** Number of layers to store in VRAM. */
-    gpuLayers?: number;
-    /** The llama_eval() call computes all logits, not just the last one. */
-    logitsAll?: boolean;
-    /** If true, reduce VRAM usage at the cost of performance. */
-    lowVram?: boolean;
-    /** Path to the model on the filesystem. */
-    modelPath: string;
-    /** If null, a random seed will be used. */
-    seed?: null | number;
-    /** The randomness of the responses, e.g. 0.1 deterministic, 1.5 creative, 0.8 balanced, 0 disables. */
-    temperature?: number;
-    /** Consider the n most likely tokens, where n is 1 to vocabulary size, 0 disables (uses full vocabulary). Note: only applies when `temperature` > 0. */
-    topK?: number;
-    /** Selects the smallest token set whose probability exceeds P, where P is between 0 - 1, 1 disables. Note: only applies when `temperature` > 0. */
-    topP?: number;
-    /** Force system to keep model in RAM. */
-    useMlock?: boolean;
-    /** Use mmap if possible. */
-    useMmap?: boolean;
-    /** Only load the vocabulary, no weights. */
-    vocabOnly?: boolean;
+export interface LlamaCppInputs extends LlamaBaseCppInputs, BaseLLMParams {
 }
 export interface LlamaCppCallOptions extends BaseLLMCallOptions {
     /** The maximum number of tokens the response should contain. */
@@ -51,18 +22,11 @@ export interface LlamaCppCallOptions extends BaseLLMCallOptions {
 export declare class LlamaCpp extends LLM<LlamaCppCallOptions> {
     CallOptions: LlamaCppCallOptions;
     static inputs: LlamaCppInputs;
-    batchSize?: number;
-    contextSize?: number;
-    embedding?: boolean;
-    f16Kv?: boolean;
-    gpuLayers?: number;
-    logitsAll?: boolean;
-    lowVram?: boolean;
-    seed?: null | number;
-    useMlock?: boolean;
-    useMmap?: boolean;
-    vocabOnly?: boolean;
-    modelPath: string;
+    maxTokens?: number;
+    temperature?: number;
+    topK?: number;
+    topP?: number;
+    trimWhitespaceSuffix?: boolean;
     _model: LlamaModel;
     _context: LlamaContext;
     _session: LlamaChatSession;

package/dist/llms/llama_cpp.js

@@ -1,4 +1,4 @@
-import { LlamaModel, LlamaContext, LlamaChatSession } from "node-llama-cpp";
+import { createLlamaModel, createLlamaContext, createLlamaSession, } from "../util/llama_cpp.js";
 import { LLM } from "./base.js";
 /**
  *  To use this model you need to have the `node-llama-cpp` module installed.
@@ -12,73 +12,31 @@ export class LlamaCpp extends LLM {
     }
     constructor(inputs) {
         super(inputs);
-        Object.defineProperty(this, "batchSize", {
+        Object.defineProperty(this, "maxTokens", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "contextSize", {
+        Object.defineProperty(this, "temperature", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "embedding", {
+        Object.defineProperty(this, "topK", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "f16Kv", {
+        Object.defineProperty(this, "topP", {
             enumerable: true,
             configurable: true,
             writable: true,
             value: void 0
         });
-        Object.defineProperty(this, "gpuLayers", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "logitsAll", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "lowVram", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "seed", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "useMlock", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "useMmap", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "vocabOnly", {
-            enumerable: true,
-            configurable: true,
-            writable: true,
-            value: void 0
-        });
-        Object.defineProperty(this, "modelPath", {
+        Object.defineProperty(this, "trimWhitespaceSuffix", {
             enumerable: true,
             configurable: true,
             writable: true,
@@ -102,29 +60,31 @@ export class LlamaCpp extends LLM {
             writable: true,
             value: void 0
         });
-        this.batchSize = inputs.batchSize;
-        this.contextSize = inputs.contextSize;
-        this.embedding = inputs.embedding;
-        this.f16Kv = inputs.f16Kv;
-        this.gpuLayers = inputs.gpuLayers;
-        this.logitsAll = inputs.logitsAll;
-        this.lowVram = inputs.lowVram;
-        this.modelPath = inputs.modelPath;
-        this.seed = inputs.seed;
-        this.useMlock = inputs.useMlock;
-        this.useMmap = inputs.useMmap;
-        this.vocabOnly = inputs.vocabOnly;
-        this._model = new LlamaModel(inputs);
-        this._context = new LlamaContext({ model: this._model });
-        this._session = new LlamaChatSession({ context: this._context });
+        this.maxTokens = inputs?.maxTokens;
+        this.temperature = inputs?.temperature;
+        this.topK = inputs?.topK;
+        this.topP = inputs?.topP;
+        this.trimWhitespaceSuffix = inputs?.trimWhitespaceSuffix;
+        this._model = createLlamaModel(inputs);
+        this._context = createLlamaContext(this._model, inputs);
+        this._session = createLlamaSession(this._context);
     }
     _llmType() {
         return "llama2_cpp";
     }
     /** @ignore */
-    async _call(prompt, options) {
+    async _call(prompt, 
+    // @ts-expect-error - TS6133: 'options' is declared but its value is never read.
+    options) {
         try {
-            const completion = await this._session.prompt(prompt, options);
+            const promptOptions = {
+                maxTokens: this?.maxTokens,
+                temperature: this?.temperature,
+                topK: this?.topK,
+                topP: this?.topP,
+                trimWhitespaceSuffix: this?.trimWhitespaceSuffix,
+            };
+            const completion = await this._session.prompt(prompt, promptOptions);
             return completion;
         }
         catch (e) {

package/dist/load/import_constants.cjs

@@ -25,6 +25,7 @@ exports.optionalImportEntrypoints = [
     "langchain/embeddings/hf_transformers",
     "langchain/embeddings/googlevertexai",
     "langchain/embeddings/googlepalm",
+    "langchain/embeddings/llama_cpp",
     "langchain/llms/load",
     "langchain/llms/cohere",
     "langchain/llms/hf",

package/dist/load/import_constants.js

@@ -22,6 +22,7 @@ export const optionalImportEntrypoints = [
     "langchain/embeddings/hf_transformers",
     "langchain/embeddings/googlevertexai",
     "langchain/embeddings/googlepalm",
+    "langchain/embeddings/llama_cpp",
     "langchain/llms/load",
     "langchain/llms/cohere",
     "langchain/llms/hf",