pull:初次提交

This commit is contained in:
Yep_Q
2025-09-08 04:48:28 +08:00
parent 5c0619656d
commit f64f498365
11751 changed files with 1953723 additions and 0 deletions

View File

@@ -0,0 +1,41 @@
import type { INodeTypeBaseDescription, IVersionedNodeType } from 'n8n-workflow';
import { VersionedNodeType } from 'n8n-workflow';
import { ChainSummarizationV1 } from './V1/ChainSummarizationV1.node';
import { ChainSummarizationV2 } from './V2/ChainSummarizationV2.node';
export class ChainSummarization extends VersionedNodeType {
constructor() {
const baseDescription: INodeTypeBaseDescription = {
displayName: 'Summarization Chain',
name: 'chainSummarization',
icon: 'fa:link',
iconColor: 'black',
group: ['transform'],
description: 'Transforms text into a concise summary',
codex: {
alias: ['LangChain'],
categories: ['AI'],
subcategories: {
AI: ['Chains', 'Root Nodes'],
},
resources: {
primaryDocumentation: [
{
url: 'https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.chainsummarization/',
},
],
},
},
defaultVersion: 2.1,
};
const nodeVersions: IVersionedNodeType['nodeVersions'] = {
1: new ChainSummarizationV1(baseDescription),
2: new ChainSummarizationV2(baseDescription),
2.1: new ChainSummarizationV2(baseDescription),
};
super(nodeVersions, baseDescription);
}
}

View File

@@ -0,0 +1,264 @@
import type { Document } from '@langchain/core/documents';
import type { BaseLanguageModel } from '@langchain/core/language_models/base';
import { PromptTemplate } from '@langchain/core/prompts';
import type { SummarizationChainParams } from 'langchain/chains';
import { loadSummarizationChain } from 'langchain/chains';
import {
NodeConnectionTypes,
type INodeTypeBaseDescription,
type IExecuteFunctions,
type INodeExecutionData,
type INodeType,
type INodeTypeDescription,
} from 'n8n-workflow';
import { N8nBinaryLoader } from '@utils/N8nBinaryLoader';
import { N8nJsonLoader } from '@utils/N8nJsonLoader';
import { getTemplateNoticeField } from '@utils/sharedFields';
import { REFINE_PROMPT_TEMPLATE, DEFAULT_PROMPT_TEMPLATE } from '../prompt';
export class ChainSummarizationV1 implements INodeType {
description: INodeTypeDescription;
constructor(baseDescription: INodeTypeBaseDescription) {
this.description = {
...baseDescription,
version: 1,
defaults: {
name: 'Summarization Chain',
color: '#909298',
},
inputs: [
NodeConnectionTypes.Main,
{
displayName: 'Model',
maxConnections: 1,
type: NodeConnectionTypes.AiLanguageModel,
required: true,
},
{
displayName: 'Document',
maxConnections: 1,
type: NodeConnectionTypes.AiDocument,
required: true,
},
],
outputs: [NodeConnectionTypes.Main],
credentials: [],
properties: [
getTemplateNoticeField(1951),
{
displayName: 'Type',
name: 'type',
type: 'options',
description: 'The type of summarization to run',
default: 'map_reduce',
options: [
{
name: 'Map Reduce (Recommended)',
value: 'map_reduce',
description:
'Summarize each document (or chunk) individually, then summarize those summaries',
},
{
name: 'Refine',
value: 'refine',
description:
'Summarize the first document (or chunk). Then update that summary based on the next document (or chunk), and repeat.',
},
{
name: 'Stuff',
value: 'stuff',
description: 'Pass all documents (or chunks) at once. Ideal for small datasets.',
},
],
},
{
displayName: 'Options',
name: 'options',
type: 'collection',
default: {},
placeholder: 'Add Option',
options: [
{
displayName: 'Final Prompt to Combine',
name: 'combineMapPrompt',
type: 'string',
hint: 'The prompt to combine individual summaries',
displayOptions: {
show: {
'/type': ['map_reduce'],
},
},
default: DEFAULT_PROMPT_TEMPLATE,
typeOptions: {
rows: 6,
},
},
{
displayName: 'Individual Summary Prompt',
name: 'prompt',
type: 'string',
default: DEFAULT_PROMPT_TEMPLATE,
hint: 'The prompt to summarize an individual document (or chunk)',
displayOptions: {
show: {
'/type': ['map_reduce'],
},
},
typeOptions: {
rows: 6,
},
},
{
displayName: 'Prompt',
name: 'prompt',
type: 'string',
default: DEFAULT_PROMPT_TEMPLATE,
displayOptions: {
show: {
'/type': ['stuff'],
},
},
typeOptions: {
rows: 6,
},
},
{
displayName: 'Subsequent (Refine) Prompt',
name: 'refinePrompt',
type: 'string',
displayOptions: {
show: {
'/type': ['refine'],
},
},
default: REFINE_PROMPT_TEMPLATE,
hint: 'The prompt to refine the summary based on the next document (or chunk)',
typeOptions: {
rows: 6,
},
},
{
displayName: 'Initial Prompt',
name: 'refineQuestionPrompt',
type: 'string',
displayOptions: {
show: {
'/type': ['refine'],
},
},
default: DEFAULT_PROMPT_TEMPLATE,
hint: 'The prompt for the first document (or chunk)',
typeOptions: {
rows: 6,
},
},
],
},
],
};
}
async execute(this: IExecuteFunctions): Promise<INodeExecutionData[][]> {
this.logger.debug('Executing Vector Store QA Chain');
const type = this.getNodeParameter('type', 0) as 'map_reduce' | 'stuff' | 'refine';
const model = (await this.getInputConnectionData(
NodeConnectionTypes.AiLanguageModel,
0,
)) as BaseLanguageModel;
const documentInput = (await this.getInputConnectionData(NodeConnectionTypes.AiDocument, 0)) as
| N8nJsonLoader
| Array<Document<Record<string, unknown>>>;
const options = this.getNodeParameter('options', 0, {}) as {
prompt?: string;
refineQuestionPrompt?: string;
refinePrompt?: string;
combineMapPrompt?: string;
};
const chainArgs: SummarizationChainParams = {
type,
};
// Map reduce prompt override
if (type === 'map_reduce') {
const mapReduceArgs = chainArgs as SummarizationChainParams & {
type: 'map_reduce';
};
if (options.combineMapPrompt) {
mapReduceArgs.combineMapPrompt = new PromptTemplate({
template: options.combineMapPrompt,
inputVariables: ['text'],
});
}
if (options.prompt) {
mapReduceArgs.combinePrompt = new PromptTemplate({
template: options.prompt,
inputVariables: ['text'],
});
}
}
// Stuff prompt override
if (type === 'stuff') {
const stuffArgs = chainArgs as SummarizationChainParams & {
type: 'stuff';
};
if (options.prompt) {
stuffArgs.prompt = new PromptTemplate({
template: options.prompt,
inputVariables: ['text'],
});
}
}
// Refine prompt override
if (type === 'refine') {
const refineArgs = chainArgs as SummarizationChainParams & {
type: 'refine';
};
if (options.refinePrompt) {
refineArgs.refinePrompt = new PromptTemplate({
template: options.refinePrompt,
inputVariables: ['existing_answer', 'text'],
});
}
if (options.refineQuestionPrompt) {
refineArgs.questionPrompt = new PromptTemplate({
template: options.refineQuestionPrompt,
inputVariables: ['text'],
});
}
}
const chain = loadSummarizationChain(model, chainArgs);
const items = this.getInputData();
const returnData: INodeExecutionData[] = [];
for (let itemIndex = 0; itemIndex < items.length; itemIndex++) {
let processedDocuments: Document[];
if (documentInput instanceof N8nJsonLoader || documentInput instanceof N8nBinaryLoader) {
processedDocuments = await documentInput.processItem(items[itemIndex], itemIndex);
} else {
processedDocuments = documentInput;
}
const response = await chain.call({
input_documents: processedDocuments,
});
returnData.push({ json: { response } });
}
return [returnData];
}
}

View File

@@ -0,0 +1,388 @@
import type {
INodeTypeBaseDescription,
IExecuteFunctions,
INodeExecutionData,
INodeType,
INodeTypeDescription,
IDataObject,
INodeInputConfiguration,
} from 'n8n-workflow';
import { NodeConnectionTypes, sleep } from 'n8n-workflow';
import { getBatchingOptionFields, getTemplateNoticeField } from '@utils/sharedFields';
import { processItem } from './processItem';
import { REFINE_PROMPT_TEMPLATE, DEFAULT_PROMPT_TEMPLATE } from '../prompt';
function getInputs(parameters: IDataObject) {
const chunkingMode = parameters?.chunkingMode;
const operationMode = parameters?.operationMode;
const inputs: INodeInputConfiguration[] = [
{ displayName: '', type: 'main' },
{
displayName: 'Model',
maxConnections: 1,
type: 'ai_languageModel',
required: true,
},
];
if (operationMode === 'documentLoader') {
inputs.push({
displayName: 'Document',
type: 'ai_document',
required: true,
maxConnections: 1,
});
return inputs;
}
if (chunkingMode === 'advanced') {
inputs.push({
displayName: 'Text Splitter',
type: 'ai_textSplitter',
required: false,
maxConnections: 1,
});
return inputs;
}
return inputs;
}
export class ChainSummarizationV2 implements INodeType {
description: INodeTypeDescription;
constructor(baseDescription: INodeTypeBaseDescription) {
this.description = {
...baseDescription,
version: [2, 2.1],
defaults: {
name: 'Summarization Chain',
color: '#909298',
},
inputs: `={{ ((parameter) => { ${getInputs.toString()}; return getInputs(parameter) })($parameter) }}`,
outputs: [NodeConnectionTypes.Main],
credentials: [],
properties: [
getTemplateNoticeField(1951),
{
displayName: 'Data to Summarize',
name: 'operationMode',
noDataExpression: true,
type: 'options',
description: 'How to pass data into the summarization chain',
default: 'nodeInputJson',
options: [
{
name: 'Use Node Input (JSON)',
value: 'nodeInputJson',
description: 'Summarize the JSON data coming into this node from the previous one',
},
{
name: 'Use Node Input (Binary)',
value: 'nodeInputBinary',
description: 'Summarize the binary data coming into this node from the previous one',
},
{
name: 'Use Document Loader',
value: 'documentLoader',
description: 'Use a loader sub-node with more configuration options',
},
],
},
{
displayName: 'Chunking Strategy',
name: 'chunkingMode',
noDataExpression: true,
type: 'options',
description: 'Chunk splitting strategy',
default: 'simple',
options: [
{
name: 'Simple (Define Below)',
value: 'simple',
},
{
name: 'Advanced',
value: 'advanced',
description: 'Use a splitter sub-node with more configuration options',
},
],
displayOptions: {
show: {
'/operationMode': ['nodeInputJson', 'nodeInputBinary'],
},
},
},
{
displayName: 'Characters Per Chunk',
name: 'chunkSize',
description:
'Controls the max size (in terms of number of characters) of the final document chunk',
type: 'number',
default: 1000,
displayOptions: {
show: {
'/chunkingMode': ['simple'],
},
},
},
{
displayName: 'Chunk Overlap (Characters)',
name: 'chunkOverlap',
type: 'number',
description: 'Specifies how much characters overlap there should be between chunks',
default: 200,
displayOptions: {
show: {
'/chunkingMode': ['simple'],
},
},
},
{
displayName: 'Options',
name: 'options',
type: 'collection',
default: {},
placeholder: 'Add Option',
options: [
{
displayName: 'Input Data Field Name',
name: 'binaryDataKey',
type: 'string',
default: 'data',
description:
'The name of the field in the agent or chains input that contains the binary file to be processed',
displayOptions: {
show: {
'/operationMode': ['nodeInputBinary'],
},
},
},
{
displayName: 'Summarization Method and Prompts',
name: 'summarizationMethodAndPrompts',
type: 'fixedCollection',
default: {
values: {
summarizationMethod: 'map_reduce',
prompt: DEFAULT_PROMPT_TEMPLATE,
combineMapPrompt: DEFAULT_PROMPT_TEMPLATE,
},
},
placeholder: 'Add Option',
typeOptions: {},
options: [
{
name: 'values',
displayName: 'Values',
values: [
{
displayName: 'Summarization Method',
name: 'summarizationMethod',
type: 'options',
description: 'The type of summarization to run',
default: 'map_reduce',
options: [
{
name: 'Map Reduce (Recommended)',
value: 'map_reduce',
description:
'Summarize each document (or chunk) individually, then summarize those summaries',
},
{
name: 'Refine',
value: 'refine',
description:
'Summarize the first document (or chunk). Then update that summary based on the next document (or chunk), and repeat.',
},
{
name: 'Stuff',
value: 'stuff',
description:
'Pass all documents (or chunks) at once. Ideal for small datasets.',
},
],
},
{
displayName: 'Individual Summary Prompt',
name: 'combineMapPrompt',
type: 'string',
hint: 'The prompt to summarize an individual document (or chunk)',
displayOptions: {
hide: {
'/options.summarizationMethodAndPrompts.values.summarizationMethod': [
'stuff',
'refine',
],
},
},
default: DEFAULT_PROMPT_TEMPLATE,
typeOptions: {
rows: 9,
},
},
{
displayName: 'Final Prompt to Combine',
name: 'prompt',
type: 'string',
default: DEFAULT_PROMPT_TEMPLATE,
hint: 'The prompt to combine individual summaries',
displayOptions: {
hide: {
'/options.summarizationMethodAndPrompts.values.summarizationMethod': [
'stuff',
'refine',
],
},
},
typeOptions: {
rows: 9,
},
},
{
displayName: 'Prompt',
name: 'prompt',
type: 'string',
default: DEFAULT_PROMPT_TEMPLATE,
displayOptions: {
hide: {
'/options.summarizationMethodAndPrompts.values.summarizationMethod': [
'refine',
'map_reduce',
],
},
},
typeOptions: {
rows: 9,
},
},
{
displayName: 'Subsequent (Refine) Prompt',
name: 'refinePrompt',
type: 'string',
displayOptions: {
hide: {
'/options.summarizationMethodAndPrompts.values.summarizationMethod': [
'stuff',
'map_reduce',
],
},
},
default: REFINE_PROMPT_TEMPLATE,
hint: 'The prompt to refine the summary based on the next document (or chunk)',
typeOptions: {
rows: 9,
},
},
{
displayName: 'Initial Prompt',
name: 'refineQuestionPrompt',
type: 'string',
displayOptions: {
hide: {
'/options.summarizationMethodAndPrompts.values.summarizationMethod': [
'stuff',
'map_reduce',
],
},
},
default: DEFAULT_PROMPT_TEMPLATE,
hint: 'The prompt for the first document (or chunk)',
typeOptions: {
rows: 9,
},
},
],
},
],
},
getBatchingOptionFields({
show: {
'@version': [{ _cnd: { gte: 2.1 } }],
},
}),
],
},
],
};
}
async execute(this: IExecuteFunctions): Promise<INodeExecutionData[][]> {
this.logger.debug('Executing Summarization Chain V2');
const operationMode = this.getNodeParameter('operationMode', 0, 'nodeInputJson') as
| 'nodeInputJson'
| 'nodeInputBinary'
| 'documentLoader';
const chunkingMode = this.getNodeParameter('chunkingMode', 0, 'simple') as
| 'simple'
| 'advanced';
const items = this.getInputData();
const returnData: INodeExecutionData[] = [];
const batchSize = this.getNodeParameter('options.batching.batchSize', 0, 5) as number;
const delayBetweenBatches = this.getNodeParameter(
'options.batching.delayBetweenBatches',
0,
0,
) as number;
if (this.getNode().typeVersion >= 2.1 && batchSize > 1) {
// Batch processing
for (let i = 0; i < items.length; i += batchSize) {
const batch = items.slice(i, i + batchSize);
const batchPromises = batch.map(async (item, batchItemIndex) => {
const itemIndex = i + batchItemIndex;
return await processItem(this, itemIndex, item, operationMode, chunkingMode);
});
const batchResults = await Promise.allSettled(batchPromises);
batchResults.forEach((response, index) => {
if (response.status === 'rejected') {
const error = response.reason as Error;
if (this.continueOnFail()) {
returnData.push({
json: { error: error.message },
pairedItem: { item: i + index },
});
} else {
throw error;
}
} else {
const output = response.value;
returnData.push({ json: { output } });
}
});
// Add delay between batches if not the last batch
if (i + batchSize < items.length && delayBetweenBatches > 0) {
await sleep(delayBetweenBatches);
}
}
} else {
for (let itemIndex = 0; itemIndex < items.length; itemIndex++) {
try {
const response = await processItem(
this,
itemIndex,
items[itemIndex],
operationMode,
chunkingMode,
);
returnData.push({ json: { response } });
} catch (error) {
if (this.continueOnFail()) {
returnData.push({ json: { error: error.message }, pairedItem: { item: itemIndex } });
continue;
}
throw error;
}
}
}
return [returnData];
}
}

View File

@@ -0,0 +1,107 @@
import type { Document } from '@langchain/core/documents';
import type { BaseLanguageModel } from '@langchain/core/language_models/base';
import type { ChainValues } from '@langchain/core/utils/types';
import { RecursiveCharacterTextSplitter, type TextSplitter } from '@langchain/textsplitters';
import { loadSummarizationChain } from 'langchain/chains';
import { type IExecuteFunctions, type INodeExecutionData, NodeConnectionTypes } from 'n8n-workflow';
import { N8nBinaryLoader } from '@utils/N8nBinaryLoader';
import { N8nJsonLoader } from '@utils/N8nJsonLoader';
import { getTracingConfig } from '@utils/tracing';
import { getChainPromptsArgs } from '../helpers';
export async function processItem(
ctx: IExecuteFunctions,
itemIndex: number,
item: INodeExecutionData,
operationMode: string,
chunkingMode: 'simple' | 'advanced' | 'none',
): Promise<ChainValues | undefined> {
const model = (await ctx.getInputConnectionData(
NodeConnectionTypes.AiLanguageModel,
0,
)) as BaseLanguageModel;
const summarizationMethodAndPrompts = ctx.getNodeParameter(
'options.summarizationMethodAndPrompts.values',
itemIndex,
{},
) as {
prompt?: string;
refineQuestionPrompt?: string;
refinePrompt?: string;
summarizationMethod: 'map_reduce' | 'stuff' | 'refine';
combineMapPrompt?: string;
};
const chainArgs = getChainPromptsArgs(
summarizationMethodAndPrompts.summarizationMethod ?? 'map_reduce',
summarizationMethodAndPrompts,
);
const chain = loadSummarizationChain(model, chainArgs);
let processedDocuments: Document[];
// Use dedicated document loader input to load documents
if (operationMode === 'documentLoader') {
const documentInput = (await ctx.getInputConnectionData(NodeConnectionTypes.AiDocument, 0)) as
| N8nJsonLoader
| Array<Document<Record<string, unknown>>>;
const isN8nLoader =
documentInput instanceof N8nJsonLoader || documentInput instanceof N8nBinaryLoader;
processedDocuments = isN8nLoader
? await documentInput.processItem(item, itemIndex)
: documentInput;
return await chain.withConfig(getTracingConfig(ctx)).invoke({
input_documents: processedDocuments,
});
} else if (['nodeInputJson', 'nodeInputBinary'].includes(operationMode)) {
// Take the input and use binary or json loader
let textSplitter: TextSplitter | undefined;
switch (chunkingMode) {
// In simple mode we use recursive character splitter with default settings
case 'simple':
const chunkSize = ctx.getNodeParameter('chunkSize', itemIndex, 1000) as number;
const chunkOverlap = ctx.getNodeParameter('chunkOverlap', itemIndex, 200) as number;
textSplitter = new RecursiveCharacterTextSplitter({ chunkOverlap, chunkSize });
break;
// In advanced mode user can connect text splitter node so we just retrieve it
case 'advanced':
textSplitter = (await ctx.getInputConnectionData(NodeConnectionTypes.AiTextSplitter, 0)) as
| TextSplitter
| undefined;
break;
default:
break;
}
let processor: N8nJsonLoader | N8nBinaryLoader;
if (operationMode === 'nodeInputBinary') {
const binaryDataKey = ctx.getNodeParameter(
'options.binaryDataKey',
itemIndex,
'data',
) as string;
processor = new N8nBinaryLoader(ctx, 'options.', binaryDataKey, textSplitter);
} else {
processor = new N8nJsonLoader(ctx, 'options.', textSplitter);
}
const processedItem = await processor.processItem(item, itemIndex);
return await chain.invoke(
{
input_documents: processedItem,
},
{ signal: ctx.getExecutionCancelSignal() },
);
}
return undefined;
}

View File

@@ -0,0 +1,71 @@
import { PromptTemplate } from '@langchain/core/prompts';
import type { SummarizationChainParams } from 'langchain/chains';
interface ChainTypeOptions {
combineMapPrompt?: string;
prompt?: string;
refinePrompt?: string;
refineQuestionPrompt?: string;
}
export function getChainPromptsArgs(
type: 'stuff' | 'map_reduce' | 'refine',
options: ChainTypeOptions,
) {
const chainArgs: SummarizationChainParams = {
type,
};
// Map reduce prompt override
if (type === 'map_reduce') {
const mapReduceArgs = chainArgs as SummarizationChainParams & {
type: 'map_reduce';
};
if (options.combineMapPrompt) {
mapReduceArgs.combineMapPrompt = new PromptTemplate({
template: options.combineMapPrompt,
inputVariables: ['text'],
});
}
if (options.prompt) {
mapReduceArgs.combinePrompt = new PromptTemplate({
template: options.prompt,
inputVariables: ['text'],
});
}
}
// Stuff prompt override
if (type === 'stuff') {
const stuffArgs = chainArgs as SummarizationChainParams & {
type: 'stuff';
};
if (options.prompt) {
stuffArgs.prompt = new PromptTemplate({
template: options.prompt,
inputVariables: ['text'],
});
}
}
// Refine prompt override
if (type === 'refine') {
const refineArgs = chainArgs as SummarizationChainParams & {
type: 'refine';
};
if (options.refinePrompt) {
refineArgs.refinePrompt = new PromptTemplate({
template: options.refinePrompt,
inputVariables: ['existing_answer', 'text'],
});
}
if (options.refineQuestionPrompt) {
refineArgs.questionPrompt = new PromptTemplate({
template: options.refineQuestionPrompt,
inputVariables: ['text'],
});
}
}
return chainArgs;
}

View File

@@ -0,0 +1,20 @@
export const REFINE_PROMPT_TEMPLATE = `Your job is to produce a final summary
We have provided an existing summary up to a certain point: "{existing_answer}"
We have the opportunity to refine the existing summary
(only if needed) with some more context below.
------------
"{text}"
------------
Given the new context, refine the original summary
If the context isn't useful, return the original summary.
REFINED SUMMARY:`;
export const DEFAULT_PROMPT_TEMPLATE = `Write a concise summary of the following:
"{text}"
CONCISE SUMMARY:`;