Merge branch 'main' into ms/add-vercel-adapter

.changeset/new-cups-dress.md

@@ -0,0 +1,5 @@
+---
+"@llamaindex/core": patch
+---
+
+The compact and refine response synthesizer (retrieved by using `getResponseSynthesizer('compact')`) has been fixed to return the original source nodes that were provided to it in its response. Previous to this it was returning the compacted text chunk documents.

.changeset/weak-cats-smash.md

@@ -0,0 +1,5 @@
+---
+"llamaindex": patch
+---
+
+withLlamaIndex now passes through webpack options to the passed in customized NextJS webpack config. Before it was only passing through the config.

apps/next/src/content/docs/llamaindex/setup/typescript.mdx

@@ -84,7 +84,7 @@ Imaging you put output file into `/dist/openai.js` but you are importing `llamai
 }
 ```
 
-In old module resolution, TypeScript will not be able to find the module because it is not follow the file structure, even you run `node index.js` successfully. (on Node.js >=16)
+In old module resolution, TypeScript will not be able to find the module because it is not following the file structure, even you run `node index.js` successfully. (on Node.js >=16)
 
 See more about [moduleResolution](https://www.typescriptlang.org/docs/handbook/modules/theory.html#module-resolution) or
 [TypeScript 5.0 blog](https://devblogs.microsoft.com/typescript/announcing-typescript-5-0/#--moduleresolution-bundler7).

packages/core/src/response-synthesizers/factory.ts

@@ -77,6 +77,16 @@ class Refine extends BaseSynthesizer {
     }
   }
 
+  async getResponse(
+    query: MessageContent,
+    nodes: NodeWithScore[],
+    stream: true,
+  ): Promise<AsyncIterable<EngineResponse>>;
+  async getResponse(
+    query: MessageContent,
+    nodes: NodeWithScore[],
+    stream: false,
+  ): Promise<EngineResponse>;
   async getResponse(
     query: MessageContent,
     nodes: NodeWithScore[],
@@ -197,6 +207,16 @@ class Refine extends BaseSynthesizer {
  * CompactAndRefine is a slight variation of Refine that first compacts the text chunks into the smallest possible number of chunks.
  */
 class CompactAndRefine extends Refine {
+  async getResponse(
+    query: MessageContent,
+    nodes: NodeWithScore[],
+    stream: true,
+  ): Promise<AsyncIterable<EngineResponse>>;
+  async getResponse(
+    query: MessageContent,
+    nodes: NodeWithScore[],
+    stream: false,
+  ): Promise<EngineResponse>;
   async getResponse(
     query: MessageContent,
     nodes: NodeWithScore[],
@@ -216,17 +236,24 @@ class CompactAndRefine extends Refine {
     const newTexts = this.promptHelper.repack(maxPrompt, textChunks);
     const newNodes = newTexts.map((text) => new TextNode({ text }));
     if (stream) {
-      return super.getResponse(
+      const streamResponse = await super.getResponse(
         query,
         newNodes.map((node) => ({ node })),
         true,
       );
+      return streamConverter(streamResponse, (chunk) => {
+        chunk.sourceNodes = nodes;
+        return chunk;
+      });
     }
-    return super.getResponse(
+
+    const originalResponse = await super.getResponse(
       query,
       newNodes.map((node) => ({ node })),
       false,
     );
+    originalResponse.sourceNodes = nodes;
+    return originalResponse;
   }
 }
 

packages/core/tests/response-synthesizers/compact-and-refine.test.ts

@@ -0,0 +1,66 @@
+import { describe, expect, test, vi } from "vitest";
+import type { LLMMetadata } from "../../llms/dist/index.js";
+import { getResponseSynthesizer } from "../../response-synthesizers/dist/index.js";
+import { Document } from "../../schema/dist/index.js";
+
+const mockLllm = () => ({
+  complete: vi.fn().mockImplementation(({ stream }) => {
+    const response = { text: "unimportant" };
+    if (!stream) {
+      return response;
+    }
+
+    function* gen() {
+      // yield a few times to make sure each chunk has the sourceNodes
+      yield response;
+      yield response;
+      yield response;
+    }
+
+    return gen();
+  }),
+  chat: vi.fn(),
+  metadata: {} as unknown as LLMMetadata,
+});
+
+describe("compact and refine response synthesizer", () => {
+  describe("synthesize", () => {
+    test("should return original sourceNodes with response when stream = false", async () => {
+      const synthesizer = getResponseSynthesizer("compact", {
+        llm: mockLllm(),
+      });
+
+      const sourceNode = { node: new Document({}), score: 1 };
+
+      const response = await synthesizer.synthesize(
+        {
+          query: "test",
+          nodes: [sourceNode],
+        },
+        false,
+      );
+
+      expect(response.sourceNodes).toEqual([sourceNode]);
+    });
+
+    test("should return original sourceNodes with response when stream = true", async () => {
+      const synthesizer = getResponseSynthesizer("compact", {
+        llm: mockLllm(),
+      });
+
+      const sourceNode = { node: new Document({}), score: 1 };
+
+      const response = await synthesizer.synthesize(
+        {
+          query: "test",
+          nodes: [sourceNode],
+        },
+        true,
+      );
+
+      for await (const chunk of response) {
+        expect(chunk.sourceNodes).toEqual([sourceNode]);
+      }
+    });
+  });
+});

packages/llamaindex/src/next.ts

@@ -41,7 +41,7 @@ export default function withLlamaIndex(config: any) {
   // eslint-disable-next-line @typescript-eslint/no-explicit-any
   config.webpack = function (webpackConfig: any, options: any) {
     if (userWebpack) {
-      webpackConfig = userWebpack(webpackConfig);
+      webpackConfig = userWebpack(webpackConfig, options);
     }
     webpackConfig.resolve.alias = {
       ...webpackConfig.resolve.alias,