Getting ERR_PACKAGE_PATH_NOT_EXPORTED]: No "exports" main defined

[liorshk]

We are getting

Error [ERR_PACKAGE_PATH_NOT_EXPORTED]: No "exports" main defined in langchain\package.json
It started happening from versions 0.0.13 and up.
It works fine for 0.0.12.

[evad1n]

Same issue as well. The exports in the package.json have switched from

    "./embeddings": {
      "import": "./embeddings.mjs",
      "default": "./embeddings.js"
    },

to

    "./embeddings": {
      "types": "./embeddings.d.ts",
      "import": "./embeddings.js"
    },

[nfcampos]

The langchain package is now ESM only. You can fix that by adding "type": "module" to your package.json. You can find more information here https://hwchase17.github.io/langchainjs/docs/getting-started/

[irl-dan]

The langchain package is now ESM only. You can fix that by adding "type": "module" to your package.json. You can find more information here https://hwchase17.github.io/langchainjs/docs/getting-started/

@nfcampos I'm not sure I understand why this requirement exists. Is this on purpose or will this be changed? Happy to open up a PR with the fix if that would be accepted.

[nfcampos]

@irl-dan we converted the package to ESM so that we can support other environments outside of Node, which are ESM only, discussion here #152

You also have the option to use it without type: module by using dynamic import as mentioned here https://hwchase17.github.io/langchainjs/docs/getting-started/#commonjs-in-nodejs

[sboudouk]

Hello.

Using dynamic import on my side produce the same error.

const { OpenAI } = await import('langchain');

and

const { AnalyzeDocumentChain, loadSummarizationChain } = await import('langchain/chains');

tsconfig.json:

{
  "compilerOptions": {
    "module": "commonjs",
    "declaration": true,
    "removeComments": true,
    "emitDecoratorMetadata": true,
    "experimentalDecorators": true,
    "allowSyntheticDefaultImports": true,
    "target": "es2017",
    "sourceMap": true,
    "outDir": "./dist",
    "baseUrl": "./",
    "incremental": true,
    "skipLibCheck": true,
    "strictNullChecks": false,
    "noImplicitAny": false,
    "strictBindCallApply": false,
    "forceConsistentCasingInFileNames": false,
    "noFallthroughCasesInSwitch": false,
    "resolveJsonModule": true
  }
}

node version:

v18.12.1

Any idea ?

Note that changing es2017 to es2020 do not change anything to the problem.

[nfcampos]

Some more information on your issue here https://stackoverflow.com/questions/65265420/how-to-prevent-typescript-from-transpiling-dynamic-imports-into-require

A solution for you appears to be "moduleResolution": "node16", see here microsoft/TypeScript#43329 (comment)

[sboudouk]

Thanks for the fast reply.

Doing so, with :

    const { OpenAI } = await import_('langchain');
    const { AnalyzeDocumentChain, loadSummarizationChain } = await import_('langchain/chains');
    const { RecursiveCharacterTextSplitter } = await import_('langchain/text_splitter');

    const text = this.speakerRecognitionPromptFormatting(
      JSON.parse((conversation as any).toString()).prediction,
      'ok',
    );

    // text is a string.
    
    const model = new OpenAI({
      temperature: 0,
      openAIApiKey: ...,
    });
    /** Load the summarization chain. */
    const chain = loadSummarizationChain(model);
    const textSplitter = new RecursiveCharacterTextSplitter({ chunkSize: 1000 });
    const docs = await textSplitter.createDocuments([text]);
    const res = await chain.call({
      input_document: docs,
    });

Results in a Error: Document key input_documents not found. error. Also tried with AnalyzeDocumentChain from documentation snippet.

Do you want me to create a separate issue ?

[nfcampos]

I think the issue might be an s missing on input_document on the last line, should be input_documents

[sboudouk]

Yep it is. Thanks. Can I submit a PR on the documentation for this single simple correction ?

[nfcampos]

Yes, thank you

[sboudouk]

Unfortunately It might not be only a documentation issue, running :

    const text = this.speakerRecognitionPromptFormatting(
      JSON.parse((conversation as any).toString()).prediction,
      'oki',
    );
    console.log('text :', text);
    const model = new OpenAI({
      temperature: 0,
      openAIApiKey: ...,
    });
    /** Load the summarization chain. */
    const combineDocsChain = loadSummarizationChain(model);
    /** Pass this into the AnalyzeDocumentChain. */
    const chain = new AnalyzeDocumentChain({
      combineDocumentsChain: combineDocsChain,
    });
    const res = await chain.call({
      input_documents: text,
    });

Regarding input_documents or input_document with the s change I get different errors. By trying to match the code I get hit by TypeError: currentDocs.map is not a function, occuring in combine_docs_chain.js

I feel like I might be doing something wrong. Note that the text that i'm sending is just a string, not coming from any file.

[nfcampos]

Ah, input_documents is expected to be an array of Document, see eg the second code block in this page https://hwchase17.github.io/langchainjs/docs/modules/chains/summarization You can get an array of Documents from eg a TextSplitter, or a Loader

[ztratar]

When will langchain support CommonJS? Turning my backend package into a module is problematic for many other packages that rely on CommonJS patterns.

[sboudouk]

If it can helps someone running CommonJS (pretty common to be fair), you can import the langchains modules dynamically and without changing anything config related by using this import package

https://www.npmjs.com/package/@brillout/import