[Feature request] CommonJS compatability

[Emsu]

CommonJS compatibility In general, the feature you want added should be supported by HuggingFace's transformers library:

• If requesting a model, it must be listed here.
• If requesting a pipeline, it must be listed here.
  • If requesting a task, it must be listed here.

We're hoping in addition to ESM, that transformers.js can compile to both CJS and ESM targets.

Reason for request Why is it important that we add this feature? What is your intended use case? Remember, we are more likely to add support for models/pipelines/tasks that are popular (e.g., many downloads), or contain functionality that does not exist (e.g., new input type).

It's important for the wider audience of Javascript and typescript developers to be able to use the library especially those working in codebases where it'd be difficult to move everything to ESM.

Additional context Add any other context or screenshots about the feature request here.

Here's the main error I get when I try to use the package

Error [ERR_REQUIRE_ESM]: require() of ES Module .pnpm/@xenova+transformers@2.2.0/node_modules/@xenova/transformers/src/transformers.js from local-vectors/lib/embed.js not supported.
Instead change the require of transformers.js in local-vectors/lib/embed.js to a dynamic import() which is available in all CommonJS modules.
    at local-vectors/lib/embed.js:71:95 {
  code: 'ERR_REQUIRE_ESM'
}

I think you can have the package.json output main, module and types for CJS, ESM and TS support.

Here's two reference articles on the web: https://adamcoster.com/blog/commonjs-and-esm-importexport-compatibility-examples https://antfu.me/posts/publish-esm-and-cjs

[kungfooman]

We're hoping in addition to ESM, that transformers.js can compile to both CJS and ESM targets.

The distributed npm package already contains a CJS file, so transformers.js "can compile to both CJS and ESM" already?

Not sure how you run into that error or what your entire build setup looks like, can you share a test repo/setup to inspect?

If you don't use the distributed ES5 bundle but ES6 as it is, you need to use commonjs plugin in rollup via plugins: [commonjs(), nodeResolve()]... but I don't know what you are even trying to do right now.

And there already is a "types": "./types/transformers.d.ts", in package.json for types.

It's important for the wider audience of Javascript and typescript developers to be able to use the library especially those working in codebases where it'd be difficult to move everything to ESM.

In any case, you can already just use the distributed ES5 file and just keep your legacy stack (which I wouldn't recommend though, because it's legacy for a reason and ES6 is better in so many ways - and every browser supports it).

[xenova]

@Emsu For your use-case, are you not able to use the dynamic import syntax? We wrote a short tutorial for how to use the library with CommonJS imports here: https://huggingface.co/docs/transformers.js/tutorials/node

That said, if it is simple to also distribute the library as a CommonJS module (hopefully only 1-2 files in dist), then I think this is definitely something we can put some time into.

[xenova]

Linked comments: https://github.com/xenova/transformers.js/pull/147/files#r1229223950

[hwang251]

I think this would help fix this issue in LangChain, https://github.com/langchain-ai/langchainjs/issues/2992

[anilbandrapalli]

Hi, in my poc application i am using direct esm module import , it is working fine. but where as my actual application is ts application using webpack build. even though after using dynamic import also i am facing "ERR_REQUIRE_ESM" issue.

Even i tried using v3 branch but still same issue

Kindly guide us in this

[mschudt]

I think this would help fix this issue in LangChain, langchain-ai/langchainjs#2992

To be specific, this comment https://github.com/langchain-ai/langchainjs/issues/2992#issuecomment-2002350356 was the only thing that worked for me out of everything I've tried.

const TransformersApi = Function('return import("@xenova/transformers")')();
const { pipeline } = await TransformersApi;