3.7.6 typescript build errors

[jozefchutka]

System Info

transformers.js: 3.7.6

Environment/Platform

• Website/web-app
• Browser extension
• Server-side (e.g., Node.js, Deno, Bun)
• Desktop app (e.g., Electron)
• Other (e.g., VSCode extension)

Description

Running a very basic typescript build using transformers.js 3.7.6 fails.

Reproduction

package.json

{
	"devDependencies": {
		"@huggingface/transformers": "^3.7.6"
	}
}

tsconfig.json

{
	"compilerOptions": {
	  "target": "ESNEXT",
	  "lib": ["ESNext", "WebWorker"],
	  "module": "ESNext",
	  "moduleResolution": "bundler",
	  "baseUrl": "./",
	  "paths": {}
	}
  }

Main.mts

import { AutomaticSpeechRecognitionConfig, env, pipeline, ProgressCallback } from "@huggingface/transformers";

runnig tsc command fails with the following errors:

node_modules/@huggingface/transformers/types/models/mgp_str/processing_mgp_str.d.ts:52:5 - error TS2416: Property 'batch_decode' in type 'MgpstrProcessor' is not assignable to the same property in base type 'Processor'.
  Type '([char_logits, bpe_logits, wp_logits]: [Tensor, Tensor, Tensor]) => { generated_text: string[]; scores: number[]; char_preds: string[]; bpe_preds: string[]; wp_preds: string[]; }' is not assignable to type '(batch: number[][] | Tensor, decode_args?: any) => string[]'.
    Types of parameters '__0' and 'batch' are incompatible.
      Type 'number[][] | Tensor' is not assignable to type '[Tensor, Tensor, Tensor]'.
        Type 'number[][]' is not assignable to type '[Tensor, Tensor, Tensor]'.
          Target requires 3 element(s) but source may have fewer.

52     batch_decode([char_logits, bpe_logits, wp_logits]: [import("../../utils/tensor.js").Tensor, import("../../utils/tensor.js").Tensor, import("../../utils/tensor.js").Tensor]): {
       ~~~~~~~~~~~~

node_modules/@huggingface/transformers/types/pipelines.d.ts:1393:59 - error TS2304: Cannot find name 'HTMLCanvasElement'.

1393 export type ImageInput = string | RawImage | URL | Blob | HTMLCanvasElement | OffscreenCanvas;
                                                               ~~~~~~~~~~~~~~~~~

node_modules/@huggingface/transformers/types/utils/image.d.ts:18:57 - error TS2304: Cannot find name 'HTMLCanvasElement'.

18     static read(input: RawImage | string | URL | Blob | HTMLCanvasElement | OffscreenCanvas): Promise<RawImage>;
                                                           ~~~~~~~~~~~~~~~~~

node_modules/@huggingface/transformers/types/utils/image.d.ts:24:31 - error TS2304: Cannot find name 'HTMLCanvasElement'.

24     static fromCanvas(canvas: HTMLCanvasElement | OffscreenCanvas): RawImage;
                                 ~~~~~~~~~~~~~~~~~

node_modules/@huggingface/transformers/types/utils/video.d.ts:11:49 - error TS2304: Cannot find name 'HTMLVideoElement'.

11 export function load_video(src: string | Blob | HTMLVideoElement, { num_frames, fps }?: {
                                                   ~~~~~~~~~~~~~~~~

node_modules/onnxruntime-common/dist/esm/tensor-factory.d.ts:203:29 - error TS2304: Cannot find name 'HTMLImageElement'.

203     fromImage(imageElement: HTMLImageElement, options?: TensorFromImageElementOptions): Promise<TypedTensor<'float32'> | TypedTensor<'uint8'>>;
                                ~~~~~~~~~~~~~~~~


Found 6 errors in 5 files.

Errors  Files
     1  node_modules/@huggingface/transformers/types/models/mgp_str/processing_mgp_str.d.ts:52
     1  node_modules/@huggingface/transformers/types/pipelines.d.ts:1393
     2  node_modules/@huggingface/transformers/types/utils/image.d.ts:18
     1  node_modules/@huggingface/transformers/types/utils/video.d.ts:11
     1  node_modules/onnxruntime-common/dist/esm/tensor-factory.d.ts:203

This issue was originally reported in ticket #1337 which was closed as a part of #1436 merge which landed in 3.7.6 release.

However, the issues originally reported are not fixed in 3.7.6 so I am recreating the closed ticket.

Workaround: #1337 (comment)

[xenova]

cc @nico-martin -- looks like most of the issues were fixed but we can make a few more improvements :)

processing_mgp_str is a bit annoying because this problem was inherited from the python transformers library. Hopefully there's a good way to fix the types here.