xenova/transformers.js

TypeError: Failed to fetch dynamically imported module

Closed this issue · 4 comments

sugarforever commented

System Info

@xenova/transformers 3.0.0-alpha.0
Chrome: Version 124.0.6367.93 (Official Build) (arm64)
OS: macOS 14.4.1 (23E224)

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

I ran pnpm run dev in the example webgpt-chat. I can download the model on http://localhost:5173. But it's not ready for chat due to the error reported in the console:

@xenova_transformers.js?v=9e5deabe:1386 Uncaught (in promise) Error: no available backend found. ERR: [webgpu] TypeError: Failed to fetch dynamically imported module: http://localhost:5173/ort-wasm-simd-threaded.jsep.mjs
    at pt (http://localhost:5173/node_modules/.vite/deps/@xenova_transformers.js?v=9e5deabe:1386:13)
    at async e.create (http://localhost:5173/node_modules/.vite/deps/@xenova_transformers.js?v=9e5deabe:1906:20)
    at async createInferenceSession (http://localhost:5173/node_modules/.vite/deps/@xenova_transformers.js?v=9e5deabe:9952:10)
    at async constructSessions (http://localhost:5173/node_modules/.vite/deps/@xenova_transformers.js?v=9e5deabe:17531:21)
    at async Promise.all (index 0)
    at async Phi3ForCausalLM.from_pretrained (http://localhost:5173/node_modules/.vite/deps/@xenova_transformers.js?v=9e5deabe:17782:14)
    at async AutoModelForCausalLM.from_pretrained (http://localhost:5173/node_modules/.vite/deps/@xenova_transformers.js?v=9e5deabe:20678:14)
    at async Promise.all (index 1)
    at async load (http://localhost:5173/src/worker.js?worker_file&type=module:137:32)
Screenshot 2024-05-09 at 9 28 23 PM

May I query if any setting required to make it work?

Btw, I can chat with the model on https://huggingface.co/spaces/Xenova/experimental-phi3-webgpu

Reproduction

  1. Git clone https://github.com/xenova/transformers.js.git
  2. Go to transformers.js/examples/webgpu-chat
  3. pnpm install
  4. pnpm run dev
  5. Visit http://localhost:5173 in Chrome and click Load Model button
HyperCrowd commented

I get the same issue, even when trying to pull in from a local checkout of onnx:

import { env, AutoModelForCausalLM, AutoTokenizer } from '@xenova/transformers'

env.backends.onnx.wasm.wasmPaths = '/onnxruntime-web/'
env.allowRemoteModels = false
env.allowLocalModels = true

const model_id = '../model';

const tokenizer = await AutoTokenizer.from_pretrained(model_id, {
    legacy: true
})

I have copied the contents of node_modules/onnxruntime-web/dist/ to public and it's trying to access a ort-wasm-simd-threaded.jsep.mjs file which does not exist in onnxruntime-web

xenova commented

This is because the demo uses an unreleased version of onnxruntime-web v1.18.0, which I have mentioned a few times when I've linked to the source code. When it is released, I will update the source code so that it works correctly. Thanks for understanding!

  • 👍2
sugarforever commented

This is because the demo uses an unreleased version of onnxruntime-web v1.18.0, which I have mentioned a few times when I've linked to the source code. When it is released, I will update the source code so that it works correctly. Thanks for understanding!

Thanks for the feedback. Looking forward to the release.

  • 👍1
xenova commented

It should now work (commits: here and here)!

  • 👍1