Error: Could not locate file (500 error)

[iamhenry]

System Info

node: 22.7 nextjs: 14

Environment/Platform

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

Description

I'm trying to load https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct and i get the error Error: Could not locate file: "https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct/resolve/main/onnx/decoder_model_merged_quantized.onnx".

seems like someone else is running into the same issue https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct/discussions/1

how do i fix this? i tried copying the download link directly but somehow it appends .json to the URL like so: https://huggingface.co/https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct/resolve/main/onnx/model_quantized.onnx/resolve/main/config.json\

how do i fix this to point to the correct model?

I'd love to use the new qwen2.5 0.5B and the 1.5B

thank you in advanced

---

code:

import { pipeline, env, Pipeline } from '@xenova/transformers';

// Set the Xenova model directory
env.localModelPath = '/tmp/models';

let generator: Pipeline | null = null;

export async function POST(req: Request) {
  let text;
  try {
    ({ text } = await req.json());
    console.log('Received text:', text);
  } catch (error) {
    console.error('Error parsing request body:', error);
    return NextResponse.json({ error: 'Invalid request body' }, { status: 400 });
  }

  try {
    if (!generator) {
      console.log('Initializing model...');
      generator = await pipeline('text-generation', 'https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct/resolve/main/onnx/model_quantized.onnx');
      console.log('Model initialized successfully');
    }

    console.log('Generating text...');
    const messages = [
      { role: 'system', content: 'You are a CBT and ACT AI therapist. Provide advice or another way to frame the situation from a therapist perspective. Ensure it\'s only 1 sentence! Omit unnecessary words that aren\'t related to the feedback or advice.' },
      { role: 'user', content: text }
    ];

    // Apply chat template
    const prompt = generator.tokenizer.apply_chat_template(messages, {
      tokenize: false,
      add_generation_prompt: true,
    });

    console.log('Constructed prompt:', prompt);

    console.log('Calling generator...');
    const output = await generator(prompt, {
      max_new_tokens: 128,
      do_sample: true,
      temperature: 0.7,
      return_full_text: false,
    });

    console.log('Raw result:', JSON.stringify(output, null, 2));

    if (!output || !Array.isArray(output) || output.length === 0 || !output[0].generated_text) {
      throw new Error(`Invalid response from model: ${JSON.stringify(output)}`);
    }

    const generatedText = output[0].generated_text.trim();
    console.log('Generated text:', generatedText);
    return NextResponse.json({ result: generatedText });
  } catch (error) {
    console.error('Error in POST request:', error);
    console.error('Error stack:', error instanceof Error ? error.stack : 'No stack available');
    console.error('Error details:', JSON.stringify(error, Object.getOwnPropertyNames(error)));
    return NextResponse.json({ error: `Server error: ${error instanceof Error ? error.message : String(error)}` }, { status: 500 });
  }
}```

### Reproduction

1. I use this url to add it to Cursor IDE (https://huggingface.co/onnx-community/Qwen2.5-0.5B-Instruct)
2. It updates the code for me
3. get 500 error 

it works when i do qwen1.5 but not the latest 2.5

[iamhenry]

im getting the same error for the new llama 3.1 1B

how do i resolve this issue and prevent it for the future models?

Error: Could not locate file: "https://huggingface.co/onnx-community/Llama-3.2-1B/resolve/main/onnx/decoder_model_merged_quantized.onnx"

[iamhenry]

i downloaded the model manually and placed in the the onnx directory

now i get this error after renaming the model to decoder_model_merged_quantized.onnx: Something went wrong during model construction (most likely a missing operation)

[nicolasembleton]

You need to specify the file name (and sometimes it's missing the folder as the onnx model files are not always in the onnx folder).

In the case of Qwen2.5, you need to specify the model: "model_quantized" (or whichever exact model version you want to use, check in files tab in HuggingFace and then open the onnx folder, pick the model you want to use and use the name, without the extension).

But then that particular model doesn't work indeed. It's failing at .loadModel() when creating an inference session can't create session.

[iamhenry]

@nicolasembleton thx.

So does that mean qwen 2.5 or llama 3.2 isn't supported yet by transformers.js?

[flatsiedatsie]

Llama 3.2 1B seems to be supported, as there seems to be a demo available. The source code is missing though.

A good starting point might be the example code for Phi 3.5: https://github.com/huggingface/transformers.js-examples/blob/main/phi-3.5-webgpu/src/worker.js

For one thing, you don't need to provide the specific .onnx file. Transformers.js will grab the correct one based on the settings you provide when you create the pipeline / model.

More examples: https://github.com/xenova/transformers.js/tree/v3/examples https://github.com/huggingface/transformers.js-examples/tree/main

[iamhenry]

thank you and great find.

do you know if llama 3.2 1B will work with nextjs? I'm new to dev so i'm thinking it could be a pipeline not being compatible yet?

perhaps the @transformers pipeline isnt supported yet but @huggingface is?

[flatsiedatsie]

perhaps the @transformers pipeline isnt supported yet but https://github.com/huggingface is?

Do you mean version 2 and version 3 of Transformers.js? if so, you need to use Transformers.js V3.

[iamhenry]

@flatsiedatsie how do i do that? is there a step by step guide?

also are all the features from v2 still functional?

i would hate to install it and cause breaking changes.

thanks in advanced!

[flatsiedatsie]

V3 is much MUCH faster because it adds support for WebGPU.

Read through the documentation: https://huggingface.co/docs/transformers.js/index

And there are tons of examples (linked above). Always search in these issues too, there is loads of knowledge in them.

[iamhenry]

Appreciate your response, thx

[WrathofBhuvan11]

Actually; I'm trying to run word embeddings locally & seeing same issue-

export class EmbeddingService {
  constructor() {
    this.model = null;
    this.initializeModel();
  }

  // Load the embedding model
  async initializeModel() {
    const { pipeline } = await import('@xenova/transformers');
    this.model = await pipeline('feature-extraction', 
    'sentence-transformers/all-MiniLM-L6-v2');
  }

Here's the code & Error message -

    throw Error(`${message}: "${remoteURL}".`);
          ^

``Error: Could not locate file: "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/onnx/model_quantized.onnx".```