Spaces:

webml-community

/

phi-3.5-webgpu

like 35

Running

Sorry if this is a silly question, I've tried to have a dig but I'm not sure whether my code just isn't working of whether I've misunderstood how the models are listed.

The Phi models seem incredible for free, lightweight models that run in the browser. I can see on the

• Transformer.js supported models page: https://huggingface.co/docs/transformers.js/en/index
• Xenova/transformers supported models page: https://www.jsdelivr.com/package/npm/@xenova/transformers
• Filtered list of transformers.js supported models: https://huggingface.co/models?library=transformers.js&sort=trending&search=phi

I also had a look on the Xenova CDN at https://cdn.jsdelivr.net/npm/@xenova/[email protected]/dist/transformers.js and found specific mentions of Phi

/* harmony export */   "PhiForCausalLM": () => (/* binding */ PhiForCausalLM),
/* harmony export */   "PhiModel": () => (/* binding */ PhiModel),
/* harmony export */   "PhiPreTrainedModel": () => (/* binding */ PhiPreTrainedModel),

I've been trying to use them with a vanilla script import from the CDN but I keep hitting notFound or "model not supported" errors

but I don't know if;

1. My code is just bad
2. They're not supported on the CDN yet

Any pointers would be amazing!

My current code is

import { pipeline } from 'https://cdn.jsdelivr.net/npm/@xenova/transformers@latest';

async function runInference() {
  const inputText = document.getElementById('input-text').value;
  const outputElement = document.getElementById('output');

  // Clear previous output
  outputElement.textContent = 'Loading model and running inference...';

  try {
    // Use the pipeline function to load the Phi model
    const generator = await pipeline('text-generation', '{model location}');

I've tried the following for the model location value (verbatim), occassionally I've had 404 not found responses, occassionally not authorised responses but the most promising error I've got seems to be the model not supported one because that suggests the script has an idea of what I'm asking for

• onnx-community/Phi-3.5-mini-instruct-onnx-web
• Xenova/Phi-3-mini-4k-instruct
• microsoft/Phi-3-mini-4k-instruct-onnx-web
• Xenova/tiny-random-PhiForCausalLM
• Xenova/phi-1_5_dev
• BricksDisplay/phi-1_5
• BricksDisplay/phi-1_5-q4
• BricksDisplay/phi-1_5-bnb4
• Xenova/Phi-3-mini-4k-instruct_fp16
• Xenova/tiny-random-LlavaForConditionalGeneration_phi

Thanks very much in advance for your help!