inference_worker.js

importScripts( "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.min.js" )
//importScripts( "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/ort.webgpu.min.js" )

let   id
let   session

const init      = async args =>{
	id = args.id
	
	try {
		ort.env.wasm.numThreads = 0
		ort.env.wasm.wasmPaths  = "https://cdn.jsdelivr.net/npm/onnxruntime-web/dist/"
		session                 = await ort.InferenceSession.create( args.model_array_buffer, { executionProviders : [ "wasm" ]})
		//session                 = await ort.InferenceSession.create( args.array_buffer, { executionProviders : [ "webgpu" ]})
	} catch( error ){
		console.log( error )
	}
	
	self.postMessage ({ selector : "init" })
}
const run_model = async args =>{
	const spec_tensor  = new ort.Tensor( "float32", args.spec, [ 1, 4, 3072, 256 ])
	const outputs      = await session.run({ input : spec_tensor })
	spec_tensor.dispose()
	
	self.postMessage ({ selector : "run_model", id, index : args.index, spec : outputs.output.cpuData })
}

self.onmessage = function( event ){
	switch( event.data.selector ){
		case "init"     : init({ id : event.data.id, model_array_buffer : event.data.model_array_buffer }); break	
		case "run_model": run_model({ index : event.data.index, spec : event.data.spec }); break	
	}
}