[Web] WebGPU batch inference only the first set of data is correct

[sunpeng222]

Describe the issue

I encountered a problem when using WebGPU. When I use batch inference, the results of the second set of data are completely incorrect. The model is not the problem, as my inference using wasm was completely fine.What is the cause of this?My model loading and inference are all placed in the webworker

To reproduce

importScripts("./ort/ort.webgpu.min.js")
let session
init()
async function init() {
    ort.env.wasm.simd = true;
    ort.env.wasm.wasmPaths = "./ort/"
    session = await ort.InferenceSession.create('./end2end_normalized_N.onnx', {
        executionProviders: [{
            name: "webgpu"
        }],
        logSeverityLevel: 3,
        logVerbosityLevel: 3,
    });
    self.postMessage({
        type: "ready",
        data: null
    })
}
self.addEventListener('message', async e => {
    if (e.data.type === "upload") {
        if (!session) {
            self.postMessage({
                type: "noSession",
                data: null
            })
            return
        }
        const tensorA = new ort.Tensor('float32', e.data.data, [2, 224, 224, 3]);
        const feeds = { input_1: tensorA };
        const start = performance.now();
        const results = await session.run(feeds);
        const end = performance.now();

        self.postMessage({
            type: "end",
            data: {
                time: end - start,
                results: results
            }
        })
    }
});

Urgency

No response

ONNX Runtime Installation

Released Package

ONNX Runtime Version or Commit ID

1.16.3

Execution Provider

'webgpu' (WebGPU)

[gyagp]

Can you give a try on dev release ("https://cdn.jsdelivr.net/npm/onnxruntime-web@dev/dist/ort.webgpu.min.js")?
If the issue still persists, can you share the model?

[sunpeng222]

Can you give a try on dev release ("https://cdn.jsdelivr.net/npm/onnxruntime-web@dev/dist/ort.webgpu.min.js")? If the issue still persists, can you share the model?

Thank you, the development version has no problem.

[gyagp]

Thanks for the confirmation! Please be patient for a new release;) FYI, @fs-eire @guschmue

[MountainAndMorning]

Is there any timeline for version 1.17.0?

[github-actions]

This issue has been automatically marked as stale due to inactivity and will be closed in 30 days if no further activity occurs. If further support is needed, please provide an update and/or more details.

[github-actions]

This issue has been automatically closed due to inactivity. Please reactivate if further support is needed.