OrtRun failed with WebGPU buffer error on Apple M1 Max

[bdim404]

Description

After loading the model, any inference attempt fails immediately with:

Something went wrong: failed to call OrtRun(). ERROR_CODE: 1, ERROR_MESSAGE:
/mnt/vss/_work/1/s/onnxruntime/core/providers/webgpu/buffer_manager.cc:543
status == wgpu::MapAsyncStatus::Success was false.
Failed to download data from buffer: Failed to execute 'mapAsync' on 'GPUBuffer':
[Invalid Buffer (unlabeled)] is invalid due to a previous error.

Environment

• OS: macOS (Darwin 25.4.0)
• GPU: Apple M1 Max (Metal 4)
• Browser: Chrome (WebGPU enabled)
• Model: gemma-4-E2B (default)

[klauss194]

I get that too on the AMD laptop 880/890 Radeon -> been trying to fix it

[kessler]

it is probably a context length issue.
if it's not private and you can, provide the page and the prompt i will try to recreate.