TypeError: Float16Array issue during generation on AMD RX 9060 XT

#1
by Amiee7 - opened

I am encountering a crash during the generation phase. The model seems to load into the VRAM (usage spikes to ~3.5GB), but as soon as the inference starts, the worker throws a TypeError and the VRAM is cleared. I've tried refreshing and checking browser settings, but the issue persists.

Error Log:
Uncaught (in promise) TypeError: A float16 tensor's data must be type of function Float16Array() { [native code] }
at new U (worker-BKr81zhZ.js:8:18990)
at A.addPastKeyValues (worker-BKr81zhZ.js:2763:36035)

System Info:

GPU: AMD Radeon RX 9060 XT (16GB)

CPU: Intel Core i7-11700 (11th Gen)

RAM: 32GB DDR4

Browser: Chrome (latest)

AMD Software Version: 26.5.1

WebGPU Status: Hardware accelerated (verified via chrome://gpu)

It seems like the ONNX runtime or the worker script has trouble handling Float16 tensors on this specific driver/architecture. Let me know if you need any further logs or testing!

Sign up or log in to comment