Speech-to-Text/Whisper NPU issue

[Narifumi]

just generated mode in line with docuementation:

optimum-cli export openvino --model openai/whisper-base whisper-base

It works as expected on CPU and GPU, but it doesn't work on NPU LNL, generates error below:

Traceback (most recent call last):
File "C:\Iwamoto\AVID_AI\whisper_openvino\Iwamoto_whisper_openvino_simple.py", line 26, in
model = openvino_genai.WhisperPipeline("whisper-base", device)
RuntimeError: Check '!self_attn_nodes.empty()' failed at C:\Jenkins\workspace\private-ci\ie\build-windows-vs2022\b\repos\openvino.genai\src\cpp\src\whisper\pipeline_static.cpp:427

LNL 268V
NPU driver version: 32.0.100.4240 (8/29/2025)
component versions:
openvino 2025.3.0 19807
openvino-genai 2025.3.0.0
openvino-telemetry 2025.2.0
openvino-tokenizers 2025.3.0.0
optimum 1.27.0
optimum-intel 1.25.2

[as-suvorov]

Try to use --disable-stateful for export for NPU.

[Narifumi]

even with --disable-statefull for export, still have same error. it doesn't run on LNL NPU....

[as-suvorov]

There is an issue with 4.53.0 transformers, what version do you use?
Could you try to use transformers=4.52 and optimum-intel=1.25.2

[Narifumi]

Ah Ok. This is very helpful information. Thank you. Let me try it.