Speech-to-text conversion command issue in the doc

[Narifumi]

Documentation has followings
#Download, convert and apply int8 static quantization to whisper-base model
optimum-cli export openvino --model openai/whisper-base --disable-stateful
--quant-mode int8 --dataset librispeech --num-samples 32 whisper-base-int8

But this command doesn't generate output unlike the example above this int8 quantization.

Also, in order to do int8 quatization, do we really need to have "--disable-stateful" option? The option is stated as non optimal performance. should we use "--weight-format int8" in many cases?

[as-suvorov]

There is accuracy issue with quantized stateless whisper model. The actual recommendation is to use --disable-stateful for quantization until issue solved.
cc: @nikita-savelyevv

[nikita-savelyevv]

Hi @Narifumi. Thanks for reporting this! Could you please elaborate a bit more what do you mean by "But this command doesn't generate output unlike the example above this int8 quantization." ?