inference on google coral usb accelerator =

[co-manifold]

Hi, I noticed on this thread, that there was a request to be able to make a tflite model that could be run on the google coral usb accelerator device. I believe there is an issue with the model that does not currently make this possible, namely the presence of flexops in the whisper tflite implementation. Is it possible that the tflite model could be changed to allow for inference on a google coral usb accelerator device? Many thanks.

[jayyang-zigbang]

actually the whisper tflite model works well on the google coral, just edge TPU runtime is still not available. if you are interested in only CPU, please just upgrade your python to 3.9.