NLP Applications

[And_Zammit 0]

I am planning to use tinker board for NLP applications (training of BERT models, etc) instead of GPU cards. Do you think that this would work?

Furthermore, is it possible to connect multiple tinker boards together and have them work in parallel? 

Thanks!!!

 

 

[tooz 52]

hello @And_Zammit

if you're planning to train BERT models using Tinker Edge  R, you will need to first convert BERT models to RKNN models(otherwise the NPU/AI accelerator will not work).

here's a document for RKNN conversion https://github.com/rockchip-linux/rknn-toolkit/blob/master/doc/Rockchip_User_Guide_RKNN_Toolkit_V1.7.0_EN.pdf

 

On 8/14/2021 at 6:36 AM, And_Zammit said:

Furthermore, is it possible to connect multiple tinker boards together and have them work in parallel? 

since each Tinker Edge R is an individual processing unit, you'll need an app to control from the host and switch between the boards.