Training Llama 2 model

Thank you for all the work.
I have not yet use Graphcore IPU and am considering its use for model training. 

I am wondering if it is possible to perform continued pre-training or fine-tuning of Llama 2 within this platform. 
Also, I am curious if the resulting weights can be reused on an NVIDIA platform.