Questions on Qualcomm chip bandwidth occupation

[zhaoxuejun1234]

Hi, Molly.
I'm writing to discuss some thoughts on rwkv. Recently, we are using some models from tier one developed based on Transformers. The high bandwith occupation is a headache issue. I haven't tried your project on device chip. But I'm wondering that whether RWKV outperforms those Transformers based models on bandwith cooupation owing to no kv-cache.
I will look into more details on kv-cache and on device test of your project. I would be grateful if you could offer some hints on this issue.