Every milk-v SoC advertises this or that number of TOPS. I am totally new to ai, got running ollama or something like that some time ago on my PC (using bare cpu). Time moved forward and i bought myself Jupiter. How do i check/ ensure that this ollama will use M1’s npu?
There’s no device exposed on any known bus (like coral tpu would do) yet something must be working under the hood as i ran Bianbu’s object detection desktop shortcut just fine on my Musebook (technically it’s K1 but for me it’s just Jupiter in disguise :D)
Or maybe there’s something propietary needed like for this super-duper Imagination GPU which doesn’t work besides default Wayland setup?
So i understand that there have to be proper k1-specific “chat” interface or whatever to utilize these instructions? In meantime i’ve got myself familiar with Spacemit-provided infer-chatglm and after obtaining llama model…Well experience wasn’t actually pleasing.
Currently i am peeking at other spacemit’s resources at spacemit-ai, however marketing this SoC as “Superior Processing Performance” is as relevant as calling Raspberry PI5 a “new generation of hammer”. Surely with enough stubbornness one can successfully drive a nail with one but it wasn’t exactly designed for that purpose.