Now General LLM is running on a canister!πŸš€

At Kinic we are focused on cutting edge ZKP, IC, and AI blends that will change the way we search for information, and also how we interact and transact online.

One recent result are various AI models being runnable on the IC.

We could run phi1.5 on a canister!
This achievement is the one of first case of running LLM on blockchain in the world.
It also has the promise of usability as it is done in Rust as a framework (multiple models).

Microsoft phi1.5 is 1.3B params and the size is 800 MiB.

Unfortunately, our canister does not support multi-user, I cannot publish the canister address, but it is already running on the public network.

With IC GPUs and further optimization, token generation speed will be much faster!


Nice work. This could be huge.
Is there work being done right now on IC GPUs?


We need GPU subnets, even if it’s just 1.