[npu] 43 weight matrices quantized to int8 and resident on the NPU
[stories15M] dim=288 layers=6 heads=6 | NPU int8 linears | temp=0.85 top-p=0.9 seed=7

--- story ---
The little penguin wanted to fly, so he built a machine out of ice and string. He went to the top of the ground and said, "I'm going to fly!" But the machine was too heavy and he started to wobble. He tried and tried, but he couldn't lift it.
The little penguin felt sad and he started to cry. Suddenly, a friendly bird flew over and said, "I can help you lift the machine!"
The bird took the machine and flew up, up, up. The little penguin was so happy. He could now fly!
He thanked the bird and said, "Thank you so much for helping me!"
The bird smiled and said, "Anytime!".

[speed] 140 tokens, 45.6 tok/s (NPU int8 linears); prompt prefill 337 ms
