Practice
Learn
← Quit
0 / 0 correct
LLMs & AI
Hard
1 of 10
Why does running a model locally with llama.cpp on a laptop CPU work at all for 7B models?
A
The model is retrained for CPUs
B
Quantized weights (~4 GB) fit in RAM, and optimized CPU kernels make memory-bound decoding fast enough for single users
C
It streams weights from the internet
D
CPUs are faster than GPUs for matrix math