Y
Hacker News
new
|
ask
|
show
|
jobs
by
alex43578
128 days ago
NVIDIA is showing training at 4 bits (NVPF4), and 4 bit quants have been standard for running LLMs at home for quite a while because performance was good enough.