Hacker News
new
past
comments
ask
show
jobs
points
by
kroaton
18 hours ago
|
comments
by
2001zhaozhao
11 hours ago
|
[-]
I wonder whether it is much more cost-effective in terms of token throughput / hardware+power cost to get actual GPUs instead, given that the model size is only 27B.
reply