Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
sifar
on Jan 29, 2025
|
parent
|
context
|
favorite
| on:
DeepSeek's multi-head latent attention and other K...
It is more efficient to compute k = xW with the weights transposed than k = Wx.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
