Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
verytrivial
15 hours ago
|
parent
|
context
|
favorite
| on:
FlashAttention-T: Towards Tensorized Attention
Tldr: 5% - 17% speedup due to removing a bottleneck by juggling where on a GPU/compute core a computation is done during Flash attention.
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search:
