Hacker News
new
|
past
|
comments
|
ask
|
show
|
jobs
|
submit
login
reasonableklout
1 day ago
|
parent
|
context
|
favorite
| on:
Fast LLM Inference From Scratch (using CUDA)
Oops, you're right and it's a difference between my blog post and source code. It should be __shfl_down_sync as seen [here](
https://github.com/andrewkchan/yalm/blob/8c908f23f5d8cc3f14c...
)
Guidelines
|
FAQ
|
Lists
|
API
|
Security
|
Legal
|
Apply to YC
|
Contact
Search: