Fusing a 27B ternary LLM's whole decode step into one CUDA kernel
1 points
1 hour ago
| 0 comments
| twitter.com
| HN
No one has commented on this post.