mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-03-12 15:00:11 +00:00
19 lines
489 B
Markdown
19 lines
489 B
Markdown
### 🔀 [#70](https://github.com/ikawrakow/ik_llama.cpp/pull/70) - Fused unary(x)*y
|
|
|
|
| **Author** | `ikawrakow` |
|
|
| :--- | :--- |
|
|
| **State** | ❌ **Closed** |
|
|
| **Created** | 2024-09-30 |
|
|
| **Updated** | 2024-10-02 |
|
|
|
|
---
|
|
|
|
#### Description
|
|
|
|
This is useful for parallel FFNs. `unary` can be `silu, gelu` or `relu`.
|
|
|
|
Implemented for CPU, CUDA and Metal.
|
|
|
|
Speedup is disappointingly small (1-3% for PP, depending on platform and model).
|
|
|
|
Let me think some more if I want to merge it. |