mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-05-01 11:51:53 +00:00
17 lines
445 B
Markdown
17 lines
445 B
Markdown
### 🔀 [#108](https://github.com/ikawrakow/ik_llama.cpp/pull/108) - Another Bitnet performance improvement on Metal
|
|
|
|
| **Author** | `ikawrakow` |
|
|
| :--- | :--- |
|
|
| **State** | ❌ **Closed** |
|
|
| **Created** | 2024-10-26 |
|
|
| **Updated** | 2024-10-26 |
|
|
|
|
---
|
|
|
|
#### Description
|
|
|
|
This time just the dequantize function.
|
|
|
|
For Bitnet-1.58b-3B on 30-core M2-Max GPU
|
|
* `IQ1_BN` goes from 702 t/s to 716 t/s
|
|
* `IQ2_BN` goes from 714 t/s to 743 t/s |