mirror of
https://github.com/ikawrakow/ik_llama.cpp.git
synced 2026-03-05 11:30:09 +00:00
19 lines
540 B
Markdown
19 lines
540 B
Markdown
### 🔀 [#630](https://github.com/ikawrakow/ik_llama.cpp/pull/630) - GEMM for IQ1_M
|
|
|
|
| **Author** | `ikawrakow` |
|
|
| :--- | :--- |
|
|
| **State** | ❌ **Closed** |
|
|
| **Created** | 2025-07-18 |
|
|
| **Updated** | 2025-07-18 |
|
|
|
|
---
|
|
|
|
#### Description
|
|
|
|
Closes #626
|
|
|
|
Hopefully the collective knowledge on Reddit and elsewhere that one cannot use `-fmoe` because of the missing `IQ1_M` GEMM has not already been perpetuated for all eternity...
|
|
|
|
After this PR, you can use `-fmoe` for any model.
|
|
|
|
Oh, no `ARM_NEON` for now, this will come later. |