ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-03-10 14:00:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add GitHub data: filename sanitization (#640)

Browse Source
This commit is contained in:
Thomas
2025-07-23 13:31:53 +02:00
committed by GitHub
parent 3600d82e98
commit eaa2510a28
626 changed files with 0 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
github-data/pull_requests/107 - Faster IQ1_BN Metal implementation.md Normal file
View File

@@ -0,0 +1,18 @@
### 🔀 [#107](https://github.com/ikawrakow/ik_llama.cpp/pull/107) - Faster IQ1_BN Metal implementation
| **Author** | `ikawrakow` |
| :--- | :--- |
| **State** | ❌ **Closed** |
| **Created** | 2024-10-26 |
| **Updated** | 2024-10-26 |
---
#### Description
On my 30-core M2-Max TG-128 for Bitnet-1.58b-3.3B improves from 82 t/s to 94.7 t/s.
PP-512 goes from 686 t/s to 702 t/s.
Integer multiplications are expensive, so the trick used is to replace them with shifts and additions.
There is also a minor `IQ2_BN` PP-512 improvement (710 -> 714 t/s).