Skip to content

losing, not loosing #2

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Sep 21, 2025
Merged

losing, not loosing #2

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,7 +2,7 @@

**Specialized llama.cpp fork with GFX906 flash attention optimizations for D=128 head dimension models ONLY!**

This fork is specifically optimized for AMD GFX906 architecture (MI50, MI60, Vega VII) and targets models with **head dimension D=128** (such as Qwen3-30B series). The aim of this fork is to be able to run a QWEN30B session with 32K ctx on a single card without loosing too much speed. For this reason the fork won't work with smaller models (you can check the huggingface model sheet for key and value lengths).
This fork is specifically optimized for AMD GFX906 architecture (MI50, MI60, Vega VII) and targets models with **head dimension D=128** (such as Qwen3-30B series). The aim of this fork is to be able to run a QWEN30B session with 32K ctx on a single card without losing too much speed. For this reason the fork won't work with smaller models (you can check the huggingface model sheet for key and value lengths).

---

Expand Down