Skip to content
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.

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

flake.nix: fix for rocm 5.7 #3853

Merged
merged 1 commit into from
Oct 31, 2023
Merged

Conversation

Tungsten842
Copy link
Contributor

NixOS/nixpkgs#258328 moved all the rocm packages to rocmPackages, this has been fixed.
Also because hipcc targets autodetection doesn't work with nix sandbox, now every rocblas target is built by default, compiling all targets increases build times but guarantees llama.cpp to work with every rocm supported gpu, until now only some random target where built.
This fixes #3320 when using nix.

Copy link
Collaborator

@Green-Sky Green-Sky left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I dont have an amd device to test this with, but the nix changes look alright.

@ggerganov ggerganov merged commit 07178c9 into ggerganov:master Oct 31, 2023
6 checks passed
@Tungsten842 Tungsten842 deleted the rocm-update branch November 1, 2023 02:28
# Build all targets supported by rocBLAS. When updating search for TARGET_LIST_ROCM
# in github.com/ROCmSoftwarePlatform/rocBLAS/blob/develop/CMakeLists.txt
# and select the line that matches the current nixpkgs version of rocBLAS.
"-DAMDGPU_TARGETS=gfx803;gfx900;gfx906:xnack-;gfx908:xnack-;gfx90a:xnack+;gfx90a:xnack-;gfx940;gfx941;gfx942;gfx1010;gfx1012;gfx1030;gfx1100;gfx1101;gfx1102"

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

The nixpkgs rocblas derivation should build everything by default, is this necessary?

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah. I misread. I thought this was rebuilding rocblas for some reason.

olexiyb pushed a commit to Sanctum-AI/llama.cpp that referenced this pull request Nov 23, 2023
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

ROCm error: ggml-cuda.cu:6246: invalid device function
4 participants