You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[Feature] Rename forked ggml library or make the related f8/INT8 features in sd.cpp as a optional #1917
Rename forked ggml for sd.cpp, or make f8/INT8 related feature as a optional.
Detailed Description
Requested feature
Using a forked version of GGML under a different name,
Or when using with the upstream ggml, SD.cpp will be disable f8/int8-related features automatically.
Reason / background
GGML is now a public library.
We need to avoid conflicts with existing upstream libraries in the system.
If the forked version of GGML has the same binary name as the upstream, there is a risk of conflict with llama.cpp, wisper.cpp, or other GGML components bundled with applications. On Windows, in particular, this can easily lead to "DLL Hell."
While forcing a static build would be safe, I think it's an impractical solution.
Often, we must use the rapidly updated upstream GGML to fix bugs on minor devices.
From what I've seen, it doesn't appear that the pull request submitted to the upstream GGML for f8/INT8 support in sd.cpp will be approved anytime soon.
Alternatives you considered
On my cloneed worktree, currently, such f8/INT8 features are all disabled by adding #if - #endif for using with upstream GGML.
I think the better alternative would be for the llama.cpp maintainers to merge ggml-org/llama.cpp#26914 (and then a follow up PR implementing this change: leejet/ggml@032b699) so that sd.cpp could use mainline ggml again.
changed the title [-][Feature] Rename ggml library or make the related f8/INT8 features in sd.cpp as a optional[/-][+][Feature] Rename forked ggml library or make the related f8/INT8 features in sd.cpp as a optional[/+]on Aug 28, 2026
Is it something new similar? if (w->type == GGML_TYPE_F8_E4M3 || w->type == GGML_TYPE_F8_E5M2) {
I patched llama.cpp for INT8 but now I'm getting another error...
#1999 introduces SD_USE_UPSTREAM_GGML. Enabling this option allows the project to compile successfully with ggml-org/ggml, although some features, such as int8 conrot, are not supported.
Activity
stduhpf commentedon Aug 27, 2026
I think the better alternative would be for the llama.cpp maintainers to merge ggml-org/llama.cpp#26914 (and then a follow up PR implementing this change: leejet/ggml@032b699) so that sd.cpp could use mainline ggml again.
[-][Feature] Rename ggml library or make the related f8/INT8 features in sd.cpp as a optional[/-][+][Feature] Rename forked ggml library or make the related f8/INT8 features in sd.cpp as a optional[/+]iwr-redmond commentedon Aug 29, 2026
There is also the matter of the downstream PR that has yet to be reviewed.
alexzk1 commentedon Aug 31, 2026
Is it something new similar?
if (w->type == GGML_TYPE_F8_E4M3 || w->type == GGML_TYPE_F8_E5M2) {I patched llama.cpp for INT8 but now I'm getting another error...
leejet commentedon Sep 19, 2026
#1999 introduces
SD_USE_UPSTREAM_GGML. Enabling this option allows the project to compile successfully withggml-org/ggml, although some features, such asint8 conrot, are not supported.