Skip to content

fix(cuda): allow f16/bf16 + q8_0 KV without GGML_CUDA_FA_ALL_QUANTS #82

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
TheTom merged 1 commit into from
Apr 17, 2026
Merged

fix(cuda): allow f16/bf16 + q8_0 KV without GGML_CUDA_FA_ALL_QUANTS #82

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
11 changes: 7 additions & 4 deletions ggml/src/ggml-cuda/fattn.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -419,11 +419,14 @@ static best_fattn_kernel ggml_cuda_get_best_fattn_kernel(const int device, const

#ifndef GGML_CUDA_FA_ALL_QUANTS
if (K->type != V->type) {
// Allow mixed turbo KV types (any combination of turbo2, turbo3, q8_0)
auto is_turbo = [](ggml_type t) {
return t == GGML_TYPE_TURBO2_0 || t == GGML_TYPE_TURBO3_0 || t == GGML_TYPE_TURBO4_0 || t == GGML_TYPE_Q8_0;
// Allow mixed KV types for combinations that have FA template instances compiled in:
// - turbo2/3/4 + q8_0 (turbo cache work)
// - f16/bf16 + q8_0 (common K=f16, V=q8_0 setup)
auto is_kv_compat = [](ggml_type t) {
return t == GGML_TYPE_TURBO2_0 || t == GGML_TYPE_TURBO3_0 || t == GGML_TYPE_TURBO4_0
|| t == GGML_TYPE_Q8_0 || t == GGML_TYPE_F16 || t == GGML_TYPE_BF16;
};
if (!is_turbo(K->type) || !is_turbo(V->type)) {
if (!is_kv_compat(K->type) || !is_kv_compat(V->type)) {
return BEST_FATTN_KERNEL_NONE;
}
}
Expand Down
Loading