DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-08-10 06:07:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: fallback when backend rejects flash attention (#1732)

Browse Source
This commit is contained in:
leejet 2026-07-03 00:46:51 +08:00 committed by GitHub
parent 7dab3668d0
commit 2574f59365
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
src/core/ggml_extend.hpp
View File

@ -1388,6 +1388,9 @@ __STATIC_INLINE__ ggml_tensor* ggml_ext_attention_ext(ggml_context* ctx,
} }
auto out = ggml_flash_attn_ext(ctx, q_in, k_in, v_in, mask_in, scale / kv_scale, 0, 0); auto out = ggml_flash_attn_ext(ctx, q_in, k_in, v_in, mask_in, scale / kv_scale, 0, 0);
if (!ggml_backend_supports_op(backend, out)) {
return nullptr;
}
ggml_flash_attn_ext_set_prec(out, GGML_PREC_F32); ggml_flash_attn_ext_set_prec(out, GGML_PREC_F32);
if (kv_scale != 1.0f) { if (kv_scale != 1.0f) {
out = ggml_ext_scale(ctx, out, 1.0f / kv_scale); out = ggml_ext_scale(ctx, out, 1.0f / kv_scale);
@ -1405,9 +1408,7 @@ __STATIC_INLINE__ ggml_tensor* ggml_ext_attention_ext(ggml_context* ctx,
if (can_use_flash_attn) { if (can_use_flash_attn) {
kqv = build_kqv(q, k, v, mask); kqv = build_kqv(q, k, v, mask);
if (!ggml_backend_supports_op(backend, kqv)) { if (kqv != nullptr) {
kqv = nullptr;
} else {
kqv = ggml_view_4d(ctx, kqv = ggml_view_4d(ctx,
kqv, kqv,
d_head, d_head,