DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-11 03:24:30 +00:00
Code Issues Packages Projects Releases Wiki Activity

perf: use ggml rope apply op on supported backends (#2113)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-10-09 04:32:36 +08:00
1 parent e16d26a96a
commit 228c707fde
7 files changed
+36 -21
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/model/diffusion/qwen_image_2_1.hpp
+2 -2
View File
@@ -188,8 +188,8 @@ namespace Qwen {
auto v = project("to_v");
q = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_q"])->forward(ctx, q);
k = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_k"])->forward(ctx, k);
q = Rope::apply_rope(ctx->ggml_ctx, q, pe);
k = Rope::apply_rope(ctx->ggml_ctx, k, pe);
q = Rope::apply_rope(ctx, q, pe);
k = Rope::apply_rope(ctx, k, pe);
if (cache.mode == QwenImage21PrefixCache::Mode::STORE) {
// Preserve query-first attention evaluation while writing each layer's
// prefix before its full-sequence K/V can accumulate across layers.