DarthAffe/stable-diffusion.cpp
1
0 Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-11-12 03:49:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: add configurable Qwen cache types and early cache scheduling (#2045)

Browse Source
This commit is contained in:
leejet authored and GitHub committed 2026-09-25 01:51:01 +08:00
1 parent 4dfe8f5d45
commit 740c7ae193
11 files changed
+185 -61
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/model/vae/ltx_vae.hpp
+2 -3
View File
@@ -1300,7 +1300,7 @@ struct LTXVideoVAE : public VAE {
feat_map[feat_idx] = get_cache_tensor_by_name(temporal_feat_cache_name(feat_idx));
}
auto runner_ctx = get_context();
auto runner_ctx = get_context(gf);
int feat_count = 0;
ggml_tensor* out = vae.decode_tiled_chunk(&runner_ctx,
z,
@@ -1313,8 +1313,7 @@ struct LTXVideoVAE : public VAE {
for (int feat_idx = 0; feat_idx < feat_count && feat_idx < static_cast<int>(feat_map.size()); ++feat_idx) {
ggml_tensor* feat_cache = feat_map[static_cast<size_t>(feat_idx)];
if (feat_cache != nullptr) {
cache(temporal_feat_cache_name(static_cast<size_t>(feat_idx)), feat_cache);
ggml_build_forward_expand(gf, feat_cache);
runner_ctx.persist_cache_tensor(temporal_feat_cache_name(static_cast<size_t>(feat_idx)), feat_cache);
}
}
src/model/vae/wan_vae.hpp
+2 -3
View File
@@ -1443,15 +1443,14 @@ namespace WAN {
ggml_tensor* z = make_input(z_tensor);
auto runner_ctx = get_context();
auto runner_ctx = get_context(gf);
ggml_tensor* out = ae.decode_tiled_chunk(&runner_ctx, z, chunk_idx);
for (size_t feat_idx = 0; feat_idx < ae._feat_map.size(); feat_idx++) {
ggml_tensor* feat_cache = ae._feat_map[feat_idx];
if (feat_cache != nullptr) {
cache("feat_idx:" + std::to_string(feat_idx), feat_cache);
ggml_build_forward_expand(gf, feat_cache);
runner_ctx.persist_cache_tensor("feat_idx:" + std::to_string(feat_idx), feat_cache);
}
}