upstream-archive/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-08-06 18:20:42 -05:00
Code Issues Packages Projects Releases Wiki Activity

feat: drive layer split from graph-cut segments (#1762)

Browse Source
This commit is contained in:
leejet
2026-07-07 23:16:52 +08:00
committed by GitHub
parent bb84971129
commit 9ef6e7398f
8 changed files with 618 additions and 213 deletions
Download Patch File Download Diff File Expand all files Collapse all files

126
src/conditioning/conditioner.hpp
View File

@@ -118,6 +118,8 @@ public:
virtual void set_max_graph_vram_bytes(size_t max_vram_bytes) {}
virtual void set_stream_layers_enabled(bool enabled) {}
virtual void set_runtime_backends(const std::vector<ggml_backend_t>& backends) {}
virtual void set_graph_cut_layer_split_enabled(bool enabled) {}
virtual void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) {}
virtual void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) {}
virtual void set_flash_attention_enabled(bool enabled) = 0;
virtual void set_weight_adapter(const std::shared_ptr<WeightAdapter>& adapter) {}
@@ -181,6 +183,27 @@ struct FrozenCLIPEmbedderWithCustomWords : public Conditioner {
}
}
void set_runtime_backends(const std::vector<ggml_backend_t>& backends) override {
text_model->set_runtime_backends(backends);
if (sd_version_is_sdxl(version)) {
text_model2->set_runtime_backends(backends);
}
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
text_model->set_graph_cut_layer_split_enabled(enabled);
if (sd_version_is_sdxl(version)) {
text_model2->set_graph_cut_layer_split_enabled(enabled);
}
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
text_model->set_graph_cut_layer_split_backend_vram_limits(limits);
if (sd_version_is_sdxl(version)) {
text_model2->set_graph_cut_layer_split_backend_vram_limits(limits);
}
}
void set_flash_attention_enabled(bool enabled) override {
text_model->set_flash_attention_enabled(enabled);
if (sd_version_is_sdxl(version)) {
@@ -639,11 +662,41 @@ struct SD3CLIPEmbedder : public Conditioner {
}
void set_runtime_backends(const std::vector<ggml_backend_t>& backends) override {
if (clip_l) {
clip_l->set_runtime_backends(backends);
}
if (clip_g) {
clip_g->set_runtime_backends(backends);
}
if (t5) {
t5->set_runtime_backends(backends);
}
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
if (clip_l) {
clip_l->set_graph_cut_layer_split_enabled(enabled);
}
if (clip_g) {
clip_g->set_graph_cut_layer_split_enabled(enabled);
}
if (t5) {
t5->set_graph_cut_layer_split_enabled(enabled);
}
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
if (clip_l) {
clip_l->set_graph_cut_layer_split_backend_vram_limits(limits);
}
if (clip_g) {
clip_g->set_graph_cut_layer_split_backend_vram_limits(limits);
}
if (t5) {
t5->set_graph_cut_layer_split_backend_vram_limits(limits);
}
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
if (t5) {
t5->get_param_tensors(tensors, "text_encoders.t5xxl.transformer");
@@ -1010,11 +1063,32 @@ struct FluxCLIPEmbedder : public Conditioner {
}
void set_runtime_backends(const std::vector<ggml_backend_t>& backends) override {
if (clip_l) {
clip_l->set_runtime_backends(backends);
}
if (t5) {
t5->set_runtime_backends(backends);
}
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
if (clip_l) {
clip_l->set_graph_cut_layer_split_enabled(enabled);
}
if (t5) {
t5->set_graph_cut_layer_split_enabled(enabled);
}
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
if (clip_l) {
clip_l->set_graph_cut_layer_split_backend_vram_limits(limits);
}
if (t5) {
t5->set_graph_cut_layer_split_backend_vram_limits(limits);
}
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
if (t5) {
t5->get_param_tensors(tensors, "text_encoders.t5xxl.transformer");
@@ -1278,6 +1352,18 @@ struct T5CLIPEmbedder : public Conditioner {
}
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
if (t5) {
t5->set_graph_cut_layer_split_enabled(enabled);
}
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
if (t5) {
t5->set_graph_cut_layer_split_backend_vram_limits(limits);
}
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
if (t5) {
t5->get_param_tensors(tensors, "text_encoders.t5xxl.transformer");
@@ -1482,6 +1568,18 @@ struct MiniT2IConditioner : public Conditioner {
}
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
if (t5) {
t5->set_graph_cut_layer_split_enabled(enabled);
}
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
if (t5) {
t5->set_graph_cut_layer_split_backend_vram_limits(limits);
}
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
if (t5) {
t5->get_param_tensors(tensors, "text_encoders.t5xxl.transformer");
@@ -1576,6 +1674,14 @@ struct AnimaConditioner : public Conditioner {
llm->set_runtime_backends(backends);
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
llm->set_graph_cut_layer_split_enabled(enabled);
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
llm->set_graph_cut_layer_split_backend_vram_limits(limits);
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
llm->get_param_tensors(tensors, "text_encoders.llm");
}
@@ -1729,6 +1835,18 @@ struct LLMEmbedder : public Conditioner {
llm->set_runtime_backends(backends);
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
if (llm) {
llm->set_graph_cut_layer_split_enabled(enabled);
}
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
if (llm) {
llm->set_graph_cut_layer_split_backend_vram_limits(limits);
}
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
llm->get_param_tensors(tensors, "text_encoders.llm");
}
@@ -2406,6 +2524,14 @@ struct LTXAVEmbedder : public Conditioner {
llm->set_runtime_backends(backends);
}
void set_graph_cut_layer_split_enabled(bool enabled) override {
llm->set_graph_cut_layer_split_enabled(enabled);
}
void set_graph_cut_layer_split_backend_vram_limits(const std::vector<size_t>& limits) override {
llm->set_graph_cut_layer_split_backend_vram_limits(limits);
}
void get_layer_split_param_tensors(std::map<std::string, ggml_tensor*>& tensors) override {
llm->get_param_tensors(tensors, "text_encoders.llm");
}