DarthAffe/stable-diffusion.cpp
1
0
Fork 0
mirror of https://github.com/leejet/stable-diffusion.cpp.git synced 2026-09-24 20:20:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: add graph cuts for MiniMax-H3 text conditioning (#1900)

Browse Source
This commit is contained in:
assouan 2026-09-23 18:03:29 +02:00 committed by GitHub
parent 2dc7f5408a
commit 36746936c0
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 15 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
src/model/diffusion/minimax_h3.hpp
View File

@ -264,6 +264,9 @@ namespace MiniMaxH3 {
for (int64_t i = 0; i < num_layers; ++i) { for (int64_t i = 0; i < num_layers; ++i) {
auto block = std::dynamic_pointer_cast<TokenRefinerBlock>(blocks["blocks." + std::to_string(i)]); auto block = std::dynamic_pointer_cast<TokenRefinerBlock>(blocks["blocks." + std::to_string(i)]);
x = block->forward(ctx, x); x = block->forward(ctx, x);
sd::ggml_graph_cut::mark_graph_cut(x,
"minimax_h3.token_refiner.blocks." + std::to_string(i),
"hidden_states");
} }
return std::dynamic_pointer_cast<RMSNorm>(blocks["final_norm"])->forward(ctx, x); return std::dynamic_pointer_cast<RMSNorm>(blocks["final_norm"])->forward(ctx, x);
} }
@ -527,7 +530,11 @@ namespace MiniMaxH3 {
GGML_ASSERT(context->ne[0] == config.text_dim); GGML_ASSERT(context->ne[0] == config.text_dim);
auto condition_proj = std::dynamic_pointer_cast<Linear>(blocks["condition_proj"]); auto condition_proj = std::dynamic_pointer_cast<Linear>(blocks["condition_proj"]);
auto token_refiner = std::dynamic_pointer_cast<TokenRefiner>(blocks["token_refiner"]); auto token_refiner = std::dynamic_pointer_cast<TokenRefiner>(blocks["token_refiner"]);
return token_refiner->forward(ctx, condition_proj->forward(ctx, context)); auto projected = condition_proj->forward(ctx, context);
sd::ggml_graph_cut::mark_graph_cut(projected,
"minimax_h3.condition_proj",
"hidden_states");
return token_refiner->forward(ctx, projected);
} }
ggml_tensor* time_embedding(GGMLRunnerContext* ctx, ggml_tensor* time_embedding(GGMLRunnerContext* ctx,

14
src/model/diffusion/qwen_image_2_1.hpp
View File

@ -178,13 +178,13 @@ namespace Qwen {
auto h = std::dynamic_pointer_cast<Linear>(blocks[name])->forward(ctx, x); auto h = std::dynamic_pointer_cast<Linear>(blocks[name])->forward(ctx, x);
return ggml_reshape_4d(ctx->ggml_ctx, h, dim_head, heads, x->ne[1], x->ne[2]); return ggml_reshape_4d(ctx->ggml_ctx, h, dim_head, heads, x->ne[1], x->ne[2]);
}; };
auto q = project("to_q"); auto q = project("to_q");
auto k = project("to_k"); auto k = project("to_k");
auto v = project("to_v"); auto v = project("to_v");
q = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_q"])->forward(ctx, q); q = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_q"])->forward(ctx, q);
k = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_k"])->forward(ctx, k); k = std::dynamic_pointer_cast<RMSNorm>(blocks["norm_k"])->forward(ctx, k);
q = Rope::apply_rope(ctx->ggml_ctx, q, pe); q = Rope::apply_rope(ctx->ggml_ctx, q, pe);
k = Rope::apply_rope(ctx->ggml_ctx, k, pe); k = Rope::apply_rope(ctx->ggml_ctx, k, pe);
if (cache.mode == QwenImage21PrefixCache::Mode::STORE) { if (cache.mode == QwenImage21PrefixCache::Mode::STORE) {
auto persist = [&](ggml_tensor* tensor, int axis, const char* name) { auto persist = [&](ggml_tensor* tensor, int axis, const char* name) {
auto part = ggml_ext_slice(ctx->ggml_ctx, tensor, axis, 0, cache.prefix_length); auto part = ggml_ext_slice(ctx->ggml_ctx, tensor, axis, 0, cache.prefix_length);