Skip to content

perf(moss): reuse shared modulation slices #2208

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
leejet merged 2 commits into from
Oct 25, 2026
Merged

perf(moss): reuse shared modulation slices #2208

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f7877e7
perf(moss): reuse shared modulation slices
mirek190 Oct 24, 2026
dd88193
docs(moss): remove unnecessary optimization documentation
mirek190 Oct 24, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 12 additions & 4 deletions src/model/diffusion/moss_image.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2,6 +2,7 @@
#define __SD_MODEL_DIFFUSION_MOSS_IMAGE_HPP__

#include <algorithm>
#include <array>
#include <cinttypes>

#include "model/common/ggml_block.hpp"
Expand Down Expand Up @@ -113,11 +114,11 @@ namespace MOSSImage {
blocks["proj.ffn_down"] = std::make_shared<Linear>(config.ffn_dim, config.dim, false);
}

ggml_tensor* forward(GGMLRunnerContext* ctx, ggml_tensor* x, ggml_tensor* pe, ggml_tensor* mod) {
ggml_tensor* forward(GGMLRunnerContext* ctx, ggml_tensor* x, ggml_tensor* pe, const std::array<ggml_tensor*, 6>& mod) {
auto c = ctx->ggml_ctx;
auto linear = [&](const char* name, ggml_tensor* h) { return std::dynamic_pointer_cast<Linear>(blocks[name])->forward(ctx, h); };
auto norm = [&](const char* name, ggml_tensor* h) { return std::dynamic_pointer_cast<RMSNorm>(blocks[name])->forward(ctx, h); };
auto part = [&](int i) { return ggml_ext_slice(c, mod, 0, i * config.dim, (i + 1) * config.dim); };
auto part = [&](int i) { return mod[i]; };
auto y = norm("norm1", x);
y = ggml_add(c, ggml_add(c, y, ggml_mul(c, y, part(1))), part(0));
auto qkv = linear("proj.qkv", y);
Expand Down Expand Up @@ -175,10 +176,17 @@ namespace MOSSImage {
time = linear("time_embedding.2", ggml_silu(c, linear("time_embedding.0", time)));
auto mod = linear("shared_block_modulation", time);
auto head = linear("video_head_modulation", time);
std::array<ggml_tensor*, 6> parts;
for (int i = 0; i < 6; ++i) {
parts[i] = ggml_ext_slice(c, mod, 0, i * config.dim, (i + 1) * config.dim);
if (ctx->graph != nullptr) {
ggml_build_forward_expand(ctx->graph, parts[i]);
}
}
x = ggml_concat(c, x, text, 1);
for (int64_t i = 0; i < config.depth; ++i) {
if (std::find(skip.begin(), skip.end(), i) == skip.end()) {
x = std::dynamic_pointer_cast<Block>(blocks["layers." + std::to_string(i)])->forward(ctx, x, pe, mod);
x = std::dynamic_pointer_cast<Block>(blocks["layers." + std::to_string(i)])->forward(ctx, x, pe, parts);
sd::ggml_graph_cut::mark_graph_cut(x, "moss_image.layers." + std::to_string(i), "x");
}
}
Expand Down Expand Up @@ -250,7 +258,7 @@ namespace MOSSImage {
auto time = ggml_new_tensor_3d(compute_ctx, GGML_TYPE_F32, config.freq_dim, 1, 1);
set_backend_tensor_data(pe, pe_data.data());
set_backend_tensor_data(time, time_data.data());
auto ctx = get_context();
auto ctx = get_context(gf);
ggml_build_forward_expand(gf, model.forward(&ctx, x, time, text, pe, skip));
return gf;
};
Expand Down