Skip to content

graph : avoid huge warm-up graphs for MoE models #14753

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
merged 2 commits into from
Jul 18, 2025
Merged

graph : avoid huge warm-up graphs for MoE models #14753

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
033b306
graph : avoid huge warm-up graphs for MoE models
ggerganov Jul 18, 2025
5883f01
cont : bump max nodes to 8x model tensors
ggerganov Jul 18, 2025
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
2 changes: 1 addition & 1 deletion src/llama-context.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1312,7 +1312,7 @@ uint32_t llama_context::output_reserve(int32_t n_outputs) {
//

uint32_t llama_context::graph_max_nodes() const {
return std::max<uint32_t>(65536u, 5u*model.n_tensors());
return std::max<uint32_t>(1024u, 6u*model.n_tensors());
Copy link
Member Author

@ggerganov ggerganov Jul 18, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We should probably bump this up to 8u*model.n_tensors() just to be safe.

}

llm_graph_result * llama_context::get_gf_res_reserve() const {
Expand Down
7 changes: 5 additions & 2 deletions src/llama-graph.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -906,8 +906,11 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
}

// aggregate experts
// note: here we explicitly use hparams.n_expert_used instead of n_expert_used
// to avoid potentially a large number of add nodes during warmup
// ref: https://github.com/ggml-org/llama.cpp/pull/14753
ggml_tensor * moe_out = nullptr;
for (int i = 0; i < n_expert_used; ++i) {
for (uint32_t i = 0; i < hparams.n_expert_used; ++i) {
ggml_tensor * cur_expert = ggml_view_2d(ctx0, experts, n_embd, n_tokens,
experts->nb[2], i*experts->nb[1]);

Expand All @@ -918,7 +921,7 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
}
}

if (n_expert_used == 1) {
if (hparams.n_expert_used == 1) {
// avoid returning a non-contiguous tensor
moe_out = ggml_cont(ctx0, moe_out);
}
Expand Down
Loading