CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

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

Browse Source 
* graph : avoid huge warm-up graphs for MoE models

ggml-ci

* cont : bump max nodes to 8x model tensors
This commit is contained in:
Georgi Gerganov
2025-07-18 14:31:15 +03:00
committed by GitHub
parent eacdeb5bfc
commit d498af3d5a
2 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-context.cpp
View File

@@ -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, 8u*model.n_tensors());
}
llm_graph_result * llama_context::get_gf_res_reserve() const {

7
src/llama-graph.cpp
View File

@@ -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]);
@@ -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);
}