Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-20 01:31:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge branch 'upstream' into concedo_experimental

Browse Source 
# Conflicts:
#	CODEOWNERS
#	docs/build.md
#	scripts/sync-ggml.last
#	tests/test-backend-ops.cpp
#	tools/imatrix/README.md
#	tools/imatrix/imatrix.cpp
This commit is contained in:
Concedo
2025-07-20 22:47:31 +08:00
parent 15b1034dd6 938b785764
commit 30675b0798
21 changed files with 1397 additions and 1181 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-graph.cpp
+16 -9
View File
@@ -428,6 +428,8 @@ void llm_graph_result::reset() {
t_embd = nullptr;
t_embd_pooled = nullptr;
params = {};
inputs.clear();
buf_compute_meta.resize(ggml_tensor_overhead()*max_nodes + ggml_graph_overhead_custom(max_nodes, false));
@@ -905,20 +907,25 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
cb(cur, "ffn_moe_weighted", il);
}
ggml_tensor * cur_experts[LLAMA_MAX_EXPERTS] = { nullptr };
assert(n_expert_used > 0);
// order the views before the adds
for (uint32_t i = 0; i < hparams.n_expert_used; ++i) {
cur_experts[i] = ggml_view_2d(ctx0, experts, n_embd, n_tokens, experts->nb[2], i*experts->nb[1]);
ggml_build_forward_expand(gf, cur_experts[i]);
}
// 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 (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]);
ggml_tensor * moe_out = cur_experts[0];
if (i == 0) {
moe_out = cur_expert;
} else {
moe_out = ggml_add(ctx0, moe_out, cur_expert);
}
for (uint32_t i = 1; i < hparams.n_expert_used; ++i) {
moe_out = ggml_add(ctx0, moe_out, cur_experts[i]);
}
if (hparams.n_expert_used == 1) {