vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-13 07:09:03 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

Merge branch 'upstream' into concedo_experimental

Browse source 
# Conflicts:
#	.github/workflows/build.yml
#	CODEOWNERS
#	ggml/CMakeLists.txt
#	ggml/src/ggml-cpu/CMakeLists.txt
#	ggml/src/ggml-cpu/kleidiai/kleidiai.cpp
#	scripts/sync-ggml.last
#	tests/test-backend-ops.cpp
This commit is contained in:
Concedo 2025-09-30 22:28:53 +08:00
commit 20c802a198
11 changed files with 455 additions and 241 deletions
Download patch file Download diff file Expand all files Collapse all files

4
src/llama-model.cpp
View file

@ -11813,6 +11813,7 @@ struct llm_graph_context_mamba : public llm_graph_context {
// TODO: skip computing output earlier for unused tokens
y = ggml_add(ctx0, y, ggml_mul(ctx0, x, model.layers[il].ssm_d));
cb(y, "mamba2_y_add_d", il);
y = ggml_swiglu_split(ctx0, ggml_cont(ctx0, z), y);
// grouped RMS norm
@ -14767,6 +14768,7 @@ struct llm_build_nemotron_h : public llm_graph_context_mamba {
ggml_tensor * inpL;
inpL = build_inp_embd(model.tok_embd);
ggml_build_forward_expand(gf, inpL);
auto * inp = build_inp_mem_hybrid();
@ -14798,7 +14800,7 @@ struct llm_build_nemotron_h : public llm_graph_context_mamba {
// add residual
cur = ggml_add(ctx0, cur, inpSA);
cb(cur, "block_out", il);
cb(cur, "nemotron_h_block_out", il);
// input for next layer
inpL = cur;