happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix: update callback for ffn_moe_weighted and add callback for attn_out in deepseek2 model

This commit is contained in:
bluebread 2025-11-18 06:19:38 +00:00
parent 1e08157134
commit 6c0715befc
2 changed files with 2 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-graph.cpp
View File

@ -1106,7 +1106,7 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
if (!weight_before_ffn) { if (!weight_before_ffn) {
experts = ggml_mul(ctx0, experts, weights); experts = ggml_mul(ctx0, experts, weights);
cb(cur, "ffn_moe_weighted", il); cb(experts, "ffn_moe_weighted", il);
} }
ggml_tensor * cur_experts[LLAMA_MAX_EXPERTS] = { nullptr }; ggml_tensor * cur_experts[LLAMA_MAX_EXPERTS] = { nullptr };

1
src/models/deepseek2.cpp
View File

@ -74,6 +74,7 @@ llm_build_deepseek2::llm_build_deepseek2(const llama_model & model, const llm_gr
cur = build_attn(inp_attn, cur = build_attn(inp_attn,
model.layers[il].wo, NULL, model.layers[il].wo, NULL,
Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il); Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, kq_scale, il);
cb(cur, "attn_out", il);
} }
else { else {
ggml_tensor * q = NULL; ggml_tensor * q = NULL;