ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-04-21 23:19:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

We need to of course pass the merged tensor to build_ffn

Browse Source
This commit is contained in:
Iwan Kawrakow
2026-01-12 16:05:14 +00:00
parent aad40bcd2d
commit ab1ec19151
1 changed files with 2 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llama-build-context.cpp
View File

@@ -6936,7 +6936,7 @@ ggml_cgraph * llm_build_context::build_glm4_moe() {
n_expert, n_expert_used,
LLM_FFN_SILU, hparams.expert_weights_norm, true, hparams.expert_weights_scale,
(llm_expert_gating_func_type) hparams.expert_gating_func,
LLM_FFN_SILU, cb, il, gf, true);
LLM_FFN_SILU, cb, il, gf, true, model.layers[il].ffn_up_gate_exps);
}
// residual and context vector
@@ -8669,7 +8669,7 @@ ggml_cgraph* llm_build_context::build_minimaxm2() {
LLM_FFN_SILU, true,
false, 0,
(llm_expert_gating_func_type)hparams.expert_gating_func,
cb, il, gf);
cb, il, gf, false, model.layers[il].ffn_up_gate_exps);
cb(cur, "ffn_moe_out", il);
cur = ggml_add(ctx0, cur, ffn_inp);