sleepy/llama.cpp
1
0
Fork 0
Code Issues 8 Pull Requests 1 Actions 177 Packages Projects Releases Wiki Activity

graph : remove redundant scale_w parameter (#20235)

Browse Source
This commit is contained in:
Sigbjørn Skjæret
2026-03-08 18:58:28 +01:00
committed by GitHub
parent 451ef08432
commit 35bee031e1
41 changed files with 85 additions and 86 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/qwen35moe.cpp
+8 -4
View File
@@ -375,11 +375,15 @@ ggml_tensor * llm_build_qwen35moe ::build_layer_ffn(ggml_tensor * cur, const int
ggml_tensor * moe_out =
build_moe_ffn(cur,
model.layers[il].ffn_gate_inp, model.layers[il].ffn_up_exps,
model.layers[il].ffn_gate_exps, model.layers[il].ffn_down_exps,
model.layers[il].ffn_gate_inp,
model.layers[il].ffn_up_exps,
model.layers[il].ffn_gate_exps,
model.layers[il].ffn_down_exps,
nullptr,
n_expert, n_expert_used, LLM_FFN_SILU,
true, false, 0.0, LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX, il,
n_expert, n_expert_used,
LLM_FFN_SILU, true,
hparams.expert_weights_scale,
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX, il,
nullptr, model.layers[il].ffn_gate_up_exps);
cb(moe_out, "ffn_moe_out", il);