mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 30 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/lfm2.cpp
+10 -4
View File
@@ -23,10 +23,16 @@ llm_build_lfm2<iswa>::llm_build_lfm2(const llama_model & model, const llm_graph_
};
auto build_moe_feed_forward = [&model, this](ggml_tensor * cur, int il) -> ggml_tensor * {
return 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_exp_probs_b, n_expert, n_expert_used, LLM_FFN_SILU, true, false, 0.0,
static_cast<llama_expert_gating_func_type>(hparams.expert_gating_func), il);
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_exp_probs_b,
n_expert, n_expert_used,
LLM_FFN_SILU, true,
hparams.expert_weights_scale,
static_cast<llama_expert_gating_func_type>(hparams.expert_gating_func),
il);
};
auto build_attn_block = [&model, this](ggml_tensor * cur,
ggml_tensor * inp_pos,