mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 32 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/mimo2-iswa.cpp
+11 -5
View File
@@ -1,4 +1,3 @@
#include "models.h"
llm_build_mimo2_iswa::llm_build_mimo2_iswa(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
@@ -88,10 +87,17 @@ llm_build_mimo2_iswa::llm_build_mimo2_iswa(const llama_model & model, const llm_
cb(cur, "ffn_out", il);
} else {
// MoE branch
cur = 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, LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID, il);
cur = 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,
hparams.expert_weights_scale,
LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID,
il);
cb(cur, "ffn_moe_out", il);
}