vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-05 23:41:45 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

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
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
41 changed files with 85 additions and 86 deletions
Download patch file Download diff file Expand all files Collapse all files

6
src/llama-graph.cpp
View file

@ -1151,7 +1151,6 @@ ggml_tensor * llm_graph_context::build_ffn(
return cur;
}
// TODO remove redundant scale_w argument
ggml_tensor * llm_graph_context::build_moe_ffn(
ggml_tensor * cur,
ggml_tensor * gate_inp,
@ -1163,7 +1162,6 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
int64_t n_expert_used,
llm_ffn_op_type type_op,
bool norm_w,
bool scale_w,
float w_scale,
llama_expert_gating_func_type gating_op,
int il,
@ -1180,7 +1178,6 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
n_expert_used,
type_op,
norm_w,
scale_w,
w_scale,
gating_op,
il,
@ -1204,7 +1201,6 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
int64_t n_expert_used,
llm_ffn_op_type type_op,
bool norm_w,
bool scale_w,
float w_scale,
llama_expert_gating_func_type gating_op,
int il,
@ -1332,7 +1328,7 @@ ggml_tensor * llm_graph_context::build_moe_ffn(
weights = ggml_reshape_3d(ctx0, weights, 1, n_expert_used, n_tokens);
}
if (scale_w) {
if (w_scale != 0.0f && w_scale != 1.0f) {
weights = ggml_scale(ctx0, weights, w_scale);
cb(weights, "ffn_moe_weights_scaled", il);
}