graph : remove redundant scale_w parameter (#20235)
This commit is contained in:
@@ -1,7 +1,5 @@
|
||||
#include "models.h"
|
||||
|
||||
|
||||
|
||||
llm_build_deepseek::llm_build_deepseek(const llama_model & model, const llm_graph_params & params) :
|
||||
llm_graph_context(params) {
|
||||
const int64_t n_embd_head = hparams.n_embd_head_v;
|
||||
@@ -100,7 +98,7 @@ llm_build_deepseek::llm_build_deepseek(const llama_model & model, const llm_grap
|
||||
nullptr,
|
||||
n_expert, n_expert_used,
|
||||
LLM_FFN_SILU, false,
|
||||
false, hparams.expert_weights_scale,
|
||||
hparams.expert_weights_scale,
|
||||
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
|
||||
il);
|
||||
cb(moe_out, "ffn_moe_out", il);
|
||||
|
||||
Reference in New Issue
Block a user