graph : remove redundant scale_w parameter (#20235)
This commit is contained in:
@@ -76,7 +76,7 @@ llm_build_jamba::llm_build_jamba(const llama_model & model, const llm_graph_para
|
||||
nullptr,
|
||||
n_expert, n_expert_used,
|
||||
LLM_FFN_SILU, false,
|
||||
false, 0.0,
|
||||
hparams.expert_weights_scale,
|
||||
LLAMA_EXPERT_GATING_FUNC_TYPE_SOFTMAX,
|
||||
il);
|
||||
cb(cur, "ffn_moe_out", il);
|
||||
|
||||
Reference in New Issue
Block a user