@@ -13530,7 +13530,7 @@ struct llm_build_exaone : public llm_graph_context {
13530
13530
13531
13531
template <bool iswa>
13532
13532
struct llm_build_exaone4 : public llm_graph_context {
13533
- llm_build_exaone4(const llama_model & model, const llm_graph_params & params, ggml_cgraph * gf ) : llm_graph_context(params) {
13533
+ llm_build_exaone4(const llama_model & model, const llm_graph_params & params) : llm_graph_context(params) {
13534
13534
const int64_t n_embd_head = hparams.n_embd_head_k;
13535
13535
13536
13536
GGML_ASSERT(n_embd_head == hparams.n_embd_head_v);
@@ -13603,7 +13603,7 @@ struct llm_build_exaone4 : public llm_graph_context {
13603
13603
cb(Kcur, "Kcur", il);
13604
13604
cb(Vcur, "Vcur", il);
13605
13605
13606
- cur = build_attn(inp_attn, gf,
13606
+ cur = build_attn(inp_attn,
13607
13607
model.layers[il].wo, NULL,
13608
13608
Qcur, Kcur, Vcur, nullptr, nullptr, 1.0f/sqrtf(float(n_embd_head)), il);
13609
13609
cb(cur, "attn_out", il);
@@ -17352,9 +17352,9 @@ ggml_cgraph * llama_model::build_graph(const llm_graph_params & params) const {
17352
17352
case LLM_ARCH_EXAONE4:
17353
17353
{
17354
17354
if (hparams.swa_type == LLAMA_SWA_TYPE_STANDARD) {
17355
- llm = std::make_unique<llm_build_exaone4<true>>(*this, params, gf );
17355
+ llm = std::make_unique<llm_build_exaone4<true>>(*this, params);
17356
17356
} else {
17357
- llm = std::make_unique<llm_build_exaone4<false>>(*this, params, gf );
17357
+ llm = std::make_unique<llm_build_exaone4<false>>(*this, params);
17358
17358
}
17359
17359
} break;
17360
17360
case LLM_ARCH_RWKV6:
0 commit comments