@@ -10863,7 +10863,7 @@ struct llm_build_context {
10863
10863
// special-case: the up and gate tensors are merged into a single tensor
10864
10864
// TOOD: support into llm_build_ffn
10865
10865
{
10866
- cur = llm_build_ffn(ctx0, cur,
10866
+ cur = llm_build_ffn(lctx, ctx0, cur,
10867
10867
model.layers[il].ffn_up, NULL, NULL,
10868
10868
NULL, NULL, NULL,
10869
10869
model.layers[il].ffn_down, NULL, NULL,
@@ -13622,7 +13622,7 @@ struct llm_build_context {
13622
13622
);
13623
13623
cb(Kcur, "Kcur_rope", il);
13624
13624
13625
- cur = llm_build_kv(ctx0, model, hparams, cparams, kv_self, gf,
13625
+ cur = llm_build_kv(lctx, ctx0, model, hparams, cparams, kv_self, gf,
13626
13626
model.layers[il].wo, NULL,
13627
13627
Kcur, Vcur, Qcur, KQ_mask, n_tokens, kv_head, n_kv, 1.0f/sqrtf(float(n_embd_head)), cb, il);
13628
13628
@@ -13647,7 +13647,7 @@ struct llm_build_context {
13647
13647
LLM_NORM_RMS, cb, il);
13648
13648
cb(cur, "ffn_norm", il);
13649
13649
13650
- cur = llm_build_ffn(ctx0, cur,
13650
+ cur = llm_build_ffn(lctx, ctx0, cur,
13651
13651
model.layers[il].ffn_up, NULL, NULL,
13652
13652
NULL, NULL, NULL,
13653
13653
model.layers[il].ffn_down, NULL, NULL,
0 commit comments