vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama: dbrx: fix build kv att out

Browse source
This commit is contained in:
Pierrick HYMBERT 2024-04-08 00:25:54 +02:00
parent 993f836029
commit b01b062ab5
1 changed files with 1 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

10
llama.cpp
View file

@ -7155,7 +7155,7 @@ struct llm_build_context {
cb(Vcur, "Vcur", il); cb(Vcur, "Vcur", il);
cur = llm_build_kv(ctx0, model, hparams, kv_self, gf, cur = llm_build_kv(ctx0, model, hparams, kv_self, gf,
model.layers[il].wo, model.layers[il].bo, model.layers[il].attn_out_norm, NULL,
Kcur, Vcur, Qcur, KQ_mask, nullptr, n_ctx, n_tokens, kv_head, n_kv, 1.0f, cb, il); Kcur, Vcur, Qcur, KQ_mask, nullptr, n_ctx, n_tokens, kv_head, n_kv, 1.0f, cb, il);
} }
@ -7235,14 +7235,6 @@ struct llm_build_context {
cur = moe_out; cur = moe_out;
} }
// DbrxNormAttentionNorm
{
cur = llm_build_norm(ctx0, cur, hparams,
model.layers[il].layer_out_norm, NULL,
LLM_NORM, cb, il);
cb(cur, "layer_out_norm", il);
}
cur = ggml_add(ctx0, cur, ffn_inp); cur = ggml_add(ctx0, cur, ffn_inp);
cb(cur, "ffn_out", il); cb(cur, "ffn_out", il);