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

llama : fix rope usage after ChatGLM change

Browse source
This commit is contained in:
Georgi Gerganov 2023-06-27 00:37:13 +03:00
parent d9779021bd
commit 181e8d9755
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 12 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

4
llama.cpp
View file

@ -1491,11 +1491,11 @@ static bool llama_eval_internal(
offload_func_kq(tmpq);
ggml_set_name(tmpq, "tmpq");
struct ggml_tensor * Kcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, tmpk, n_embd/n_head, n_head, N), n_past, n_rot, 0);
struct ggml_tensor * Kcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, tmpk, n_embd/n_head, n_head, N), n_past, n_rot, 0, 0);
offload_func_kq(Kcur);
ggml_set_name(Kcur, "Kcur");
struct ggml_tensor * Qcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, tmpq, n_embd/n_head, n_head, N), n_past, n_rot, 0);
struct ggml_tensor * Qcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, tmpq, n_embd/n_head, n_head, N), n_past, n_rot, 0, 0);
offload_func_kq(Qcur);
ggml_set_name(Qcur, "Qcur");