diff options
author | Georgi Gerganov <ggerganov@gmail.com> | 2023-06-27 00:37:13 +0300 |
---|---|---|
committer | Georgi Gerganov <ggerganov@gmail.com> | 2023-06-27 00:37:33 +0300 |
commit | 181e8d975528a4e27eabb8ae6e9865f9ceae4b37 (patch) | |
tree | 4f6fc037755fbbca36518be79e1cdb04ef99f764 /llama.cpp | |
parent | d9779021bd59ed96daae75e820a5ac5da47ca8ff (diff) |
llama : fix rope usage after ChatGLM change
Diffstat (limited to 'llama.cpp')
-rw-r--r-- | llama.cpp | 4 |
1 files changed, 2 insertions, 2 deletions
@@ -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"); |