From 181e8d975528a4e27eabb8ae6e9865f9ceae4b37 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Tue, 27 Jun 2023 00:37:13 +0300 Subject: llama : fix rope usage after ChatGLM change --- llama.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) (limited to 'llama.cpp') diff --git a/llama.cpp b/llama.cpp index 1a15844b..2482bdd1 100644 --- a/llama.cpp +++ b/llama.cpp @@ -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"); -- cgit v1.2.3