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

fix rope inplace

Browse source
This commit is contained in:
Concedo 2023-06-27 19:44:33 +08:00
parent 282376c85a
commit 9527a783ea
2 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

4
otherarch/gptj_v3.cpp
View file

@ -462,8 +462,8 @@ bool gptj_eval(
// self-attention // self-attention
{ {
struct ggml_tensor * Qcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, ggml_mul_mat(ctx0, model.layers[il].c_attn_q_proj_w, cur), n_embd/n_head, n_head, N), n_past, n_rot, 0); struct ggml_tensor * Qcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, ggml_mul_mat(ctx0, model.layers[il].c_attn_q_proj_w, cur), n_embd/n_head, n_head, N), n_past, n_rot, 0, 0);
struct ggml_tensor * Kcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, ggml_mul_mat(ctx0, model.layers[il].c_attn_k_proj_w, cur), n_embd/n_head, n_head, N), n_past, n_rot, 0); struct ggml_tensor * Kcur = ggml_rope_inplace(ctx0, ggml_reshape_3d(ctx0, ggml_mul_mat(ctx0, model.layers[il].c_attn_k_proj_w, cur), n_embd/n_head, n_head, N), n_past, n_rot, 0, 0);
// store key and value to memory // store key and value to memory
{ {

4
otherarch/neox_v3.cpp
View file

@ -492,8 +492,8 @@ bool gpt_neox_eval(
struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd/n_head, n_head, N, cur->nb[1]/n_head, cur->nb[1], 2*sizeof(float)*n_embd/n_head)); struct ggml_tensor * Vcur = ggml_cont(ctx0, ggml_view_3d(ctx0, cur, n_embd/n_head, n_head, N, cur->nb[1]/n_head, cur->nb[1], 2*sizeof(float)*n_embd/n_head));
// using mode = 2 for GPT-NeoX mode // using mode = 2 for GPT-NeoX mode
Qcur = ggml_rope_inplace(ctx0, Qcur, n_past, n_rot, 2); Qcur = ggml_rope_inplace(ctx0, Qcur, n_past, n_rot, 2, 0);
Kcur = ggml_rope_inplace(ctx0, Kcur, n_past, n_rot, 2); Kcur = ggml_rope_inplace(ctx0, Kcur, n_past, n_rot, 2, 0);
// store key and value to memory // store key and value to memory
{ {