github/gpt4all
1
0
Fork 0
mirror of https://github.com/nomic-ai/gpt4all.git synced 2025-09-06 11:00:48 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update to latest llama.cpp (#1706)

Browse Source
This commit is contained in:
Jared Van Bortel
2023-12-01 16:51:15 -05:00
committed by GitHub
parent 012f399639
commit 9e28dfac9c
5 changed files with 113 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

17
gpt4all-backend/gptj.cpp
View File

@@ -345,6 +345,13 @@ bool gptj_eval(
struct ggml_context * ctx0 = ggml_init(params);
struct ggml_cgraph gf = {};
// KQ_pos - contains the positions
struct ggml_tensor * KQ_pos = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
int * data = (int *) KQ_pos->data;
for (int i = 0; i < N; ++i) {
data[i] = n_past + i;
}
struct ggml_tensor * embd = ggml_new_tensor_1d(ctx0, GGML_TYPE_I32, N);
memcpy(embd->data, embd_inp.data(), N*ggml_element_size(embd));
@@ -370,8 +377,14 @@ bool gptj_eval(
// self-attention
{
struct ggml_tensor * Qcur = ggml_rope(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(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);
struct ggml_tensor * Qcur = ggml_rope(
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),
KQ_pos, n_rot, 0, 0
);
struct ggml_tensor * Kcur = ggml_rope(
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),
KQ_pos, n_rot, 0, 0
);
// store key and value to memory
{