root/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-24 10:24:35 +00:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

train : fix KQ_pos allocation (#3392)

Browse Source 
* train : fix KQ_pos allocation

* make sure KQ_pos is not reallocated in finetune

---------

Co-authored-by: xaedes <xaedes@gmail.com>
This commit is contained in:
Georgi Gerganov 2023-09-29 19:05:18 +03:00 committed by GitHub
parent 2777a84be4
commit bc34dd4f5b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 6 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
examples/finetune/finetune.cpp
View File

@ -626,7 +626,8 @@ static struct ggml_tensor * llama_build_lora_finetune_graphs(
// KQ_pos - contains the positions // KQ_pos - contains the positions
struct ggml_tensor * KQ_pos = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, N); struct ggml_tensor * KQ_pos = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, N);
{ ggml_allocr_alloc(alloc, KQ_pos);
if (!ggml_allocr_is_measure(alloc)) {
int * data = (int *) KQ_pos->data; int * data = (int *) KQ_pos->data;
for (int i = 0; i < N; ++i) { for (int i = 0; i < N; ++i) {
data[i] = n_past + i; data[i] = n_past + i;
@ -786,6 +787,8 @@ static struct ggml_tensor * llama_build_lora_finetune_graphs(
ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, t36->grad, one)); ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, t36->grad, one));
GGML_ASSERT(t36->grad->data == NULL && t36->grad->view_src == NULL); GGML_ASSERT(t36->grad->data == NULL && t36->grad->view_src == NULL);
ggml_allocr_alloc(alloc, t36->grad); ggml_allocr_alloc(alloc, t36->grad);
// KQ_pos
ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, KQ_pos, one));
// make sure base model tensors data cannot be used in viewable operations // make sure base model tensors data cannot be used in viewable operations
ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, model->tok_embeddings, one)); ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, model->tok_embeddings, one));

3
examples/train-text-from-scratch/train-text-from-scratch.cpp
View File

@ -334,7 +334,8 @@ static struct ggml_tensor * llama_build_train_graphs(
// KQ_pos - contains the positions // KQ_pos - contains the positions
struct ggml_tensor * KQ_pos = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, N); struct ggml_tensor * KQ_pos = ggml_new_tensor_1d(ctx, GGML_TYPE_I32, N);
{ ggml_allocr_alloc(alloc, KQ_pos);
if (!ggml_allocr_is_measure(alloc)) {
int * data = (int *) KQ_pos->data; int * data = (int *) KQ_pos->data;
for (int i = 0; i < N; ++i) { for (int i = 0; i < N; ++i) {
data[i] = n_past + i; data[i] = n_past + i;