CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

train : make KQ_pos memory buffer permanent via dummy scale op

Browse Source
This commit is contained in:
Georgi Gerganov
2023-09-20 14:14:50 +03:00
parent 54206962c7
commit 2f3a46fccf
1 changed files with 2 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
examples/train-text-from-scratch/train-text-from-scratch.cpp
View File

@@ -795,6 +795,8 @@ struct ggml_tensor * llama_build_train_graphs(
ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, t36, one)); ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, t36, one));
// input gradient // input gradient
ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, t36->grad, one)); ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, t36->grad, one));
// KQ_pos
ggml_build_forward_expand(gb, ggml_scale_inplace(ctx, KQ_pos, one));
GGML_ASSERT(t36->grad->data == NULL && !ggml_is_view(t36->grad)); GGML_ASSERT(t36->grad->data == NULL && !ggml_is_view(t36->grad));
ggml_allocr_alloc(alloc, t36->grad); ggml_allocr_alloc(alloc, t36->grad);
// gradient tensors (will be set to zero by ggml_graph_reset) // gradient tensors (will be set to zero by ggml_graph_reset)