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

llama : adapt to F16 KQ_pos

Browse source
This commit is contained in:
Georgi Gerganov 2024-02-19 13:10:24 +02:00
parent 31109ca00a
commit f249c997a8
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
4 changed files with 13 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

2
tests/test-backend-ops.cpp
View file

@ -1505,7 +1505,7 @@ struct test_attn : public test_case {
struct ggml_tensor * cur;
cur = ggml_mul_mat (ctx, k, q);
cur = ggml_soft_max_ext(ctx, cur, mask, 1.0f/sqrtf(hs));
cur = ggml_soft_max_ext(ctx, cur, mask, nullptr, 1.0f/sqrtf(hs), 0.0f);
cur = ggml_mul_mat (ctx, v, cur);
cur = ggml_permute (ctx, cur, 0, 2, 1, 3);
cur = ggml_cont_2d (ctx, cur, hs*nh, nb);