From beab06884b51aa537be96c11fd2bd2d3fb2f9ac5 Mon Sep 17 00:00:00 2001 From: Daniel Bevenius Date: Thu, 31 Oct 2024 08:13:31 +0100 Subject: [PATCH] docs: fix typo in llama.md --- notes/llama.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/notes/llama.md b/notes/llama.md index 902b45e..1badb4d 100644 --- a/notes/llama.md +++ b/notes/llama.md @@ -5604,8 +5604,8 @@ $168 = 3 #define GGML_KQ_MASK_PAD 32 #define GGML_PAD(x, n) (((x) + (n) - 1) & ~((n) - 1)) -GGML_PAD(n_tokens, GGML_KQ_MASK_PAD)); -GGML_PAD(3, 32)); +GGML_PAD(n_tokens, GGML_KQ_MASK_PAD); +GGML_PAD(3, 32); GGML_PAD(3, 32) (((3) + (32) - 1) & ~((32) - 1)) = 32 ``` @@ -5621,7 +5621,7 @@ So this would look something like this: 30 [0 ... 255] 31 [0 ... 255] ``` -And this is for the Qeury matrix matrix masking to prevent the model from +And this is for the Query matrix masking to prevent the model from looking at tokens "in the future". ```c++ for (int il = 0; il < n_layer; ++il) {