From 4b1df60e94d2fae6db7b8bf50eb617cb426b704a Mon Sep 17 00:00:00 2001
From: PENG Bo <33809201+BlinkDL@users.noreply.github.com>
Date: Fri, 25 Mar 2022 05:41:15 +0800
Subject: [PATCH] Update README.md

---
 README.md | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/README.md b/README.md
index 50b03be..ee435c8 100644
--- a/README.md
+++ b/README.md
@@ -10,7 +10,9 @@ Hence it can be 100x faster than GPT, and 100x more VRAM friendly.
 
 The a b c d factors work together to build a time-decay curve: u, 1, w, w^2, w^3, ...
 
-Write out the formulas for "token at pos 2" and "token at pos 3" and you will get the idea.
+Write out the formulas for "token at pos 2" and "token at pos 3" and you will get the idea:
+* a and b: EMAs of kv and k.
+* c and d: a and b combined with self-attention.
 
 The model: