From e6cc7c9ff78b171ba2218ac421242ab399143029 Mon Sep 17 00:00:00 2001 From: Gustaf Rydholm Date: Fri, 25 Aug 2023 23:19:51 +0200 Subject: Update reqs --- README.md | 3 +++ 1 file changed, 3 insertions(+) (limited to 'README.md') diff --git a/README.md b/README.md index 164ee24..969e192 100644 --- a/README.md +++ b/README.md @@ -80,3 +80,6 @@ Ideas of mine that did not work unfortunately: - [x] residual attn - [x] single kv head - [x] fix rotary embedding +- [ ] simplify attention with norm +- [ ] tie embeddings +- [ ] cnn -> tf encoder -> tf decoder -- cgit v1.2.3-70-g09d2