From 37429b7da27c4c1917c2cae00781f3a73a09c8d0 Mon Sep 17 00:00:00 2001
From: deep1 <>
Date: Sun, 24 Sep 2023 16:39:23 +0800
Subject: [PATCH] :poop:

---
 mjc_notes.md | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/mjc_notes.md b/mjc_notes.md
index 37ea3a3..7435e60 100644
--- a/mjc_notes.md
+++ b/mjc_notes.md
@@ -1546,3 +1546,7 @@ print(pd.Series(ds_tokens['instructed_to_lie']).value_counts()) # should be 50%
 
 
 Ah found it :brain: it was using the same random seed. so I was selecting the Nth each time, which happened to be diff for each dataset. But was the same template and type. OK now I can redo.
+
+# 2023-09-24 16:38:46
+
+I got poor results in all the latest datasets. Maybe grad methods require too small a model and the model is too small to know anything? It certainly gets poor acc on simple sentiment tasks.