X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=README.txt;h=d4740f386def252fb3b3229bcb97ba95b40fb3dc;hb=ce969e8372fb161d86be29042a20b044ee6efe2a;hp=223565e63ad6236d33f90fa38e18d8b15e9d3193;hpb=dbb361548096536b62f50d810885439043ec08a3;p=picoclvr.git

diff --git a/README.txt b/README.txt
index 223565e..d4740f3 100644
--- a/README.txt
+++ b/README.txt
@@ -1,5 +1,23 @@
+
+======================================================================
 For the stack experiment:
 
-./main.py --task=stack --stack_fraction_values_for_train=0.75 --stack_nb_stacks=3
+./main.py --task=stack
 
 Takes ~1h10min on a 4090.
+
+======================================================================
+For the arithmetic expressions experiments
+
+# 38M parameters / 250k samples
+
+./main.py --task=expr
+
+# 352M parameters / 2.5M samples, reaches 99.80% after 12 epochs, the
+  learning rate schedule is obviously terrible
+
+./main.py --task=expr --nb_blocks=48 --dim_model=1024 --nb_train_samples=2500000 --result_dir=results_expr_48b_d1024_2.5M
+======================================================================
+25.07.2023
+
+./main.py --task=sandbox --nb_train_samples=10000 --nb_test_samples=1000 --nb_blocks=4 --nb_heads=1 --nb_epochs=20