X-Git-Url: https://fleuret.org/cgi-bin/gitweb/gitweb.cgi?a=blobdiff_plain;f=README.txt;fp=README.txt;h=231d292110ec4d88e7a22e2273683a8f7c53e0bb;hb=7c8049b2dde68455147c74c6245fc5d44aa03077;hp=bf6180a1c5b3edbacc5da38493528b921bdcd871;hpb=45f841397dd6f8fb163f2c5b0793cdc250072c73;p=picoclvr.git diff --git a/README.txt b/README.txt index bf6180a..231d292 100644 --- a/README.txt +++ b/README.txt @@ -11,6 +11,7 @@ For the arithmetic expressions experiments ./main.py --task=expr -# 352M parameters / 2.5M samples +# 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 --result_dir=results_expr_48b_d1024_2.5M --dim_model=1024 --nb_train_samples=2500000 --learning_rate_schedule="1:2e-5,3:4e-6" +./main.py --task=expr --nb_blocks=48 --result_dir=results_expr_48b_d1024_2.5M --dim_model=1024 --nb_train_samples=2500000