Share your thoughts, 1 month free Claude Pro on usSee more

Character-level Language Modeling on text8 100M regime (Backward stream)

2.17Backward BPC

Transformer (ctx=1024)

Updated 23h ago

Evaluation Results

Method	Links
Transformer (ctx=1024) 2026.05		2.17
Transformer (ctx=1024) 2026.05		2.2
SHARP 2026.05		2.33
GRU 2026.05		2.56
Clockwork RNN 2026.05		2.78
LSTM 2026.05		3.14
RNN 2026.05		4.63