Tag: scaling laws

Chinchilla's 70B model beat 280B Gopher using 4x fewer parameters. The secret: training tokens matter more than model size. Proof inside.

Read more →