5fa1a76
1
2
3
If a model's max input size is \(k\), we then approximate the likelihood of a token \(x_t\) by conditioning only on the \(k-1\) tokens that precede it rather than the entire context.
