https://www.youtube.com/watch?v=N9L32bFieEY&t=1s
https://web.stanford.edu/class/archive/cs/cs224n/cs224n.1234/slides/cs224n-2023-lecture10-nlg.pdf
Pada bagian 3 Decoding from NLG models
Misal memggunakan:
- greedy decoding vs beam search
- Sampling: Top-k sampling, Top-p sampling, typical sampling, epsilon sampling
- Scaling randomness: temperature
- Reranking
