abertsch72 / abertsch72/minimum-bayes-risk
add course to fine MBR pipeline
Open
- Dominant language
- Python
- Stars
- 2
- Forks
- 0
- PR merge metrics
- No merged PRs in 30d
Description
from https://aclanthology.org/2022.emnlp-main.754.pdf:
1. sample a ton from the model according to some set of sampling arguments
2. use a cheap MBR metric (allow ROUGE, BLEU, BEER?) to rerank
3. take the top k from that list as the top-k list
todo: is the score the score from reranking or the original model score of each?
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.