A library for minimum Bayes risk (MBR) decoding
-
Updated
Nov 2, 2025 - Python
#
mbr-decoding
Here are 3 public repositories matching this topic...
Characterizing the performance of minimum Bayes risk (MBR) decoding for neural machine translation (NMT) when using various candidate generation methods by examining BLEU score, length bias, and token frequency bias. [CS287: Deep Learning for Natural Language Processing, Final Project, Fall 2021]
transformers beam-search neural-machine-translation nmt huggingface-transformers minimum-bayes-risk-decoding mbr-decoding length-bias token-frequency-bias
-
Updated
Dec 11, 2021 - Jupyter Notebook
The Anatomy of the CTC Oracle Gap: the near-converged CTC ASR decoding bottleneck is linguistic, not acoustic. MBR + RoBERTa PLL → ~9% relative WER reduction. Paper + code
pytorch speech-recognition beam-search asr k2 ctc librispeech roberta mbr-decoding language-model-rescoring
-
Updated
Jun 24, 2026 - Python
Add this topic to your repo
To associate your repository with the mbr-decoding topic, visit your repo's landing page and select "manage topics."