Spike No More: Stabilizing the Pre-training of Large Language Models.
Sho TakaseShun KiyonoSosuke KobayashiJun SuzukiPublished in: CoRR (2023)
Keyphrases
- language model
- language modeling
- n gram
- information retrieval
- probabilistic model
- speech recognition
- language modelling
- document retrieval
- retrieval model
- query expansion
- query terms
- statistical language models
- smoothing methods
- test collection
- vector space model
- context sensitive
- document ranking
- word error rate
- document length
- language model for information retrieval
- relevance model
- cross lingual
- training set
- pseudo feedback
- text categorization