HadSkip: Homotopic and Adaptive Layer Skipping of Pre-trained Language Models for Efficient Inference.
Haoyu WangYaqing WangTianci LiuTuo ZhaoJing GaoPublished in: EMNLP (Findings) (2023)
Keyphrases
- language model
- efficient inference
- pre trained
- probabilistic model
- conditional random fields
- probabilistic inference
- hidden variables
- information retrieval
- fully connected
- speech recognition
- markov random field
- training data
- approximate inference
- training examples
- graph structure
- graphical models
- markov networks
- object recognition