Faster Language Models with Better Multi-Token Prediction Using Tensor Decomposition
This work addresses efficiency issues in large language models for users in text and code generation, representing an incremental improvement over existing multi-token prediction techniques.
The paper tackles the problem of slow inference in language models by proposing a new multi-token prediction method based on tensor decomposition, which improves inference speed for text and code generation without compromising accuracy.
We propose a new model for multi-token prediction in transformers, aiming to enhance sampling efficiency without compromising accuracy. Motivated by recent work that predicts the probabilities of subsequent tokens using multiple heads, we connect this approach to rank-$1$ canonical tensor decomposition. By generalizing it to a rank-$r$ canonical probability decomposition, we develop an improved model that predicts multiple tokens simultaneously. This model can also be interpreted as a mixture of experts, allowing us to leverage successful techniques from that domain for efficient and robust training. Importantly, the overall overhead for training and sampling remains low. Our method demonstrates significant improvements in inference speed for both text and code generation tasks, proving particularly beneficial within the self-speculative decoding paradigm. It maintains its effectiveness across various model sizes and training epochs, highlighting its robustness and scalability.