LlaMaVAE: Guiding Large Language Model Generation via Continuous Latent Sentence Spaces
This work addresses the need for better controllability in text generation for users of large language models, though it is incremental as it builds on existing VAE and LLM methods.
The authors tackled the problem of controlling text generation in large language models by integrating Variational AutoEncoders with LLMs, resulting in LlaMaVAE outperforming the previous state-of-the-art VAE model, Optimus, across tasks like language modeling and semantic textual similarity.
Deep generative neural networks, such as Variational AutoEncoders (VAEs), offer an opportunity to better understand and control language models from the perspective of sentence-level latent spaces. To combine the controllability of VAE latent spaces with the state-of-the-art performance of recent large language models (LLMs), we present in this work LlaMaVAE, which combines expressive encoder and decoder models (sentenceT5 and LlaMA) with a VAE architecture, aiming to provide better text generation control to LLMs. In addition, to conditionally guide the VAE generation, we investigate a new approach based on flow-based invertible neural networks (INNs) named Invertible CVAE. Experimental results reveal that LlaMaVAE can outperform the previous state-of-the-art VAE language model, Optimus, across various tasks, including language modelling, semantic textual similarity and definition modelling. Qualitative analysis on interpolation and traversal experiments also indicates an increased degree of semantic clustering and geometric consistency, which enables better generation control.