Efficient Self-Evaluation for Diffusion Language Models via Sequence Regeneration
This work addresses the challenge of evaluating non-sequential generation in dLLMs, which is crucial for applications requiring robust quality control, though it appears incremental as it builds on existing self-evaluation concepts.
The paper tackles the problem of quality assessment in diffusion large language models (dLLMs) by proposing DiSE, a self-evaluation method that quantifies confidence through token regeneration probabilities, resulting in improved efficiency and reliability in tasks like likelihood estimation and uncertainty quantification.
Diffusion large language models (dLLMs) have recently attracted significant attention for their ability to enhance diversity, controllability, and parallelism. However, their non-sequential, bidirectionally masked generation makes quality assessment difficult, underscoring the need for effective self-evaluation. In this work, we propose DiSE, a simple yet effective self-evaluation confidence quantification method for dLLMs. DiSE quantifies confidence by computing the probability of regenerating the tokens in the entire generated sequence, given the full context. This method enables more efficient and reliable quality assessment by leveraging token regeneration probabilities, facilitating both likelihood estimation and robust uncertainty quantification. Building upon DiSE, we further introduce a flexible-length generation framework, which adaptively controls the sequence length based on the model's self-assessment of its own output. We analyze and validate the feasibility of DiSE from the perspective of dLLM generalization, and empirically demonstrate that DiSE is positively correlated with both semantic coherence and answer accuracy. Extensive experiments on likelihood evaluation, uncertainty quantification, and flexible-length generation further confirm the effectiveness of the proposed DiSE.