Token sampling strategies critically influence text generation quality in large language models (LLMs). However, existing methods introduce additional hyperparameters, requiring extensive tuning and complicating deployment. We present Entropy Equilibrium Sampling (EES), a tuning-free approach grounded in information theory that dynamically determines the candidate set size using a first-crossing rule: it stops candidate-set expansion at the first index where cumulative probability mass exceeds the normalized entropy of the top-k distribution and retains the preceding prefix. We evaluate EES on reasoning, mathematical, and open-ended generation tasks across multiple model architectures and scales. Our results show that EES achieves robust performance across temperature settings without retuning truncation hyperparameters, delivering competitive accuracy with tuned baselines while maintaining diversity. Analysis reveals that the retained candidate-set size k* automatically encodes temperature effects, adapting from near-greedy behavior at low temperatures to broad candidate retention at high temperatures. By eliminating truncation-specific hyperparameter tuning, EES simplifies deployment while providing stable and competitive performance. Our code is publicly available.

Reinforcement learning with verifiable rewards (RLVR) has become a standard paradigm for post-training large language models (LLMs). While Group Relative Policy Optimization (GRPO) is widely adopted, it suffers from sparse reward signals and loses gradients entirely when all responses within a group receive identical rewards. On-policy distillation (OPD) offers a natural remedy by providing dense, token-level supervision from a teacher model. However, naively combining GRPO with OPD leads to degraded performance, due to three underlying causes: not all samples benefit from distillation; fitting too quickly to the teacher undermines the exploratory capacity of RL; and OPD’s advantages are asymmetric, suppressing most tokens. To address these challenges, we propose RSTG (Recovering Learning Signals via Adaptive Teacher Guidance), which applies distillation selectively and precisely where it matters most. At the sample level, OPD is restricted to negative zero-variance prompts with each sample weighted by the teacher’s confidence score. At the token level, distillation targets only tokens with high student entropy or large teacher-student divergence. We further augment training with SFT on correct trajectories generated by the teacher model, injecting positive gradient signals where RL yields none. Experiments demonstrate that RSTG substantially outperforms naive GRPO+OPD by +4.02% on math and +3.05% on code.

Existing methods for incorporating entities into EAE rely on prompts or NER. They typically fail to explicitly explore the role of entity types, which results in shallow argument comprehension and often encounter three issues: (1) weak semantic associations due to missing role-entity correspondence cues; (2) compromised semantic integrity from abandoning context after recognizing entities regardless of their types; (3) one-sided semantic understanding relying solely on argument role semantics. To tackle these issues, we propose Scented-EAE, an EAE model with stage-customized entity type embedding to explicitly underscore and explore the role of entity types, thus intervening in argument selection. Specifically, at the input stage, we strengthen semantic associations by prompting role-entity correspondence after extending a non-autoregressive decoder as part of the encoder. At the intermediate stage, we preserve semantic integrity by optimizing our proposed BIO-aware NER and EAE via a novel IPE joint learning. At the output stage, we expand semantic understanding dimensions by determining arguments using span selectors from argument roles and entity types. Experiments show that our model achieves state-of-the-art performance on mainstream benchmarks. In addition, it also exhibits robustness in low-resource settings with the help of prompts and entity types.

This paper presents FlowSUM, a normalizing flows-based variational encoder-decoder framework for Transformer-based summarization. Our approach tackles two primary challenges in variational summarization: insufficient semantic information in latent representations and posterior collapse during training. To address these challenges, we employ normalizing flows to enable flexible latent posterior modeling, and we propose a controlled alternate aggressive training (CAAT) strategy with an improved gate mechanism. Experimental results show that FlowSUM significantly enhances the quality of generated summaries and unleashes the potential for knowledge distillation with minimal impact on inference time. Furthermore, we investigate the issue of posterior collapse in normalizing flows and analyze how the summary quality is affected by the training strategy, gate initialization, and the type and number of normalizing flows used, offering valuable insights for future research.

We consider the problem of scaling automated suggested replies for a commercial email application to multiple languages. Faced with increased compute requirements and low language resources for language expansion, we build a single universal model for improving the quality and reducing run-time costs of our production system. However, restricted data movement across regional centers prevents joint training across languages. To this end, we propose a multi-lingual multi-task continual learning framework, with auxiliary tasks and language adapters to train universal language representation across regions. The experimental results show positive cross-lingual transfer across languages while reducing catastrophic forgetting across regions. Our online results on real user traffic show significant CTR and Char-saved gain as well as 65% training cost reduction compared with per-language models. As a consequence, we have scaled the feature in multiple languages including low-resource markets.