While Reinforcement Learning (RL) effectively incentivizes reasoning in Large Language Models, current pipelines are hindered by training instability and rapid entropy collapse. These limitations often stem from "Rollout Silencing" and low-quality gradient signals in standard sampling procedures. In this work, we propose a robust, data-centric framework to stabilize RL training. We first introduce Potential-Aware Query Mining (PAQM), which filters data dynamically to focus on the "Distillation Zone"—samples with high potential for capability elicitation. Furthermore, we present Hybrid Stratified Replay (HSR), a novel mechanism that restructures batches by stratifying rollouts based on Path Entropy, a rollout-level confidence proxy, and outcome reward. Within each optimization step, HSR reuses current-policy "Stability Anchors" and "Hard Negatives" to construct high-contrast optimization groups, then clears its buffers before the next step. This approach mitigates entropy collapse while improving the utilization of learning signals under limited compute. Our method outperforms strong baselines on complex reasoning tasks, offering a principled solution for stable and efficient RL fine-tuning.

Latent test-time scaling improves reasoning by refining hidden states during inference, but existing methods typically apply a single scalar reward to all editable latent tokens. For multimodal large language models, this global update ignores that generated tokens play different roles: some are sensitive to visual evidence, while others correspond to uncertain reasoning decisions. We present Token-Disentangled Latent Test-Time Scaling, an inference-time framework that makes latent refinement token-role-aware. Starting from an initial generated trajectory, we optimize a short hidden-state prefix while routing perception-side visual feedback to image-sensitive tokens and reasoning feedback to high-entropy tokens. Tokens selected by neither route are constrained by an anchor regularizer. Across six multimodal reasoning benchmarks and two frozen MLLM backbones, our method consistently improves over the initial chain-of-thought rollout: macro accuracy increases from 66.38 to 68.95 on Qwen2.5-VL-7B and from 67.93 to 69.44 on InternVL3.5-8B. These results show that token-disentangled latent refinement is an effective strategy for multimodal test-time scaling and remains competitive with output-space test-time scaling baselines.

Diffusion language models (dLLMs) have recently emerged as a promising alternative to autoregressive (AR) language models, offering competitive performance with a fundamentally different generation paradigm. Instead of producing tokens strictly left-to-right, dLLMs iteratively refine partially masked sequences, enabling bidirectional context modeling and supporting parallel token updates. This design creates new opportunities for accelerating generation and improving controllability. However, practical deployment of dLLMs remains challenging due to the high cost of training at scale and the lack of mature inference optimizations such as cache-friendly decoding and robust parallel sampling. In this work, we provide a comprehensive overview of recent progress on efficient dLLMs. To reflect how efficiency bottlenecks arise across the full lifecycle of dLLMs, spanning training, decoding, and deployment, we organize existing approaches into five categories: Training, Inference, Context, Framework, and Multimodality. For each category, we summarize representative methods and highlight their motivations, key techniques, and empirical trade-offs. Finally, we discuss open challenges and outline promising directions for future research toward high-throughput, robust, and practically deployable diffusion language models.