DOA: Training-Free Decoder-Only Attention Policy for Long-Form Simultaneous Speech Translation with SpeechLLMs
The paper proposes Decoder-Only Attention (DOA), a training-free streaming policy for simultaneous speech-to-text translation (SimulST) that works with off-the-shelf decoder-only Speech LLMs. DOA derives proxy alignment signals from self-attention rather than cross-attention, enabling long-form simultaneous translation without retraining. Experiments on Phi4-Multimodal and Qwen3-Omni demonstrate low-latency performance approaching offline decoding quality, validating that decoder self-attention contains sufficient alignment information for streaming decisions.
Related guides (3)
Related events (8)
AlignAtt4LLM adapts simultaneous speech translation policy to decoder-only LLMs for IWSLT 2026
Researchers present AlignAtt4LLM, a simultaneous speech translation system for IWSLT 2026 covering English to German, Italian, and Chinese. The system cascades Qwen3-ASR for incremental transcription with Gemma-4 E4B-it for translation, applying a novel AlignAtt policy adapted for decoder-only LLMs that lack encoder-decoder cross-attention. Key contributions include explicit source span prompting, offline alignment head selection, and query/key capture to recover a usable attention-based read/write policy. The system outperforms IWSLT 2026 baselines for European language pairs in both low- and high-latency regimes.
AGDO: Attention-guided denoising and optimization framework improves diffusion language model reasoning
Researchers propose AGDO, a framework that replaces random masking in diffusion large language models (dLLMs) with attention-guided denoising order and token weighting during fine-tuning and reinforcement learning. The work is motivated by an empirical finding that tokens with stronger attention to unmasked context are more stable and critical for reasoning. Experiments on math and coding benchmarks show AGDO outperforms existing post-training methods for dLLMs, advancing the case for attention-aware training in parallel-decoding language models.
ADAS: Attention-Discounted Adaptive Sampler improves parallel decoding for masked diffusion language models
Researchers propose ADAS, a training-free reranking rule for masked diffusion language model decoding that addresses token interaction failures in parallel token commitment. The method greedily penalizes candidates that attend strongly to already-selected uncertain positions, using attention weights as soft marginal penalties rather than hard constraints. Evaluated on LLaDA-8B-Base and Dream-7B-Base across GSM8K, MATH500, HumanEval, and MBPP, ADAS improves low-NFE performance by 9–10 percentage points on average when plugged into existing samplers with only 3.1% runtime overhead.
Loong: A Human-Like Long Document Translation Agent with Observe-and-Act Adaptive Context Selection
Loong is a long document translation agent that uses a 3E memory module (Essence-Exemplar-Entity) to store structured historical context, replacing passive full-context attention with RL-optimized adaptive context selection. The agent learns its context retrieval policy via reinforcement learning on self-sampled reasoning trajectories. Evaluations show average gains of up to 13.0 points across three metrics in English↔Chinese, German, and French translation directions, with strong generalization and robustness to noise in ultra-long documents.
SimSD: Speculative Decoding Adapted for Diffusion Language Models
SimSD introduces a training-free speculative decoding algorithm for diffusion large language models (dLLMs), which previously could not use standard token-level speculative decoding due to their bidirectional attention and masked language modeling formulation. The method uses a plug-and-play masking strategy that introduces reference tokens from a draft model and a custom attention mask, enabling valid logit computation for drafted tokens in a single forward pass. Evaluated on SDAR-family dLLMs across four benchmarks, SimSD achieves up to 7.46x decoding throughput improvement while maintaining or improving generation quality. The approach is compatible with other acceleration techniques such as KV cache and blockwise decoding.
ASRD: Training-free anchor-guided revocable decoding for diffusion LLMs improves accuracy and throughput
A new arXiv preprint introduces ASRD (Anchor Supervised Revocable Decoding), a training-free framework for improving decoding quality in diffusion large language models. The method addresses error propagation and local error reinforcement in revocable decoding by separating trusted 'anchor tokens' (identified via temporal consistency) from uncertain candidates, then applying anchor-guided generation and anchor-perturbed verification. Experiments on math and coding benchmarks show up to 6.4% accuracy improvement and 7.2× inference throughput gains over remasking baselines.
d-OPSD: First on-policy self-distillation framework tailored for diffusion LLMs
Researchers introduce d-OPSD, the first on-policy self-distillation (OPSD) framework designed specifically for diffusion large language models (dLLMs). The method addresses a fundamental mismatch between existing autoregressive OPSD approaches and dLLMs' arbitrary-order generation by using suffix conditioning on self-generated answers and step-level rather than token-level divergence supervision. Across four reasoning benchmarks, d-OPSD outperforms RLVR and SFT baselines while requiring only ~10% of the optimization steps of RLVR, suggesting strong sample efficiency gains for dLLM post-training.
LESS: Adaptive mutual-stability sampling cuts diffusion LLM decoding steps by 72%
Researchers introduce LESS, a training-free adaptive sampler for diffusion large language models that treats token commitment as an online stopping problem. The method uses a joint stability rule combining confidence, persistence, and distributional stability to decide when to unmask tokens, avoiding wasted computation on already-stable positions. Evaluated on Dream-7B, LLaDA-8B, and LLaDA-1.5-8B across seven benchmarks, LESS reduces reverse denoising steps by 72.1% versus fixed-budget decoding while improving accuracy over prior adaptive samplers. The step reductions translate directly to fewer Transformer forward passes and lower wall-clock latency.