DREAM: Training dense retrieval embeddings via autoregressive LLM supervision
DREAM is a new method for training dense retrieval embedding models using the autoregressive next-token prediction objective of a frozen LLM, bypassing the need for labeled positive/negative document pairs required by contrastive training. The approach injects retriever-generated query-document similarity scores into selected attention heads of the LLM, allowing prediction loss gradients to flow back to the retriever. Evaluated on BEIR and RTEB benchmarks with 0.5B–3B parameter backbones, DREAM consistently outperforms contrastive baselines across model scales.
Related guides (2)
Related events (8)
RL-trained LLMs learn retriever-specific query formulation strategies for RAG
A new arXiv paper presents the first systematic study of using reinforcement learning to teach LLMs to adapt query formulation strategies to different retrieval backends. The authors find that different retrievers have surprisingly distinct optimal query styles (e.g., descriptive vs. question-like), making cross-retriever strategy transfer ineffective. They introduce a branching-based rollout technique to stabilize training over multi-step retrieval trajectories and show gains from retriever-specific human guidance and model scaling.
Sleep paradigm for LLMs enables continual learning and memory consolidation via distillation and RL
A new arXiv preprint proposes a 'Sleep' paradigm for language models that enables continual learning by consolidating short-term in-context memories into long-term parameters. The framework has two stages: Knowledge Seeding (distilling a smaller model's memories into a larger network via on-policy distillation combined with RL-based imitation learning) and Dreaming (self-improvement via RL-generated synthetic curricula without human supervision). Experiments cover long-horizon tasks, continual learning, knowledge incorporation, and few-shot generalization, addressing a known weakness of current LLMs in retaining temporal knowledge across contexts.
RA-RFT: Retrieval-Augmented Reinforcement Fine-Tuning teaches LLMs to reason by analogy
Researchers propose Retrieval-Augmented Reinforcement Fine-Tuning (RA-RFT), a post-training framework that trains a retriever to rank contexts by expected reasoning benefit rather than semantic similarity, then fine-tunes a policy model via reinforcement learning using retrieved analogous demonstrations. The key insight is that reasoning-relevant retrieval surfaces complementary solution strategies rather than superficially similar problems. On mathematical reasoning benchmarks, RA-RFT improves AIME 2025 average@32 accuracy by 7.1 and 2.8 points over GRPO for Qwen3-1.7B and Qwen3-4B respectively, suggesting reasoning-aware retrieval is orthogonal to reward design and training curriculum improvements.
SARDI: Self-Augmenting Retrieval for Diffusion Language Models using lookahead tokens
Researchers introduce SARDI, a training-free RAG framework for discrete diffusion language models that repurposes discarded low-confidence tokens during denoising as lookahead signals to guide retrieval before output is finalized. The method is retriever-agnostic and applicable to any reasoning-capable discrete diffusion LM. Evaluated across five multi-hop QA benchmarks, SARDI outperforms training-free diffusion and autoregressive retrieval baselines at up to 8x higher throughput.
RELEX: Extrapolating LLM RLVR Training via Rank-1 Parameter Trajectories
This paper demonstrates that RLVR weight update trajectories are extremely low-rank and near-linearly predictable, with a rank-1 approximation capturing most downstream performance gains. The authors propose RELEX, a compute-efficient method that observes a short training window, estimates the rank-1 subspace, and extrapolates future checkpoints via linear regression—requiring no additional training. Evaluated on Qwen2.5-Math-1.5B, Qwen3-4B-Base, and Qwen3-8B-Base, RELEX matches or exceeds full RLVR performance using as few as 15% of training steps, and can extrapolate up to 10–20× beyond the observed prefix. The authors attribute the method's effectiveness to a denoising effect from rank-1 projection that discards stochastic optimization noise.
A Sleep-Like Consolidation Mechanism for LLMs
A preprint on arXiv proposes a sleep-like memory consolidation mechanism for large language models, drawing an analogy to biological sleep-based memory consolidation in neural systems. The work appears to address how LLMs might better retain and integrate new information over time, a key challenge in continual learning and knowledge updating. The paper attracted notable community attention on Hacker News with 164 points and 122 comments, suggesting broad interest in the approach.
MemDreamer: Hierarchical graph memory and agentic retrieval for long video understanding
MemDreamer is a plug-and-play framework that decouples perception and reasoning for long-video understanding by incrementally building a three-tier Hierarchical Graph Memory capturing spatiotemporal and causal relations. During inference, a reasoning model uses an Observation-Reason-Action loop with agentic tool-augmented retrieval to navigate the memory graph, constraining the context window to 2% of full-context ingestion while achieving a 12.5-point absolute accuracy gain. The system reaches SOTA on four benchmarks, narrowing the gap with human experts to 3.7 points. The authors also report a strong linear correlation between logical reasoning performance and long-video understanding, proposing agentic capability scaling as a new paradigm for multimodal comprehension.
ContextRL: Context-aware reinforcement learning improves grounding in agentic and multimodal LLMs
Researchers introduce ContextRL, a reinforcement learning method that trains LLMs to select the context that supports a given query-answer pair from two highly similar candidates, rather than supervising only final answers. The approach constructs contrastive context pairs in two domains: coding agent trajectories (1k pairs) and multimodal image pairs (7k pairs). ContextRL achieves +2.2% average gains over standard GRPO on 5 long-horizon benchmarks and +1.8% across 12 visual QA benchmarks, with ablations showing the gains stem from the context-selection objective rather than the contrastive data alone.