4arXiv cs.CL (Computation and Language)·
Model merging as training-free strategy for conversational information retrieval
A new arXiv preprint proposes using model merging (Model Soup and Slerp) as a training-free alternative to fine-tuning for conversational information retrieval, avoiding catastrophic forgetting of ad-hoc retrieval capabilities. Experiments on standard ad-hoc and conversational retrieval benchmarks show up to 15% higher NDCG@3 under zero-shot conditions. The approach enables a single retrieval model to operate across both ad-hoc and conversational settings without retraining.
Related guides (1)
Related events (8)
Steerable Model Merging (ST-Merge) improves multilingual reasoning via adaptive gated cross-attention
Researchers propose ST-Merge, a framework for adaptively merging a multilingual model and a reasoning model using a gated cross-attention mechanism that weights each source model's contribution based on input characteristics. The approach addresses the limitation of static one-size-fits-all merging strategies that fail to resolve conflicts between source models. Experiments across 21 languages on four multilingual reasoning benchmarks show consistent improvements over strong baselines.
Training-free mixture-of-agents framework combines LLMs and knowledge graphs for multi-document summarization
A new arXiv preprint proposes a training-free multi-agent framework for multi-document summarization (MDS) that decomposes the task into specialized agents for extractive selection, knowledge-aware abstraction, and iterative refinement, unified via a multi-perspective consistency mechanism. The system integrates LLMs with knowledge graphs without task-specific fine-tuning. Experiments across four datasets in English and Vietnamese show state-of-the-art or competitive performance, with the authors emphasizing cross-domain and cross-lingual generalization.
ConvMemory v2: Recall-preserving cross-encoder reranker for conversational memory retrieval
ConvMemory v2 is a fine-tuned cross-encoder reranker (22M parameters, based on ms-marco-MiniLM-L-6-v2) that reorders the top-10 candidates from the prior ConvMemory v1 system without changing which memories are retrieved, preserving Recall@10 by construction. On the LoCoMo conversational memory benchmark, v2 raises MRR from 0.5824 to 0.6560 and Hit@1 from 0.4440 to 0.5474, closing most of the gap to a much more expensive full-pool cross-encoder baseline. An ablation study confirms that candidate-specific memory text is the key mechanism driving the improvement.
M³Exam: Benchmark for Multimodal Memory in Realistic User-Agent Interactions
Researchers introduce M³Exam, a query-centric multimodal conversational memory benchmark designed to evaluate language agents on realistic user-agent interactions, including cross-modal grounding and implicit information inference. Existing benchmarks are critiqued for assuming sparse visuals and human-human interaction formats. The paper also proposes M³Proctor, a companion memory method that detects query modality bias and retrieves raw visual sources on demand, achieving 13% accuracy improvement while reducing index-construction time and retrieved tokens by over 70%.
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.
Training and Finetuning Multimodal Embedding & Reranker Models with Sentence Transformers
Hugging Face published a blog post detailing how to train and finetune multimodal embedding and reranker models using the Sentence Transformers library. The post covers techniques for building models that can jointly embed text and images for retrieval and reranking tasks. This represents an extension of the Sentence Transformers ecosystem into multimodal territory, enabling practitioners to build cross-modal search and ranking systems.
Training and Finetuning Reranker Models with Sentence Transformers
Hugging Face published a tutorial on training and fine-tuning reranker models using the Sentence Transformers library. Rerankers are cross-encoder models used in retrieval-augmented generation (RAG) and search pipelines to re-score candidate documents for improved relevance. The post covers dataset preparation, loss functions, and training configurations specific to reranking tasks.
RL-based alignment improves interactivity in full-duplex spoken dialogue models
Researchers propose a post-training alignment method using reinforcement learning to improve interactivity in full-duplex spoken dialogue models, which can listen and speak simultaneously. The method addresses four canonical axes of interactivity—pause handling, turn-taking, backchanneling, and user interruption—each with axis-specific reward functions, plus an LLM-based reward to prevent semantic degradation. The approach is applied to two open-source models, Moshi and PersonaPlex, showing consistent improvements in both offline and real-time multi-turn evaluation.