Large language models (LLMs) have been widely deployed across numerous fields. Reinforcement Learning from Human Feedback (RLHF) leverages reward models (RMs) as proxies for human preferences to align LLM behaviors with human values, making the accuracy, reliability, and interpretability of RMs critical for effective alignment. However, traditional RMs lack interpretability, offer limited insight into the reasoning behind reward assignments, and are inflexible toward user preference shifts. While recent multidimensional RMs aim for improved interpretability, they often fail to provide feature-level attribution and require costly annotations. To overcome these limitations, we introduce the Sparse Autoencoder-Enhanced Reward Model (SARM), a novel architecture that integrates a pretrained Sparse Autoencoder (SAE) into a reward model. SARM maps the hidden activations of LLM-based RM into an interpretable, sparse, and monosemantic feature space, from which a scalar head aggregates feature activations to produce transparent and conceptually meaningful reward scores. Empirical evaluations demonstrate that SARM facilitates direct feature-level attribution of reward assignments, allows dynamic adjustment to preference shifts, and achieves superior alignment performance compared to conventional reward models.

Hengxing Cai, Jinhan Dong, Jingjun Tan, Jingcheng Deng, Sihang Li, Zhifeng Gao, Haidong Wang, Zicheng Su, Agachai Sumalee, Renxin Zhong. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Mechanistic interpretability of large language models (LLMs) aims to uncover the internal processes of information propagation and reasoning.Sparse autoencoders (SAEs) have demonstrated promise in this domain by extracting interpretable and monosemantic features.However, prior works primarily focus on feature extraction from a single layer, failing to effectively capture activations that span multiple layers.In this paper, we introduce Route Sparse Autoencoder (RouteSAE), a new framework that integrates a routing mechanism with a shared SAE to efficiently extract features from multiple layers.It dynamically assigns weights to activations from different layers, incurring minimal parameter overhead while achieving high interpretability and flexibility for targeted feature manipulation.We evaluate RouteSAE through extensive experiments on Llama-3.2-1B-Instruct.Specifically, under the same sparsity constraint of 64, RouteSAE extracts 22.5% more features than baseline SAEs while achieving a 22.3% higher interpretability score.These results underscore the potential of RouteSAE as a scalable and effective method for LLM interpretability, with applications in feature discovery and model intervention.Our codes are available at https: //github.com/swei2001/RouteSAEs.

Scientific literature understanding is crucial for extracting targeted information and garnering insights, thereby significantly advancing scientific discovery. Despite the remarkable success of Large Language Models (LLMs), they face challenges in scientific literature understanding, primarily due to (1) a lack of scientific knowledge and (2) unfamiliarity with specialized scientific tasks. To develop an LLM specialized in scientific literature understanding, we propose a hybrid strategy that integrates continual pre-training (CPT) and supervised fine-tuning (SFT), to simultaneously infuse scientific domain knowledge and enhance instruction-following capabilities for domain-specific tasks. In this process, we identify two key challenges: (1) constructing high-quality CPT corpora, and (2) generating diverse SFT instructions. We address these challenges through a meticulous pipeline, including PDF text extraction, parsing content error correction, quality filtering, and synthetic instruction creation. Applying this strategy, we present a suite of LLMs: SciLitLLM, specialized in scientific literature understanding. These models demonstrate promising performance on scientific literature understanding benchmarks. (1) We present an effective framework that integrates CPT and SFT to adapt LLMs to scientific literature understanding, which can also be easily adapted to other domains. (2) We propose an LLM-based synthesis method to generate diverse and high-quality scientific instructions, resulting in a new instruction set -- SciLitIns -- for less-represented scientific domains. (3) SciLitLLM achieves promising performance in scientific literature understanding benchmarks.

3D molecule generation is crucial for drug discovery and material design. While prior efforts focus on 3D diffusion models for their benefits in modeling continuous 3D conformers, they overlook the advantages of 1D SELFIES-based Language Models (LMs), which can generate 100\% valid molecules and leverage the billion-scale 1D molecule datasets. To combine these advantages for 3D molecule generation, we propose a foundation model -- NExT-Mol: 3D Diffusion Meets 1D Language Modeling for 3D Molecule Generation. NExT-Mol uses an extensively pretrained molecule LM for 1D molecule generation, and subsequently predicts the generated molecule's 3D conformers with a 3D diffusion model. We enhance NExT-Mol's performance by scaling up the LM's model size, refining the diffusion neural architecture, and applying 1D to 3D transfer learning. Notably, our 1D molecule LM significantly outperforms baselines in distributional similarity while ensuring validity, and our 3D diffusion model achieves leading performances in conformer prediction. Given these improvements in 1D and 3D modeling, NExT-Mol achieves a 26\% relative improvement in 3D FCD for de novo 3D generation on GEOM-DRUGS, and a 13\% average relative gain for conditional 3D generation on QM9-2014. Our codes and pretrained checkpoints are available at https://github.com/acharkq/NExT-Mol.

3D molecule generation is crucial for drug discovery and material science, requiring models to process complex multi-modalities, including atom types, chemical bonds, and 3D coordinates. A key challenge is integrating these modalities of different shapes while maintaining SE(3) equivariance for 3D coordinates. To achieve this, existing approaches typically maintain separate latent spaces for invariant and equivariant modalities, reducing efficiency in both training and sampling. In this work, we propose **U**nified Variational **A**uto-**E**ncoder for **3D** Molecular Latent Diffusion Modeling (**UAE-3D**), a multi-modal VAE that compresses 3D molecules into latent sequences from a unified latent space, while maintaining near-zero reconstruction error. This unified latent space eliminates the complexities of handling multi-modality and equivariance when performing latent diffusion modeling. We demonstrate this by employing the Diffusion Transformer--a general-purpose diffusion model without any molecular inductive bias--for latent generation. Extensive experiments on GEOM-Drugs and QM9 datasets demonstrate that our method significantly establishes new benchmarks in both *de novo* and conditional 3D molecule generation, achieving leading efficiency and quality. On GEOM-Drugs, it reduces FCD by 72.6% over the previous best result, while achieving over 70% relative average improvements in geometric fidelity. Our code is released at [https://github.com/lyc0930/UAE-3D/](https://github.com/lyc0930/UAE-3D/).

Large language models have demonstrated impressive reasoning capabilities but are inherently limited by their knowledge reservoir. Retrieval-augmented reasoning mitigates this limitation by allowing LLMs to query external resources, but existing methods often retrieve irrelevant or noisy information, hindering accurate reasoning. In this paper, we propose **AutoRefine**, a reinforcement learning post-training framework that adopts a new "search-and-refine-during-think" paradigm. AutoRefine introduces explicit knowledge refinement steps between successive search calls, enabling the model to iteratively filter, distill, and organize evidence before generating an answer. Furthermore, we incorporate tailored retrieval-specific rewards alongside answer correctness rewards using group relative policy optimization. Experiments on single-hop and multi-hop QA benchmarks demonstrate that AutoRefine significantly outperforms existing approaches, particularly in complex, multi-hop reasoning scenarios. Detailed analysis shows that AutoRefine issues frequent, higher-quality searches and synthesizes evidence effectively.

Large Language Models (LLMs) have emerged as a new paradigm for recommendation by converting interacted item history into language modeling. However, constrained by the limited context length of LLMs, existing approaches have to truncate item history in the prompt, focusing only on recent interactions and sacrificing the ability to model long-term history. To enable LLMs to model long histories, we pursue a concise embedding representation for items and sessions. In the LLM embedding space, we construct an item's embedding by aggregating its textual token embeddings; similarly, we construct a session's embedding by aggregating its item embeddings. While efficient, this way poses two challenges since it ignores the temporal significance of user interactions and LLMs do not natively interpret our custom embeddings. To overcome these, we propose PatchRec, a multi-grained patch training method consisting of two stages: (1) Patch Pre-training, which familiarizes LLMs with aggregated embeddings -- patches, and (2) Patch Fine-tuning, which enables LLMs to capture time-aware significance in interaction history. Extensive experiments show that PatchRec effectively models longer behavior histories with improved efficiency. This work facilitates the practical use of LLMs for modeling long behavior histories.

Graph augmentation is the key component to reveal instance-discriminative features of a graph as its rationale—an interpretation for it—in graph contrastive learning (GCL). Existing rationale-aware augmentation mechanisms in GCL frameworks roughly fall into two categories and suffer from inherent limitations: (1) non-heuristic methods with the guidance of domain knowledge to preserve salient features, which require expensive expertise and lack generality, or (2) heuristic augmentations with a co-trained auxiliary model to identify crucial substructures, which face not only the dilemma between system complexity and transformation diversitybut also the instability stemming from the co-training of two separated sub-models. Inspired by recent studies on transformers, we propose self-attentive rationale-guided GCL (SR-GCL), which integrates rationale generator and encoder together, leverages the self-attention values in transformer module as a natural guidance to delineate semantically informative substructures from both node- and edge-wise perspectives, and contrasts on rationale-aware augmented pairs. On real-world biochemistry datasets, visualization results verify the effectiveness and interpretability of self-attentive rationalization, and the performance on downstream tasks demonstrates the state-of-the-art performance of SR-GCL for graph model pre-training. Codes are available at https://github.com/lsh0520/SR-GCL .

Masked modeling has recently achieved remarkable success in specific fields of vision and language, sparking a surge of interest in graph-related research. However, Masked Graph Modeling (MGM), which captures fine-grained local information by masking low-level elements such as nodes, edges, and features, limits itself to a sub-optimal position, particularly on tasks requiring high-quality graph-level representations. Such a local perspective disregards the graph's global information and structure. To address these limitations, we propose a novel graph pre-training framework called Graph Contrastive Masked Autoencoder (GCMAE). GCMAE leverages the strengths of both MGM and Graph Contrastive Learning (GCL) to provide a more comprehensive perspective of both local and global. Our frame-work uses instance discrimination to learn global representations of graphs and reconstructs the graph using masked low-level elements. We augment the framework with a novel multi-view augmentation module to further enhance the pre-trained model's robustness and generalization ability. We evaluate GCMAE on real-world biochemistry and social network datasets, conducting extensive experiments on both node and graph classification tasks and transfer learning on downstream graph classification tasks. Our experimental results demonstrate that GCMAE's comprehensive perspective of both local and global benefits model pre-training. Moreover, GCMAE outperforms existing MGM and GCL baselines, proving its effectiveness on downstream tasks. Our code is available at https://github.com/lyc0930/GCMAE.

Language Models (LMs) have greatly influenced diverse domains. However, their inherent limitation in comprehending 3D molecular structures has considerably constrained their potential in the biomolecular domain. To bridge this gap, we focus on 3D molecule-text interpretation, and propose 3D-MoLM: 3D-Molecular Language Modeling. Specifically, 3D-MoLM enables an LM to interpret and analyze 3D molecules by equipping the LM with a 3D molecular encoder. This integration is achieved by a 3D molecule-text projector, bridging the 3D molecular encoder’s representation space and the LM’s input space. Moreover, to enhance 3DMoLM’s ability of cross-modal molecular understanding and instruction following, we meticulously curated a 3D molecule-centric instruction tuning dataset – 3D-MoIT. Through 3D molecule-text alignment and 3D molecule-centric instruction tuning, 3D-MoLM establishes an integration of 3D molecular encoder and LM. It significantly surpasses existing baselines on downstream tasks, including moleculetext retrieval, molecule captioning, and more challenging open-text molecular QA tasks, especially focusing on 3D-dependent properties. We will release our codes and datasets at https://github.com/lsh0520/3D-MoLM.

Sequential recommendation aims to predict users' next interaction with items based on their past engagement sequence. Recently, the advent of Large Language Models (LLMs) has sparked interest in leveraging them for sequential recommendation, viewing it as language modeling. Previous studies represent items within LLMs' input prompts as either ID indices or textual metadata. However, these approaches often fail to either encapsulate comprehensive world knowledge or exhibit sufficient behavioral understanding. To combine the complementary strengths of conventional recommenders in capturing behavioral patterns of users and LLMs in encoding world knowledge about items, we introduce Large Language-Recommendation Assistant (LLaRA). Specifically, it uses a novel hybrid prompting method that integrates ID-based item embeddings learned by traditional recommendation models with textual item features. Treating the "sequential behaviors of users" as a distinct modality beyond texts, we employ a projector to align the traditional recommender's ID embeddings with the LLM's input space. Moreover, rather than directly exposing the hybrid prompt to LLMs, a curriculum learning strategy is adopted to gradually ramp up training complexity. Initially, we warm up the LLM using text-only prompts, which better suit its inherent language modeling ability. Subsequently, we progressively transition to the hybrid prompts, training the model to seamlessly incorporate the behavioral knowledge from the traditional sequential recommender into the LLM. Empirical results validate the effectiveness of our proposed framework. Codes are available at https://github.com/ljy0ustc/LLaRA.

Language Models (LMs) have demonstrated impressive molecule understanding ability on various 1D text-related tasks.However, they inherently lack 2D graph perception -a critical ability of human professionals in comprehending molecules' topological structures.To bridge this gap, we propose MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter.MolCA enables an LM (i.e., Galactica) to understand both text-and graph-based molecular contents via the cross-modal projector.Specifically, the cross-modal projector is implemented as a Q-Former to connect a graph encoder's representation space and an LM's text space.Further, MolCA employs a uni-modal adapter (i.e., LoRA) for the LM's efficient adaptation to downstream tasks.Unlike previous studies that couple an LM with a graph encoder via cross-modal contrastive learning, MolCA retains the LM's ability of open-ended text generation and augments it with 2D graph information.To showcase its effectiveness, we extensively benchmark MolCA on tasks of molecule captioning, IUPAC name prediction, and molecule-text retrieval, on which MolCA significantly outperforms the baselines.Our codes and checkpoints can be found at https: //github.com/acharkq/MolCA.

Leading graph contrastive learning (GCL) methods perform graph augmentations in two fashions: (1) randomly corrupting the anchor graph, which could cause the loss of semantic information, or (2) using domain knowledge to maintain salient features, which undermines the generalization to other domains. Taking an invariance look at GCL, we argue that a high-performing augmentation should preserve the salient semantics of anchor graphs regarding instance-discrimination. To this end, we relate GCL with invariant rationale discovery, and propose a new framework, Rationale-aware Graph Contrastive Learning (RGCL). Specifically, without supervision signals, RGCL uses a rationale generator to reveal salient features about graph instance-discrimination as the rationale, and then creates rationale-aware views for contrastive learning. This rationale-aware pre-training scheme endows the backbone model with the powerful representation ability, further facilitating the fine-tuning on downstream tasks. On MNIST-Superpixel and MUTAG datasets, visual inspections on the discovered rationales showcase that the rationale generator successfully captures the salient features (\ie distinguishing semantic nodes in graphs). On biochemical molecule and social network benchmark datasets, the state-of-the-art performance of RGCL demonstrates the effectiveness of rationale-aware views for contrastive learning. Our codes are available at https://github.com/lsh0520/RGCL.