small language model

Simple use of small and medium sized local LLMs for Q-matrix generation

Can AI Help Chemists To Solve NMR?

Chemists read 1 H and 13 C NMR daily, and it is tempting to paste the peak lists into a large language model. Given only those two peak lists for 105 molecules balanced by molecular complexity and compound classes, six frontier LLMs (Claude Opus 4.8, DeepSeek V4 Pro, Gemini 3.1 Pro, GPT-5.5, Grok 4.3, Qwen3.7-Max) and...

Investigating the effect of VR-mediated pragmatics instruction on L2 refusal strategies

Pragmatic competence remains a vital yet often overlooked area in instructed second language learning, particularly in the domain of refusal speech acts. This quasi-experimental study investigated the effect of VR-mediated explicit pragmatics instruction on L2 learners’ refusal strategies. Fifty-two intermediate-leve...

Graph-Enhanced Retrieval-Augmented Generation for Postoperative Patient Question Answering: A Comparative Evaluation of Knowledge Graph, Retrieval-Augmented Generation, and Hybrid GraphRAG Approaches

GraphRAG is a multi-component pipeline that also includes query expansion, a second retrieval stage, evidence fusion, and a monotonic RAG fallback, and is therefore evaluated as a complete hybrid system rather than as graph retrieval in isolation, which reflects the complete hybrid system rather than graph augmentation...

ANI-Gamut: Benchmarking Agent Reliability across the Gamut of Agent-Network Interface Abstractions

ANI-Gamut is presented, a reproducible, open-source playground that exposes the same task at several levels on a single, densely populated brownfield substrate, where many coexisting services share resources and collateral damage actually arises.

MaskVLA: Visual Masking Against Trajectory Overfitting of Vision-Language-Action Model

MaskVLA, a masking-based fine-tuning strategy that randomly masking a small portion of the main camera's visual information leads to the emergence of robust policies, thereby enhancing the model's capability to tackle complex manipulation tasks and improving its generalization performance.

Which Terrain Is Better? Preference Learning with VLM Prototypes for Off-Road Traversability Ranking

TravPro, which converts standard annotations into ordered region pairs and fits a small readout on frozen vision--language model (VLM) patch tokens to these pairs, producing an ordering sensitive to surface condition that a per-class value cannot represent.

A small-distortion geometric model for elasto-plasticity in single crystals driven by the motion of dislocations

A central problem in nonlinear elasto-plasticity theory is to formulate a geometrically nonlinear, large-strain model of elasto-plasticity in single crystals in which plastic flow is driven directly by the motion of dislocations and which allows for a homogenization procedure from discrete dislocation lines to dislocat...

HOIBlender: Blending Lightweight Detection with Vision-Language Priors for Efficient Human-Object Interaction Detection

The results show that lightweight detection, structured spatial-semantic decoding, and deeply integrated vision-language priors can be blended into a single efficient HOI pipeline.

SPHQuant: Efficient extreme low bit weight quantization for Vision-Language Models

SPHQuant is a rotation-free spherical weight-only quantization framework for VLMs that isolates outlier magnitude into the radius while keeping directions bounded and statistically regular and designs a hardware-friendly GEMV kernel that keeps the direction codebook small enough for shared-memory lookup and packs radia...

SKstars at SHROOM: Visions Agreement-Guided Ensembling of Zero-Shot and LoRA-Adapted Vision--Language Models

The results show that combining a large zero-shot model with a smaller adapted model provides a practical framework for multilingual and fine-grained hallucination localization, while also highlighting the difficulty of transferring development-set improvements to hidden test data.

DocMIDE: Learning Multi-Hop Implicit Derivation in Visually Rich Documents

DocMIDE is introduced, a fine-tuning framework that trains compact vision-language models to retrieve visual evidence explicitly before deriving an answer, and constrains generation to a plan-retrieve-derive structure and optimizes it with Group Relative Policy Optimization under a four-component, rule-based reward.