Natural Language Processing

Fietje: An open, efficient LLM for Dutch

AceMath: Advancing Frontier Math Reasoning with Post-Training and Reward Modeling

TheAgentCompany: Benchmarking LLM Agents on Consequential Real World Tasks

Smarter, Better, Faster, Longer: A Modern Bidirectional Encoder for Fast, Memory Efficient, and Long Context Finetuning and Inference

RAG-RewardBench: Benchmarking Reward Models in Retrieval Augmented Generation for Preference Alignment

AntiLeak-Bench: Preventing Data Contamination by Automatically Constructing Benchmarks with Updated Real-World Knowledge

DateLogicQA: Benchmarking Temporal Biases in Large Language Models

Whisper-GPT: A Hybrid Representation Audio Large Language Model

The Open Source Advantage in Large Language Models (LLMs)

SPaR: Self-Play with Tree-Search Refinement to Improve Instruction-Following in Large Language Models

SepLLM: Accelerate Large Language Models by Compressing One Segment into One Separator

RetroLLM: Empowering Large Language Models to Retrieve Fine-grained Evidence within Generation

Smaller Language Models Are Better Instruction Evolvers

Reliable, Reproducible, and Really Fast Leaderboards with Evalica

SCBench: A KV Cache-Centric Analysis of Long-Context Methods

Large Action Models: From Inception to Implementation

Byte Latent Transformer: Patches Scale Better Than Tokens

Phi-4 Technical Report

GReaTer: Gradients over Reasoning Makes Smaller Language Models Strong Prompt Optimizers

SmolTulu: Higher Learning Rate to Batch Size Ratios Can Lead to Better Reasoning in SLMs