Luke Zettlemoyer

Reconstruction Alignment Improves Unified Multimodal Models

Latent Speech-Text Transformer

High Accuracy, Less Talk (HALT): Reliable LLMs through Capability-Aligned Finetuning

OneFlow: Concurrent Mixed-Modal and Interleaved Generation with Edit Flows

HybridCoT: Interleaving Latent and Text Chain-of-Thought for Efficient Reasoning

Spurious Rewards: Rethinking Training Signals in RLVR

Train for Truth, Keep the Skills: Binary Retrieval-Augmented Reward Mitigates Hallucinations

Mixture-of-Mamba: Enhancing Multi-Modal State-Space Models with Modality-Aware Sparsity

Continual Learning via Sparse Memory Finetuning

FlexOLMo: Open Language Models for Flexible Data Use

When Worse is Better: Navigating the Compression Generation Trade-off In Visual Tokenization

Meta CLIP 2: A Worldwide Scaling Recipe

Transfusion: Predict the Next Token and Diffuse Images with One Multi-Modal Model

Memory Layers at Scale

Heterogeneous Swarms: Jointly Optimizing Model Roles and Weights for Multi-LLM Systems

CAT: Content-Adaptive Image Tokenization

ReasonIR: Training Retrievers for Reasoning Tasks

2 OLMo 2 Furious (COLM’s Version)

LMFusion: Adapting Pretrained Language Models for Multimodal Generation

Precise Information Control in Long-Form Text Generation

MUSE: Machine Unlearning Six-Way Evaluation for Language Models

Recycling the Web: A Method to Enhance Pre-training Data Quality and Quantity for Language Models

Latent Action Pretraining from Videos

(Mis)Fitting Scaling Laws: A Survey of Scaling Law Fitting Techniques in Deep Learning

Generative Adapter: Contextualizing Language Models in Parameters with A Single Forward Pass

ParaPO: Aligning Language Models to Reduce Verbatim Reproduction of Pre-training Data

Fantastic Copyrighted Beasts and How (Not) to Generate Them

MoMa: Efficient Early-Fusion Pre-training with Mixture of Modality-Aware Experts