Edoardo M. Ponti

Latest

• AdaSplash-2: Faster Differentiable Sparse Attention
• Self-Improving World Modelling with Latent Actions
• Bolmo: Byteifying the Next Generation of Language Models
• Fast and Expressive Multi-Token Prediction with Probabilistic Circuits
• Inference-Time Hyper-Scaling with KV Cache Compression
• Bootstrapping Action-Grounded Visual Dynamics in Unified Vision-Language Models
• The Sparse Frontier: Sparse Attention Trade-offs in Transformer LLMs
• Zero-Shot Tokenizer Transfer
• Dynamic Memory Compression: Retrofitting LLMs for Accelerated Inference
• Efficient Transformers with Dynamic Token Pooling