Abstract: In this paper, we focus on monolithic Multimodal Large Language Models (MLLMs) that integrate visual encoding and language decoding into a single LLM. In particular, we identify that ...

This is the official repository of paper Distilled Decoding 1: One-step Sampling of Image Auto-regressive Models with Flow Matching. We propose Distilled Decoding (DD) to distill a pre-trained image ...

Abstract: Speculative decoding has proven to be a powerful method for speeding up autoregressive inference by allowing tokens to be generated in parallel using a draft-then-verify approach. However, ...