ConceptMTP / Multi-Token Prediction
Multi-Token Prediction
A language model decoding technique that allows the model to predict multiple tokens in a single forward pass. It is commonly combined with speculative decoding to improve inference efficiency.
Source Articles
Adaptive Multi-Token Prediction: How Does the Speculative Horizon Dynamically Adjust for Inference Acceleration?Oct 21BitMTP: Co-Accelerating Multi-Token Prediction and Low-Bit LLMsOct 21Lafa: An 8B-Parameter Continuous Flow Language Model with 2.54x Throughput Over AR ModelsOct 2112GB VRAM Running a 180B Model? A Deep Dive into the Strata Inference EngineOct 5IQuest-Q1 Deep Dive: How a 320B MoE Model Is Reshaping Agentic CodingOct 4