AI Briefing#multi-token-prediction
The latest AI and developer news about #multi-token-prediction, with the original source and a short summary.
The latest AI and developer news about #multi-token-prediction, with the original source and a short summary.
A new method called Soft-MTP improves standard MTP by filtering out noisy gradients, resulting in lower perplexity and faster speculative decoding from 167M to 1.2B parameters.
The method removes a noisy gradient component in standard MTP, improving perplexity and speculative decoding speed from 167M to 1.2B parameters.
An adaptive MTP feature that dynamically adjusts MTP depth based on output complexity has been proposed for llama.cpp.
LG AI Research has developed K-EXAONE, a large-scale AI model that maximizes both efficiency and performance.
LG AI Research has developed K-EXAONE, a large-scale AI model that maximizes efficiency and reasoning capability.
LG AI Research has released EXAONE 4.5, a multimodal model that understands text and images simultaneously, as an open-weight model.
Google combined Multi-Token Prediction with the existing Gemini Nano v3 model to solve inference bottlenecks in mobile environments.
A Pull Request has been submitted to the llama.cpp open-source project to support the StepFun 3.5 MTP model.
Native MTP support for Qwen3.6 MLX models has been added to the distributed inference tool exo.
Google AI Edge Gallery has released a new update including Gemma 4's Multi-Token Prediction and Pixel TPU support.
The llama.cpp b9180 release adds MTP support and partial rollback.
MTP (Multi-Token Prediction) support has been officially merged into the llama.cpp project.
llama.cpp added MTP support, and Qwen3.6 MTP GGUF models were released.
z-lab has released the DFlash drafter model and paper for Gemma 4.
Google released an MTP drafter for Gemma 4, claiming up to 3x faster inference.
Google released an MTP drafter for Gemma 4 that boosts inference speed by up to 3x.
Google has released MTP drafters for Gemma 4, boosting inference speed by up to 3x.
llama.cpp has added support for MTP (Multi Token Prediction) as a beta feature.
IK_LLAMA has added Qwen3.5 MTP support, and MTP-preserved GGUFs are required.