AI Briefing#mtp
The latest AI and developer news about #mtp, with the original source and a short summary.
The latest AI and developer news about #mtp, with the original source and a short summary.
ik_llama.cpp has officially merged MTP support for Qwen3.8-Flash-Next, resulting in approximately 2x faster inference speeds on RTX 5090 environments.
An adaptive MTP feature that dynamically adjusts MTP depth based on output complexity has been proposed for llama.cpp.
llama.cpp has been changed to load MTP tensors for certain architectures by default even when MTP is disabled, increasing VRAM usage.
LG AI Research has unveiled EXAONE 4.5, an open-weight VLM that processes visual information and language simultaneously.
Xiaomi has released the MiMo-V2.5-DFlash model weights, which have a large number of parameters, via Hugging Face.
MTP heads for Gemma 4 QAT models have been released, along with news of a fix for a llama.cpp parallel execution bug.
A PR applying post-norm hidden states to improve Qwen model MTP performance has been submitted to llama.cpp.
A Pull Request has been submitted to the llama.cpp open-source project to support the StepFun 3.5 MTP model.
A Qwen3.6 GGUF model has been released that integrates an MTP head, enabling self-speculative decoding without a separate draft model.
llama.cpp has applied an update that reduces VRAM usage by processing the KQ mask in f16 when using Flash Attention.
By applying MTP to the Qwen3.6-35B-A3B model, a decoding speed improvement of about 1.49x was achieved on llama-server.
llama.cpp now supports NVIDIA FP4 quantization and Multi-Token Prediction (MTP) features.
An update has been proposed in llama.cpp to improve MTP performance by moving the draft path to backend sampling.
LM Studio added support for MTP speculative decoding in 0.4.14 Beta.
Bug fixes related to MTP and speculative decoding improvements have been merged into llama.cpp.
llama.cpp b9200 includes an optimization that improves MTP prompt processing speed.
An update has been released for llama.cpp that optimizes prompt decoding performance when using MTP.
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.
llama.cpp is undergoing updates to resolve conflicts between MTP and multimodal features and support their integration.
Unsloth has released GGUF versions of Qwen models with Multi-Token Prediction (MTP) technology applied.
A method for optimizing Qwen3.6 inference performance by grafting an MTP layer onto an Unsloth quantized model is introduced.
MTP support in llama.cpp can boost inference speed for the Qwen 3.6 27B model by 2.5x.
Google excluded Gemma 4's MTP feature from the public release, then began supporting it via an auxiliary model after the community discovered it.