AI Briefing#mixture-of-experts
The latest AI and developer news about #mixture-of-experts, with the original source and a short summary.
The latest AI and developer news about #mixture-of-experts, with the original source and a short summary.
Meta has released MobileMoE, a MoE model with 0.9B active parameters optimized for mobile devices.
The ToMoE technique has been released, converting dense LLMs into MoE via dynamic structural pruning to reduce active parameters without performance degradation.
llama.cpp now supports dots3-note, the first open-weight model in the dots3 family.
This analysis identifies the threshold at which communication costs limit computational performance during data and tensor parallelization of transformer models.
Ornith-1.5 has been released as a series of open models ranging from 397B to 9B parameters.
Thinking Machines released its first self-developed model, Inkling, under the Apache 2.0 license.
By leveraging a Mixture-of-Experts (MoE) architecture, a 397B-parameter AI model has been successfully run locally on an iPhone.
Microsoft has released MAI-Thinking-1, a high-performance reasoning model optimized for enterprise tasks.
Qwen has released the open model Qwen3.8 with a total of 2.4T parameters and 95B active parameters.
Qwen has released the Qwen3.8 open model with a total size of 2.4T and 95B active parameters.
WARP has implemented local inference for the 2.78-trillion-parameter Kimi K3 on a 64GB MacBook.
NVIDIA released Nemotron 3.5 Lightning, a 30B MoE model with 3B active parameters for low-latency AI agent tasks.
NVIDIA is partnering with the local AI community to expand the ecosystem of open models and agents.
NVIDIA has released a high-efficiency agentic AI model and an intelligent model routing library.
Cursor released a fused MoE communication and computation kernel for NVL72, reporting a 1.41x throughput improvement.
Multimodal models developed more effectively when modalities were trained together from the beginning.
Ant Group has released Ling-3.0-flash, a 124B model, under an unrestricted MIT license.
InclusionAI released a 124B MoE model with 5.1B active parameters.
Cursor released an open-source MoE megakernel for NVL72, boosting throughput by up to 2.37x.
DiffusionGemma accelerates text generation by iteratively refining 256 tokens in parallel.
AirLLM has revealed a memory-saving method that runs a 70B model on a 4GB GPU.
AI9Stars has released a 39B open-weight language model based on Apache 2.0.
WASTE streams activated expert weights from NVMe, enabling Kimi K3 to run larger than RAM.
WASTE ran the 2.78T Kimi K3 via disk streaming on a 64GB MacBook Pro.