Apple researchers showed distributed LLM inference across Mac Studios, with related small-model and tool-learning papers appearing the same day.
Apple Machine Learning Research published work on running large LLMs across multiple Mac Studio machines using MLX and RDMA. The same day brought a design-space study of tri-modal masked diffusion models, and arXiv papers on tool learning for small language models and agent-to-agent negotiation.
Large-model inference is starting to look feasible beyond datacenter GPUs, on clustered high-end desktop machines. Alongside work on training small models from teachers, a push toward lighter, local execution overlapped within the same week. These are research reports, though, and real-world adoption is not yet confirmed.
Watch whether MLX distributed inference becomes available as public tooling or a consumer feature, and whether further results on effective performance, including interconnect bandwidth, follow.