Thinking Machines releases open-weight Inkling 2 Mini, a 26.4B looped MoE model that runs on 32GB Macs

Thinking Machines Lab has released Inkling 2 Mini, an open-weight language model with 26.4 billion parameters, under the Apache 2.0 license on Hugging Face. The model uses a Looped Transformer design that reuses its middle layers, delivering the computation of 48 layer applications from 24 stored layers. Its checkpoint is about 20GB, allowing it to run on a Mac with 32GB unified memory or a single GPU such as the NVIDIA DGX Spark. The model combines this looping with a sparse Mixture-of-Experts feed-forward network of 128 experts, activating 8 per token. It accepts text and image inputs and produces text outputs, and it follows the larger 594-billion-parameter Inkling 2 Small released in November.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)
Log in to join the discussion and vote.
Log in