Thinking Machines publishes Inkling 2 Small's weights on Hugging Face under Apache 2.0

Thinking Machines says the weights of Inkling 2 Small can now be downloaded from Hugging Face under an Apache 2.0 license, the release it had promised when it launched the model on November 5 as proprietary.

Songlin Yang, who posted soon after, lists changes to the architecture since the previous generation: attention residuals with a q/k/v split, gated attention, a dynamic attention sink, dynamic shortconv and value norm. These are the poster's own description.