Unlike the first model, the Khazri 2 family is built on modern language-model approaches and the team's own SYNAPSE architecture.
Why a new architecture?
Khazri 1 was an important starting point for Azerbaijani, but its first-generation GPT-2-based design was limited for advanced understanding, consistent answers and efficient scaling.
Khazri 2 is not only about adding parameters. It aims for a better balance of quality and efficiency through modern approaches to attention, activation and information flow.
What does SYNAPSE represent?
SYNAPSE is the name of the architecture direction developed for the Khazri 2 family. Not every technical detail is public at this stage; its core objective is to provide a stronger model foundation for Azerbaijani and local use cases.
Two sizes, one family
The 250M-parameter Khazri 2 targets flagship capabilities, while the 100M-parameter Khazri 2 Mini focuses on tighter compute environments. Both models have open weights on Hugging Face; contact our team for integration support.