Content Deep Dive

Ayb: Language models for Armenian

Blog post from Hugging Face

Post Details
Company
Date Published
Author
Spartak Bughdaryan, Zaruhi Navasardyan, and Hrant Davtyan
Word Count
1,249
Company Posts That Month
68
Language
-
Hacker News Points
-
Post removed?
No
Summary

Ayb 0.5 is an open-source family of Armenian-language models intended to strengthen access to AI for Armenian speakers and support a more independent Armenian AI ecosystem. Built on the Qwen3.5 architecture, the three models—0.8B, 4B, and 9B—were adapted using an Armenian-specific tokenizer, continued Armenian and English pretraining, instruction tuning, and post-training; they range from lightweight text processing to conversational and tool-using applications. In zero-shot evaluations, Ayb-9B ranked third on ArmBench with a score of 0.6164, while the 4B and 0.8B models ranked eighth and sixteenth; the family performed relatively well in reading comprehension, text generation, and correction, but showed room to improve in factual knowledge, complex reasoning, and instruction following. The models can also handle Armenian image questions because the release retains Qwen3.5’s vision components, and examples demonstrate source-based answering, correction of false premises, and cautious responses to a medical question. Their reliability is weaker in Western Armenian, regional dialects, and Latin-script Armenian, and recent-event coverage is uneven; the developers identify reasoning, preference tuning, and support for Armenian variants as priorities for future work.

Trends Found in this Post
Trend Post Mentions Total Month Mentions Posts Companies MoM
LLM 1 3,674 823 199 -44%
Use This Data

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.