{"slug": "yalda-embedding-open-development-of-a-competitive-persian-retrieval-model", "title": "Yalda-Embedding: Open Development of a Competitive Persian Retrieval Model", "summary": "A 595.8-million-parameter Persian retrieval model, Yalda Embedding, adapted from Qwen3-Embedding-0.6B using more than one million query-document pairs, achieved the highest macro-average of 0.5488 among evaluated 0.6B models across eight Persian retrieval tasks from FaMTEB, according to an arXiv paper (2610.32933v1). Yalda Embedding outperformed Qwen3-Embedding-0.6B and F2LLM-v2-0.6B on every task and approached Qwen3-Embedding-4B, winning on four tasks, tying on one, and trailing by 0.0075 in macro-average. Relational knowledge distillation from F2LLM-v2-8B raised the mean from 0.4813 to 0.5488 across all eight tasks without increasing inference-time model size, and the authors released model weights, training and evaluation code, the pairwise dataset, and its teacher-embedded counterpart.", "body_md": "arXiv:2610.32933v1 Announce Type: new \nAbstract: Despite advances in multilingual text representations, effective dense retrieval for Persian remains constrained by the limited availability of language-specific training resources and compact embedding models. We present Yalda Embedding, a 595.8-million-parameter Persian retrieval model adapted from Qwen3-Embedding-0.6B using more than one million query-document pairs. On eight Persian retrieval tasks from FaMTEB, the model achieves the highest macro-average among the evaluated 0.6B models, scoring 0.5488 and outperforming Qwen3-Embedding-0.6B and F2LLM-v2-0.6B on every task. It also approaches Qwen3-Embedding-4B, winning on four tasks, tying on one, and trailing by 0.0075 in macro-average. An incremental ablation study examines contrastive batch size, optimization stabilization, and relational knowledge distillation from F2LLM-v2-8B. Under the one-epoch training setup, contrastive fine-tuning alone yields a modest mean improvement over initialization and regressions on two tasks. The final distilled configuration improves all eight tasks over the non-distilled model, raising the mean from 0.4813 to 0.5488 without increasing inference-time model size. These gains include hard-negative evaluation tasks despite training without explicitly mined or supplied hard negatives. We release model weights, training and evaluation code, the pairwise dataset, and its teacher-embedded counterpart, enabling inspection of data composition and training procedures, independent evaluation, and reuse without repeating expensive teacher inference.", "url": "https://wpnews.pro/news/yalda-embedding-open-development-of-a-competitive-persian-retrieval-model", "canonical_source": "https://www.machinebrief.com/news/yalda-embedding-open-development-of-a-competitive-persian-re-ermk", "published_at": "2026-10-26 04:00:00+00:00", "updated_at": "2026-10-26 05:42:15.463106+00:00", "lang": "en", "topics": ["natural-language-processing", "ai-research", "machine-learning", "artificial-intelligence"], "entities": ["Yalda Embedding", "Qwen3-Embedding-0.6B", "F2LLM-v2-0.6B", "F2LLM-v2-8B", "Qwen3-Embedding-4B", "FaMTEB", "arXiv"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/yalda-embedding-open-development-of-a-competitive-persian-retrieval-model", "markdown": "https://wpnews.pro/news/yalda-embedding-open-development-of-a-competitive-persian-retrieval-model.md", "text": "https://wpnews.pro/news/yalda-embedding-open-development-of-a-competitive-persian-retrieval-model.txt", "jsonld": "https://wpnews.pro/news/yalda-embedding-open-development-of-a-competitive-persian-retrieval-model.jsonld"}}