Saidzoda-Kazakh-EmbeddingGemma-300M：单轮微调将哈萨克语检索 recall@1 提升至 78.0%
Saidzoda-Kazakh-EmbeddingGemma-300M: How Far One Epoch of Generated Questions Takes a Small Kazakh Embedder
339M 参数的小模型，用一个 epoch 的合成问答对微调，哈萨克语检索 recall@1 从 60.8% 拉到 78.0%，还压过 OpenAI text-embedding-3-large。做低资源语言检索或想复刻这套合成数据微调流程的工程师都该看看细节。
研究团队基于开源嵌入模型 EmbeddingGemma-300M，用 208,511 对样本微调一个 epoch，得到 Saidzoda-Kazakh-EmbeddingGemma-300M。训练数据包括由自研语言模型针对百科、Wikipedia、新闻和书籍文本生成的 166,263 条哈萨克语问题，每条经第二个模型校验，另加入英语和俄语回放数据。
在 Kazakh Belebele 检索测试上，模型的 recall@1 从原始的 60.8% 升至 78.0%，recall@10 从 76.3% 升至 91.4%。在 MTEB 同一测试中，该模型超过 194 个已公布成绩模型中的 161 个，包括 OpenAI text-embedding-3-large（47.1%）。在 Hugging Face 上 6 个哈萨克语专用嵌入模型的同代码评测中，它以 77.3% 领先其中 4 个，包括两个同规模 E5-base 模型（72.6% 和 69.0%），但落后于总参数量 1,695M 的 kaz-embedder（86.0%）和 BGE-M3 微调版（81.0%）。
模型在乌兹别克语和吉尔吉斯语 Belebele 上也高于原始版本，但未使用任何突厥语对的塔吉克模型获得相近提升，说明这部分增益并非哈萨克语训练特有。此外，哈萨克语 Wikipedia 测试问题与训练问题出自同一生成器家族，因此主要结论以 Belebele 为准。
Saidzoda-Kazakh-EmbeddingGemma-300M: How Far One Epoch of Generated Questions Takes a Small Kazakh Embedder
EmbeddingGemma-300M is a small open embedder, and used as released it finds the right passage first for 60.8% of the Kazakh Belebele retrieval questions. We present Saidzoda-Kazakh-EmbeddingGemma-300M, the same model fully fine-tuned for one epoch on 208,511 pairs: 166,263 Kazakh questions that our own language model wrote about source passages from encyclopedias, Wikipedia, news and books, each checked by a second model, plus English and Russian replay. On Kazakh Belebele, recall@1 rises from 60.8% to 78.0% and recall@10 from 76.3% to 91.4%. Our model scores above 161 of the 194 models with a published MTEB score on the same test, among them OpenAI text-embedding-3-large (47.1%). The other 33 score higher: the ten best are large multi-billion-parameter or closed models, reaching up to 92.2%, and at least 9 are open models of up to 0.6B parameters, the smallest with 339M. We also scored 6 Kazakh-specific embedders shared on Hugging Face with the same code in one run: our model (77.3% there) is ahead of 4 of them, including the two E5-base models of similar size (72.6% and 69.0%), and behind kaz-embedder (86.0%, three towers with 1,695M parameters in total) and a BGE-M3 fine-tune (81.0%); at recall@10 its interval overlaps those of its nearest neighbours on both sides. Our four-language model, trained with a capped share of the same Kazakh pairs, is 2.8 points lower at recall@1, with overlapping recall@10 intervals. The Kazakh model also scores higher than the released model on the Uzbek and Kyrgyz versions of Belebele, but our Tajik model, trained without any Turkic pairs, gains as much there, so these gains are not specific to Kazakh training. Our Kazakh Wikipedia test questions come from the same generator family as the training questions, so we rest the main claims on Belebele.
This report is categorized under AITOP's "Models" desk. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include EmbeddingGemma, MTEB, Belebele, 嵌入模型.
Primary source verification: published directly by arXiv: OpenAI with first-party authority. Cross-source corroboration: 11 independent sources have covered this development (including Pinecone, Artificial Analysis, vLLM), confirming industry-wide signal strength.
