{"slug": "ufakzeka-1-building-and-evaluating-a-151m-parameter-turkish-language-model-from", "title": "ufakzeka-1: Building and Evaluating a 151M-Parameter Turkish Language Model from Scratch", "summary": "Researchers released ufakzeka-1, a 151M-parameter (182M with embeddings) decoder-only Turkish language model pretrained from scratch on 13.5B tokens of openly licensed text and instruction-tuned for chat at a total cost of about $286 in cloud GPU, API and notebook time, according to the arXiv paper 2609.25081v1. The paper reports three transferable findings: a safety gate \"fixed\" with training data written from its own questions read 64/64 while the honest figure was 34/64, training-seed variance matched the spread across every recipe tried, and data rounds repaired only what was absent from the data while identity tracking over long context and multi-turn arithmetic did not move. Weights, the data recipe, evaluation code and spend ledger are released under Apache-2.0.", "body_md": "arXiv:2609.25081v1 Announce Type: new \nAbstract: We describe ufakzeka-1, a 151M-parameter (182M with embeddings) decoder-only Turkish language model pretrained from scratch on 13.5B tokens of openly licensed text and instruction-tuned for chat, at a total cost of about \\$286 in cloud GPU, API and notebook time. The contribution is not the model's capability, which is what a model this size can be expected to have, but the record of building and measuring it: a Turkish byte-level tokenizer at 1.77 tokens per word, a three-stage pretraining schedule, a post-training mixture of openly licensed and generated data, and an evaluation battery of release gates, a rule-checked sweep of 5,508 conversations, judged conversations and hand tests, all with prompts held out from the training data, enforced by decontamination inside the data build and by a checked-in invariant script we run before each build. We report three findings that we believe transfer to other small-model efforts: a safety gate that had been \"fixed\" with training data written from its own questions read 64/64 while the honest figure was 34/64; training-seed variance was as large as the spread across every recipe we tried, so single-seed comparisons at this scale are uninformative; and data rounds repaired only what was absent from the data, while identity tracking over long context and multi-turn arithmetic did not move across any data change we tried, which we read as limits of the model size rather than gaps in the data, a reading the next, larger model will test. Weights, the data recipe, the evaluation code and the spend ledger are released under Apache-2.0.", "url": "https://wpnews.pro/news/ufakzeka-1-building-and-evaluating-a-151m-parameter-turkish-language-model-from", "canonical_source": "https://arxiv.org/abs/2609.25081", "published_at": "2026-09-23 04:00:00+00:00", "updated_at": "2026-09-23 04:25:51.691788+00:00", "lang": "en", "topics": ["large-language-models", "ai-research", "ai-safety", "natural-language-processing"], "entities": ["ufakzeka-1", "arXiv", "Apache-2.0"], "alternates": {"html": "https://wpnews.pro/news/ufakzeka-1-building-and-evaluating-a-151m-parameter-turkish-language-model-from", "markdown": "https://wpnews.pro/news/ufakzeka-1-building-and-evaluating-a-151m-parameter-turkish-language-model-from.md", "text": "https://wpnews.pro/news/ufakzeka-1-building-and-evaluating-a-151m-parameter-turkish-language-model-from.txt", "jsonld": "https://wpnews.pro/news/ufakzeka-1-building-and-evaluating-a-151m-parameter-turkish-language-model-from.jsonld"}}