Search papers, labs, and topics across Lattice.
The paper introduces TUDUM, a pipeline designed to adapt the Qwen3.5-27B model for Turkish reasoning by focusing on both language and reasoning trace localization. Through supervised fine-tuning on 15,991 Turkish reasoning examples and reinforcement learning in a Turkish mathematics environment, the model exhibited improved Turkish reasoning behavior but faced trade-offs in benchmark accuracy. While the fine-tuning reduced response length and thinking exhaustion, the overall performance compared to the base model remained mixed, highlighting the complexities of achieving effective reasoning in a non-English context.
TUDUM reveals that fine-tuning for Turkish reasoning can enhance linguistic alignment but may compromise benchmark accuracy, challenging assumptions about model performance in multilingual settings.
This paper presents TUDUM (T\"urk\c{c}e D\"u\c{s}\"unen \"Uretken Model), a project pipeline for adapting a Qwen-family 27B thinking model toward Turkish reasoning. The central problem is not only to answer Turkish prompts in Turkish, but to make the explicit reasoning trace itself Turkish. A thinking model may translate a Turkish prompt into an English-centered internal or visible scratchpad, solve the problem mostly in English, and only localize the final answer. TUDUM instead treats the generated...block as a trainable behavior. The pipeline starts from the project base checkpoint unsloth/Qwen3.5-27B, applies supervised fine-tuning (SFT) on 15,991 Turkish reasoning examples using LoRA adapters, and then applies GRPO-family reinforcement learning on a proxy-filtered Turkish mathematics environment. The results are mixed. SFT made the model shorter and more consistently Turkish in its reasoning behavior, with large reductions in average response length and thinking exhaustion, but reduced benchmark accuracy. RL recovered some mathematical performance, especially AIME24 at the best early checkpoint, yet did not uniformly improve all benchmarks and did not exceed the base model on the reported Macro-6 average. The contribution is therefore best framed as a technically honest Turkish-thinking reasoning pipeline and evaluation, not as a claim of state-of-the-art Turkish reasoning. The released step-50 model is publicly available.
