Tag
This paper introduces MoganColBERT-TR, a late-interaction multi-vector retrieval model for Turkish, which achieves competitive zero-shot performance on Turkish BEIR datasets through distillation training from previous encoders.
This paper presents TÜDÜM, a pipeline for adapting Qwen3.5-27B to perform explicit reasoning in Turkish, using SFT and GRPO-based RL on Turkish reasoning data. Results show improved Turkish reasoning consistency but mixed benchmark performance, offering an honest evaluation rather than a SOTA claim.