KIT_IWSLT26_IF_SHORT_ unconstrained_ contrastive1

We participate in the unconstrained data condition. Our system is based on an cascaded system model built on Parakeet ASR followed by Qwen2.5-Omni-7B text-to-text, only the Qwen2.5-Omni-7B was fine-tuned.

We employ task-balanced data interleaving with temperature-based sampling to balance heterogeneous task distributions and mitigate dominance of high-resource tasks, including ASR, speech translation (ST), spoken question answering (SQA), speech summarization (SSUM), audio captioning (ACHAP), multiple-choice QA (MC), and general instruction-following tasks.

From\To	de	en	it	zh
From\To	en	N/A	N/A	N/A	N/A