The mdok-style systems submitted to the SemEval-2026 Task 10 shared task.
If you use the data, code, or the information in this repository, cite the following paper.
TBA
| File | Description |
|---|---|
| mdok-style.py | the script for training and inference of the pure mdok-style system |
| mdok-style_robust.py | the script for training and inference of the self-trained mdok-style system submitted to subtask 2 of PsyCoMark |
Clone and install the IMGTB framework, activate the conda environment.
git clone https://github.com/kinit-sk/IMGTB.git
cd IMGTB
conda env create -f environment.yaml
conda activate IMGTB
-
To retrain the Qwen3-32B model, run the enclosed code mdok-style.py.
-
Run predictions using the fine-tuned Qwen3-32B model on the dev and test splits and save them as dev_silver.jsonl and test_silver.jsonl.
-
Retrain self-trained Qwen3-32B model by running the enclosed mdok-style_robust.py script.
-
To run just inference, append
--test_onlyoption to the enclosed scripts.