@inproceedings{chaksangchaichot-akarajaradwong-2025-budget,
    title = "A Budget Recipe for Finetuning a Long-form Legal Summarization Model",
    author = "Chaksangchaichot, Chompakorn  and
      Akarajaradwong, Pawitsapak",
    editor = "Modi, Ashutosh  and
      Ghosh, Saptarshi  and
      Ekbal, Asif  and
      Goyal, Pawan  and
      Jain, Sarika  and
      Joshi, Abhinav  and
      Mishra, Shivani  and
      Datta, Debtanu  and
      Paul, Shounak  and
      Singh, Kshetrimayum Boynao  and
      Kumar, Sandeep",
    booktitle = "Proceedings of the 1st Workshop on NLP for Empowering Justice (JUST-NLP 2025)",
    month = dec,
    year = "2025",
    address = "Mumbai, India",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-ijcnlp-aacl/2025.justnlp-main.11/",
    pages = "113--120",
    ISBN = "979-8-89176-312-8",
    abstract = "We describe an inexpensive system that ranked first in the JUST-NLP 2025 L-SUMM task, summarizing very long Indian court judgments (up to 857k characters) using a single 80GB GPU and a total budget of about {\$}50. Our pipeline first filters out length{--}summary outliers, then applies two-stage LoRA SFT on Qwen3-4B-Instruct-2507 to learn style and extend context, and finally runs RLVR tuned to BLEU, ROUGE-2, and ROUGE-L, with BLEU upweighted. We showed that two-stage SFT is better than a single-stage run, and RLVR gives the largest gains, reaching 32.71 internal vs. 16.15 base and 29.91 on the test leaderboard. In ablation on prompting, we find that a simple, naive prompt converges faster but saturates earlier, while the curated legal-structured prompt keeps improving with longer training and yields higher final scores, and the finetuned model remains fairly robust to unseen prompts. Our code are fully open-sourced, available for reproducibility."
}