Melola Discovery

← Melola Discovery

Akses terbuka Jurnal SINTA 1

Optimizing Fine-Tuning Memory Usage with QLoRA for Automated Pedagogical Feedback Generation in Bahasa Indonesia

Journal of Information Systems Engineering and Business Intelligence · Vol. 12 No. 3 (2026) · hlm. 580–595

DOI
10.20473/jisebi.12.3.580-595
ISSN
2443-2555, 2598-6333
Terbit
2026-10-28
Penerbit
Universitas Airlangga
Bahasa
eng
Lisensi
creativecommons.org/licenses/by/4.0

Abstrak

Background: While recent advances in open-source large language models (LLMs) have enabled automated feedback generation for educational assessment, most other research is focused on English and requires high-end Graphics Processing Unit (GPU) resources. However, there is limited empirical evidence of the efficiency of LLMs in Bahasa Indonesia feedback tasks in low-resource settings like Indonesia, given the strict hardware limitations and differences in data sources Objective: This study investigates whether 4-bit quantization combined with Low-Rank Adaptation (LoRA)-based task adaptation can enable open-weight LLMs to generate Indonesian pedagogical feedback under strict 8GB Video Random Access Memory (VRAM) constraints, comparing the performance across various model architectures and supervision data (human-curated vs synthetic). Methods: A quantitative and experimental design using a 2 × 2 factorial structure was used to compare Llama-3-8B-Instruct and Gemma-2-9B-Instruct on 2 datasets of equal size (1,950 human-curated and 1,950 synthetic feedback pairs). For model fine-tuning, we implemented 4-bit Quantized Low-Rank Adaptation (QLoRA) quantization on consumer-grade GPUs with 8GB VRAM Performance was assessed by semantic similarity using Sentence-BERT (SBERT), similarity of the model with BLEU-4, ROUGE-L, peak VRAM usage, and training time. Results: QLoRA successfully enabled fine-tuning within 8GB VRAM, with peak memory usage between 7.0 and 7.8GB. Gemma-2-9B-Instruct achieved the highest semantic alignment (SBERT similarity = 0.77 on synthetic and 0.74 on human data), while Llama-3-8B-Instruct showed competitive performance (0.76 and 0.71, respectively). Human-curated datasets produced higher lexical alignment (ROUGE-1 up to 44.8) than synthetic datasets, though synthetic supervision remained semantically consistent. Conclusion: QLoRA enables efficient fine-tuning of large language models on consumer-grade hardware, making Indonesian feedback generation feasible in low-resource settings. While 4-bit quantization reduces memory requirements, LoRA enables task-specific pedagogical adaptation. Instruction-tuned models consistently outperform language-adapted base models. Future work should include human evaluation and broader educational contexts to improve generalizability. Keywords: Automated Feedback Generation, Large Language Models, QLoRA Fine-Tuning, Bahasa Indonesia, Educational Assessment

Jaringan kolaborasi para penulis

6 penulis · 9 tautan kepenulisan bersama · 5 lintas organisasi

Arahkan kursor atau ketuk sebuah nama untuk menyorot rekan penulisnya. Klik dua kali untuk membuka halaman penulis.

  • Atma Jaya University Yogyakarta
  • Binus University
  • Management Center Innsbruck
  • Afiliasi tidak diketahui
  • Organisasi berbeda
  • Organisasi sama
  • Afiliasi tidak diketahui
  • Garis lebih tebal = lebih banyak artikel bersama · lingkaran lebih besar = lebih banyak artikel
Tampilkan sebagai tabel
Rekan penulisOrganisasiArtikel bersama
Bernadectus Yudi DwiandiyantaAtma Jaya University Yogyakarta2
Andrea CorradiniManagement Center Innsbruck2
Yulia Ery KurniawatiAfiliasi tidak diketahui1

Afiliasi berasal dari OpenAlex; metadata OAI-PMH jurnal tidak memuatnya. Penulis tanpa ID OpenAlex dicocokkan berdasarkan nama, sehingga dua orang bisa tergabung atau satu orang bisa terpisah.

Sumber metadata: OAI-PMH (metadata jurnal) https://e-journal.unair.ac.id/JISEBI/oai, pertama di-harvest 29 Okt 2026, terakhir terlihat 15 Nov 2026.
Metadata asli (oai_dc)
<oai_dc:dc xmlns="http://www.openarchives.org/OAI/2.0/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:oai_dc="http://www.openarchives.org/OAI/2.0/oai_dc/" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:schemaLocation="http://www.openarchives.org/OAI/2.0/oai_dc/  http://www.openarchives.org/OAI/2.0/oai_dc.xsd">
	<dc:title xml:lang="en-US">Optimizing Fine-Tuning Memory Usage with QLoRA for Automated Pedagogical Feedback Generation in Bahasa Indonesia</dc:title>
	<dc:creator>Maslim, Martinus</dc:creator>
	<dc:creator>Wibisono, Albert William</dc:creator>
	<dc:creator>Prabowo, Yulius Denny</dc:creator>
	<dc:description xml:lang="en-US">Background: While recent advances in open-source large language models (LLMs) have enabled automated feedback generation for educational assessment, most other research is focused on English and requires high-end Graphics Processing Unit (GPU) resources. However, there is limited empirical evidence of the efficiency of LLMs in Bahasa Indonesia feedback tasks in low-resource settings like Indonesia, given the strict hardware limitations and differences in data sources 
Objective: This study investigates whether 4-bit quantization combined with Low-Rank Adaptation (LoRA)-based task adaptation can enable open-weight LLMs to generate Indonesian pedagogical feedback under strict 8GB Video Random Access Memory (VRAM) constraints, comparing the performance across various model architectures and supervision data (human-curated vs synthetic).  
Methods: A quantitative and experimental design using a 2 × 2 factorial structure was used to compare Llama-3-8B-Instruct and Gemma-2-9B-Instruct on 2 datasets of equal size (1,950 human-curated and 1,950 synthetic feedback pairs). For model fine-tuning, we implemented 4-bit Quantized Low-Rank Adaptation (QLoRA) quantization on consumer-grade GPUs with 8GB VRAM Performance was assessed by semantic similarity using Sentence-BERT (SBERT), similarity of the model with BLEU-4, ROUGE-L, peak VRAM usage, and training time. 
Results: QLoRA successfully enabled fine-tuning within 8GB VRAM, with peak memory usage between 7.0 and 7.8GB. Gemma-2-9B-Instruct achieved the highest semantic alignment (SBERT similarity = 0.77 on synthetic and 0.74 on human data), while Llama-3-8B-Instruct showed competitive performance (0.76 and 0.71, respectively). Human-curated datasets produced higher lexical alignment (ROUGE-1 up to 44.8) than synthetic datasets, though synthetic supervision remained semantically consistent. 
Conclusion: QLoRA enables efficient fine-tuning of large language models on consumer-grade hardware, making Indonesian feedback generation feasible in low-resource settings. While 4-bit quantization reduces memory requirements, LoRA enables task-specific pedagogical adaptation. Instruction-tuned models consistently outperform language-adapted base models. Future work should include human evaluation and broader educational contexts to improve generalizability. 
Keywords: Automated Feedback Generation, Large Language Models, QLoRA Fine-Tuning, Bahasa Indonesia, Educational Assessment </dc:description>
	<dc:publisher xml:lang="en-US">Universitas Airlangga</dc:publisher>
	<dc:date>2026-10-28</dc:date>
	<dc:type>info:eu-repo/semantics/article</dc:type>
	<dc:type>info:eu-repo/semantics/publishedVersion</dc:type>
	<dc:format>application/pdf</dc:format>
	<dc:identifier>https://e-journal.unair.ac.id/JISEBI/article/view/86890</dc:identifier>
	<dc:identifier>10.20473/jisebi.12.3.580-595</dc:identifier>
	<dc:source xml:lang="en-US">Journal of Information Systems Engineering and Business Intelligence; Vol. 12 No. 3 (2026): October; 580-595</dc:source>
	<dc:source>2443-2555</dc:source>
	<dc:source>2598-6333</dc:source>
	<dc:language>eng</dc:language>
	<dc:relation>https://e-journal.unair.ac.id/JISEBI/article/view/86890/41632</dc:relation>
	<dc:rights xml:lang="en-US">Copyright (c) 2026 The Authors. Published by Universitas Airlangga.</dc:rights>
	<dc:rights xml:lang="en-US">http://creativecommons.org/licenses/by/4.0</dc:rights>
</oai_dc:dc>