Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2512.19455 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915716389666816 |
|---|---|
| author | Pairatsuppawat, Thittipat Tachaapornchai, Abhibhu Kusolsomboon, Paweekorn Chaiwong, Chutikan Chay-intr, Thodsaporn Viriyayudhakorn, Kobkrit Ketui, Nongnuch Wong, Aslan B. |
| author_facet | Pairatsuppawat, Thittipat Tachaapornchai, Abhibhu Kusolsomboon, Paweekorn Chaiwong, Chutikan Chay-intr, Thodsaporn Viriyayudhakorn, Kobkrit Ketui, Nongnuch Wong, Aslan B. |
| contents | Open-weights large language models remain difficult to deploy for Thai due to unstable generation under complex instructions, despite strong English performance. To mitigate these limitations, We present SiamGPT-32B, an open-weights model based on Qwen3-32B, fine-tuned with a Quality-First strategy emphasizing curated supervision over data scale. The fine-tuning pipeline combines high-complexity English instruction data with a Thai-adapted AutoIF framework for instruction and linguistic constraints. Using supervised fine-tuning only, without continual pretraining or corpus expansion, SiamGPT-32B improves instruction adherence, multi-turn robustness, and linguistic stability. Evaluations on the SEA-HELM benchmark show that SiamGPT-32B achieves the strongest overall performance among similar-scale open-weights Thai models, with consistent gains in instruction following, multi-turn dialogue, and natural language understanding. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2512_19455 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | SiamGPT: Quality-First Fine-Tuning for Stable Thai Text Generation Pairatsuppawat, Thittipat Tachaapornchai, Abhibhu Kusolsomboon, Paweekorn Chaiwong, Chutikan Chay-intr, Thodsaporn Viriyayudhakorn, Kobkrit Ketui, Nongnuch Wong, Aslan B. Computation and Language Open-weights large language models remain difficult to deploy for Thai due to unstable generation under complex instructions, despite strong English performance. To mitigate these limitations, We present SiamGPT-32B, an open-weights model based on Qwen3-32B, fine-tuned with a Quality-First strategy emphasizing curated supervision over data scale. The fine-tuning pipeline combines high-complexity English instruction data with a Thai-adapted AutoIF framework for instruction and linguistic constraints. Using supervised fine-tuning only, without continual pretraining or corpus expansion, SiamGPT-32B improves instruction adherence, multi-turn robustness, and linguistic stability. Evaluations on the SEA-HELM benchmark show that SiamGPT-32B achieves the strongest overall performance among similar-scale open-weights Thai models, with consistent gains in instruction following, multi-turn dialogue, and natural language understanding. |
| title | SiamGPT: Quality-First Fine-Tuning for Stable Thai Text Generation |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2512.19455 |