SiamGPT: Quality-First Fine-Tuning for Stable Thai Text Generation

TL;DR

SiamGPT-32B is a Thai language model fine-tuned with a quality-first approach, combining English instruction data and linguistic constraints to improve stability and instruction adherence without increasing data scale.

Contribution

Introduces SiamGPT-32B, a Thai language model fine-tuned with a novel quality-first strategy emphasizing curated supervision over data quantity.

Findings

Outperforms similar-scale open-weights Thai models on SEA-HELM benchmark.

Improves instruction adherence and multi-turn robustness.

Achieves stable and natural Thai text generation.

Abstract

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.