Fine-tuned tool-calling models from the SakThai family — Qwen2.5-based LoRA adapters and merged checkpoints trained on curated SFT datasets.