lab22-dpo-vn

DPO-aligned Vietnamese LLM · VinUni AICB Program · Lab 22.

Item	Value
Base model	{BASE_MODEL}
SFT data	5CD-AI/Vietnamese-alpaca-cleaned (1k samples, 1 epoch)
DPO data	argilla/ultrafeedback-binarized-preferences-cleaned (2k pairs)
beta	{BETA}
lr	{LR}
LoRA r	16 (alpha=32)