GoTo Group Releases 27 Billion Parameter Indonesian Language Model

GoTo Group released a 27 billion parameter language model for Indonesian and regional languages during Bali Tech Week on October 19, according to TechNode Global. The model is available through GoTo's Hugging Face repository.

Built on the Gemma 3 architecture, the instruction tuned model supports English, Indonesian, Javanese, Sundanese, Batak Toba and Balinese. It has a context window of 128,000 tokens. GoTo initiated the project and developed the model with AI Singapore.

GoTo describes the model as open source, while its model card lists the Gemma license. The model card estimates that full precision inference requires about 54 GB to 60 GB of video memory, with suggested hardware including one Nvidia A100 80GB, two Nvidia L40S cards or three RTX 4090 cards.

The repository includes guidance for running the model with vLLM and Hugging Face Transformers. It also warns that the model can produce inaccurate information, may struggle with reasoning tasks and requires testing and safeguards before use.