Sahabat-AI/gemma2-9b-cpt-sahabatai-v1-base Overview

This model is a 9 billion parameter decoder-only language model, co-initiated by Indonesian tech and telecommunication companies GoTo Group and Indosat Ooredoo Hutchison, and developed by PT GoTo Gojek Tokopedia Tbk and AI Singapore. It is a continued pre-trained version of the Gemma2 9B CPT SEA-Lionv3 base model, specifically enhanced for the Indonesian language and its dialects.

Key Capabilities

• Multilingual Proficiency: Optimized for Indonesian, Javanese, and Sundanese languages, in addition to English.
• Extensive Pre-training: Continued pre-training on approximately 50 billion tokens, including significant Indonesian, Javanese, and Sundanese datasets.
• General Language Tasks: Evaluated on the SEA HELM (BHASA) benchmark for tasks like Question Answering, Sentiment Analysis, Toxicity Detection, Translation, Summarization, Causal Reasoning, and Natural Language Inference.
• Strong Localized Performance: Achieves an overall score of 64.123 on the SEA HELM benchmark for Bahasa Indonesia, Javanese, and Sundanese, outperforming several other 7-9B models in this domain.
• Gemma Community License: Licensed under the Gemma Community License.

Good For

• Applications requiring robust language understanding and generation in Indonesian, Javanese, and Sundanese.
• Developers and researchers focusing on Southeast Asian language processing.
• Tasks such as content creation, customer support, and information retrieval in the Indonesian linguistic context.