Gemma2 9B CPT Sahabat-AI v1 Instruct
| Property | Value |
|---|---|
| Parameter Count | 9.24B |
| Model Type | Decoder |
| Languages | English, Indonesian, Javanese, Sundanese |
| License | Gemma Community License |
| Context Length | 8192 tokens |
What is gemma2-9b-cpt-sahabatai-v1-instruct?
Sahabat-AI v1 Instruct is a sophisticated multilingual language model developed through collaboration between GoTo Group and Indosat Ooredoo Hutchison. It's specifically optimized for Indonesian languages, trained on 448,000 Indonesian instruction pairs, 96,000 Javanese pairs, 98,000 Sundanese pairs, and 129,000 English instruction pairs.
Implementation Details
Built on the Gemma2 architecture, this model underwent extensive fine-tuning using 8x H100-80GB GPUs, with 4 hours of fine-tuning and 2 hours of alignment training. It maintains the original Gemma-2-9B tokenizer and supports a context length of 8192 tokens.
- Achieves state-of-the-art performance on SEA HELM benchmark with 61.169% overall score
- Demonstrates superior performance on IndoMMLU with 62.6% accuracy
- Supports multiple Indonesian dialects with robust instruction-following capabilities
Core Capabilities
- Multilingual instruction processing in 4 languages
- Advanced reasoning and task completion across various domains
- Strong performance in question answering, sentiment analysis, and translation tasks
- Context-aware responses with cultural relevance to Indonesian languages
Frequently Asked Questions
Q: What makes this model unique?
Its specialized focus on Indonesian languages and dialects, combined with state-of-the-art performance on regional benchmarks, makes it particularly valuable for Indonesian language applications.
Q: What are the recommended use cases?
The model excels in multilingual task completion, educational applications, content generation in Indonesian languages, and cross-lingual understanding between English and Indonesian dialects.
