gemma2-9b-cpt-sahabatai-v1-instruct-GGUF
| Property | Value |
|---|---|
| Parameter Count | 9.24B |
| Languages | English, Indonesian, Javanese, Sundanese |
| License | Gemma Community License |
| Author | QuantFactory |
| Base Model | Gemma2 9B |
What is gemma2-9b-cpt-sahabatai-v1-instruct-GGUF?
This is a quantized version of the Sahabat-AI language model, specifically designed for Indonesian and regional language processing. The model has been instruction-tuned on approximately 448,000 Indonesian instruction pairs, 96,000 Javanese pairs, 98,000 Sundanese pairs, and 129,000 English instruction pairs, making it particularly effective for multilingual applications in the Indonesian region.
Implementation Details
The model is built on the Gemma2 architecture with 9.24B parameters and supports a context length of 8192 tokens. It has undergone both full parameter fine-tuning and on-policy alignment, trained on 8x H100-80GB GPUs.
- Extensive instruction tuning across four languages
- Optimized for regional language understanding
- Strong performance on SEA HELM and IndoMMLU benchmarks
- Quantized for efficient deployment
Core Capabilities
- Achieves 61.169% overall score on SEA HELM benchmark
- 62.6% accuracy on IndoMMLU
- Excels in Indonesian (64.154%), Javanese (64.439%), and Sundanese (54.913%) tasks
- Maintains strong English language capabilities (33.67% average on standard benchmarks)
Frequently Asked Questions
Q: What makes this model unique?
The model's primary strength lies in its comprehensive coverage of Indonesian and regional languages, with state-of-the-art performance across multiple benchmarks. It's specifically optimized for Indonesian, Javanese, and Sundanese while maintaining English language capabilities.
Q: What are the recommended use cases?
The model is well-suited for multilingual applications in the Indonesian region, including content generation, question answering, and language understanding tasks. However, users should note that it hasn't been aligned for safety, and appropriate security measures should be implemented.
