Hugging Face Overview
Model Overview
Sahabat-AI/Llama-Sahabat-AI-v2-70B-IT is a 70 billion parameter instruction-tuned large language model developed by PT GoTo Gojek Tokopedia Tbk and AI Singapore. It is part of the Sahabat-AI collection, focused on pretraining and instruct-tuning for Indonesian languages and local dialects. The model uses the default Llama 3.1 tokenizer and features an extended context length of 128k tokens.
Key Capabilities
- Multilingual Support: Supports English, Indonesian, Javanese, Sundanese, Batak Toba, and Balinese.
- Instruction Following: Evaluated on instruction-following capabilities using localized datasets like SEA-IFEval and SEA-MTBench.
- Local Context Understanding: Assessed for Indonesian context-rooted capabilities via the IndoMMLU benchmark, covering humanities, language, culture, social science, and STEM topics.
- General Language Tasks: Performance on tasks such as Question Answering, Sentiment Analysis, Toxicity Detection, Translation, Summarization, Causal Reasoning, and Natural Language Inference is evaluated using the SEA-HELM benchmark.
Usage Considerations
This model requires significant computational resources, with a minimum of approximately 140 GB of VRAM for FP16 or BF16 precision. It is aligned for general safety but users should implement their own safety fine-tuning. The model may exhibit limitations such as hallucination and occasional inconsistencies in reasoning.