Sahabat-AI-Gemma2-9B-Instruct
Parameters
9.2B
Context Length
8K
Modality
Text
Architecture
Dense
License
Gemma-Community
Release Date
14 Nov 2024
Knowledge Cutoff
Mar 2024
System Requirements
VRAM requirements for different quantization methods and context sizes
1,024 tokens
Consumer
1x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
8,192 tokens
Consumer
1x RTX 4090
24GB VRAM
Datacenter
1x NVIDIA A100
80GB VRAM
Apple Silicon
1x Apple M3 Max
128GB VRAM
Architecture Diagram
Evaluation Benchmarks
No evaluation benchmarks for Sahabat-AI-Gemma2-9B-Instruct available.
Rankings
Overall Rank
-
Coding Rank
-
About Sahabat-AI-Gemma2-9B-Instruct
Sahabat-AI-Gemma2-9B-Instruct is a specialized large language model developed through a strategic collaboration between GoTo Group, Indosat Ooredoo Hutchison, and AI Singapore. Built upon the Google Gemma 2 architecture, this variant is the result of continued pre-training (CPT) and intensive instruction tuning specifically tailored for the Indonesian linguistic ecosystem. It is engineered to provide high-fidelity conversational capabilities not only in standard Bahasa Indonesia but also in major regional dialects, including Javanese and Sundanese, addressing the cultural and linguistic nuances inherent to the Indonesian archipelago.
The underlying architecture follows a decoder-only transformer design that incorporates several modern refinements for efficiency and stability. It utilizes Grouped-Query Attention (GQA) to optimize inference throughput and memory bandwidth, which is particularly effective for maintaining performance during long-context processing. For training stability and representational accuracy, the model employs RMSNorm for pre- and post-normalization across layers and integrates logit soft-capping to prevent divergence. The instruction-tuning phase involved a supervised fine-tuning process using a localized dataset of over 600,000 instruction-completion pairs, followed by on-policy alignment and model merging to refine its response quality and adherence to complex prompts.
Technically, the model is optimized for a wide array of natural language processing tasks, including sentiment analysis, toxicity detection, causal reasoning, and abstractive summarization within Southeast Asian contexts. By leveraging the base Gemma 2 9B weights, it inherits a robust world-knowledge foundation while specializing in regional idioms and cultural contexts that are often underrepresented in global models. This makes it a suitable candidate for developers building localized digital assistants, automated customer service interfaces, and educational tools designed for the Indonesian market.
Technical Specifications
Attention
Attention Structure
Multi-Head Attention
Attention Heads
16
Key-Value Heads
8
Attention Head Dimension
256
Position Embedding
Absolute Position Embedding
RoPE Theta
10,000
Sliding Window Attention
Yes
Sliding Window Size
4,096
Sliding Window Ratio
-
Linear Attention
-
Linear Attention Ratio
-
Normalization
RMS Normalization
Activation Function
Gated GELU
Dimensions
Hidden Dimension Size
3,584
Number of Layers
42
FFN Intermediate Size (Dense)
14,336
Multi-Token Prediction Heads
-
Tokenizer
Vocabulary Size
256,000
About Sahabat-AI
Sahabat-AI is an Indonesian language model family co-initiated by GoTo and Indosat Ooredoo Hutchison. Developed with AI Singapore and NVIDIA, it is a collection of models (based on Gemma 2 and Llama 3) specifically optimized for Bahasa Indonesia and regional languages like Javanese and Sundanese.