Bahasalab/Bahasa-4b-chat
Bahasa-4b-chat is a 4 billion parameter language model developed by Bahasa AI, built upon the Qwen-4b architecture. It is specifically optimized for Indonesian language tasks, having undergone continued training on 10 billion high-quality Indonesian text tokens. This model demonstrates strong performance in Indonesian NLP applications like question answering and sentiment analysis, often outperforming other 4B and some 7B models in this domain.
Loading preview...
Model Overview
Bahasa-4b-chat is a 4 billion parameter language model developed by Bahasa AI, specifically designed and optimized for the Indonesian language. It is built upon the Qwen-4b architecture and has undergone extensive continued training using a substantial dataset of 10 billion high-quality Indonesian text tokens.
Key Capabilities and Performance
This model excels in various Indonesian Natural Language Processing (NLP) tasks, including question answering, sentiment analysis, and document summarization. Benchmarks demonstrate that Bahasa-4b-chat consistently outperforms the Sailor_4b model across multiple Indonesian datasets like tydiqa-id, xcopa-id, and m3exam-id-ppl, showing improvements in both Exact Match (EM) and F1 scores. It also proves competitive with larger models such as Mistral-7B-v0.1 in several Indonesian language benchmarks.
Intended Use Cases
- Indonesian Language Understanding: Ideal for applications requiring deep comprehension of Indonesian text.
- Text Generation: Suitable for generating coherent and contextually relevant Indonesian responses.
- NLP Tasks: Recommended for tasks like question answering, sentiment analysis, and summarization in Indonesian.