Overview
The mesolitica/malaysian-mistral-7b-32k-instructions is a 7 billion parameter Mistral-based model developed by Mesolitica, specifically fine-tuned for instruction following in the Malaysian language. It leverages a 32,768 token context length, making it suitable for processing longer prompts and generating comprehensive responses. The model utilizes the exact Mistral Instruct chat template for consistent interaction.
Key Capabilities
- Malaysian Instruction Following: Optimized to understand and respond accurately to instructions provided in Malaysian, covering a range of topics from general knowledge to specific queries.
- Extended Context Window: Benefits from a 32k context length, allowing for more detailed conversations and the processing of extensive input texts.
- Function Calling: Demonstrates the ability to parse and respond to function call requests, extracting entities from text based on provided function schemas.
- Efficient Inference: Designed to be used with
use_flash_attention_2 and BitsAndBytesConfig for efficient inference, including 4-bit quantization.
Good For
- Applications requiring a strong understanding of the Malaysian language and cultural context.
- Chatbots and conversational AI systems targeting Malaysian users.
- Tasks involving entity extraction or structured output generation via function calling in a Malaysian context.
- Research and development in low-resource language NLP, specifically for Malaysian.