indonlp/cendol-llama2-13b-merged-inst

Cendol LLaMA-2 13B Instruct: Indonesian Language Model

This model is a 13 billion parameter LLaMA-2 based instruction-tuned generative large language model (LLM) from the Cendol collection, developed by IndoNLP. It is specifically fine-tuned for Indonesian languages, aiming to provide high-performance natural language processing capabilities for the region.

Key Capabilities

• Indonesian Language Specialization: Optimized for various NLP tasks in Indonesian, including sentiment analysis, topic modeling, machine translation, summarization, question answering, and paraphrasing.
• Instruction Following: Designed for single-turn, task-specific instruction following, making it suitable for direct command-based applications.
• Performance: Cendol models, including this 13B variant, have demonstrated superior performance compared to other open-source multilingual and region-specific LLMs on tested benchmarks.
• Architecture: Built on the LLaMA-2 architecture and instruction-tuned using LoRA (Low-Rank Adaptation) with the Cendol Collection v1 dataset.

Intended Use Cases

• Research: Primarily intended for research purposes, particularly in the domain of Indonesian natural language processing.
• Task-Specific Instructions: Ideal for applications requiring the model to execute specific NLP tasks based on single-turn instructions in Indonesian.

Limitations

• Language Scope: Primarily designed for Indonesian languages; performance in other languages is not guaranteed.
• Safety: As with all LLMs, potential for inaccurate, biased, or objectionable outputs exists, requiring safety testing for specific applications.