azale-ai/DukunLM-13B-V1.0-Uncensored

DukunLM-13B-V1.0-Uncensored Overview

DukunLM-13B-V1.0-Uncensored is a 13 billion parameter language model developed by azale-ai, specifically optimized for generating Indonesian text. It is an instruction-tuned model, fine-tuned from the ehartford/WizardLM-13B-V1.0-Uncensored base model using the QLoRA method. The training utilized the Indonesian subset of the MBZUAI/Bactrian-X dataset, adapting the model for robust performance in the Indonesian language.

Key Characteristics

• Indonesian Language Focus: Primarily designed for generating and understanding Indonesian text.
• Uncensored Nature: The model operates without explicit content filters or alignment, meaning it can produce direct and unfiltered responses.
• Base Model: Built upon the WizardLM-13B-V1.0-Uncensored architecture.
• Fine-tuning: Utilizes QLoRA for efficient fine-tuning on a specialized Indonesian dataset.
• Prompt Format: Employs the Alpaca prompt format for instruction-following tasks.

Important Considerations

• Uncensored Content: Users should be aware that due to its uncensored nature, the model may generate content that contains errors, cultural biases, or potentially offensive material. Responsible usage is advised.
• Limitations: The model's base language is English, and while fine-tuned for Indonesian, it may still exhibit cultural and contextual biases inherited from its training data.

Use Cases

• Indonesian Text Generation: Ideal for applications requiring the creation of Indonesian language content.
• Research and Development: Suitable for researchers exploring uncensored language models and their behavior in specific linguistic contexts.
• Instruction Following: Capable of responding to instructions formatted in the Alpaca style for various tasks in Indonesian.