adzcai/AfriGuardInst-AfriqueQwen3.5-4B-lora-merged
AfriGuardInst-AfriqueQwen3.5-4B-lora-merged is a 4.5 billion parameter instruction-tuned language model, developed by adzcai, based on the McGill-NLP/AfriqueQwen3.5-4B architecture. It was fine-tuned using LoRA on the israel/AfriGuard-inst dataset, which focuses on safety system instructions and user prompts. This model is specifically designed to generate safety-aligned responses, categorizing and responding to prompts within explicit safety tags, making it suitable for content moderation and responsible AI applications.
Loading preview...
Overview
AfriGuardInst-AfriqueQwen3.5-4B-lora-merged is a 4.5 billion parameter instruction-tuned language model, built upon the McGill-NLP/AfriqueQwen3.5-4B base model. It has been fine-tuned using LoRA (rank 16, all linear layers) on the israel/AfriGuard-inst dataset. This dataset structures each example with safety system instructions, a user prompt, and a target output format including <safety>…</safety>, <category>…</category>, and <response>…</response> tags.
Key Capabilities
- Safety-aligned response generation: Designed to produce outputs that adhere to specified safety guidelines.
- Structured output: Generates responses encapsulated within
<safety>,<category>, and<response>tags, facilitating programmatic parsing and integration. - Instruction following: Optimized to follow safety system instructions provided in the prompt.
Good for
- Content moderation systems: Ideal for applications requiring automated identification and categorization of potentially unsafe content.
- Responsible AI development: Useful for building AI systems that need to generate responses within predefined safety boundaries.
- Research on safety alignment: Provides a specialized model for studying and improving safety mechanisms in language models.