adzcai/AfriGuardInst-AfriqueQwen3.5-4B-50Langs-Instruct-v1-lora-merged

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 28, 2026License:cc-by-4.0Architecture:Transformer Open Weights Featherless Exclusive Cold

The adzcai/AfriGuardInst-AfriqueQwen3.5-4B-50Langs-Instruct-v1-lora-merged is a 4.5 billion parameter instruction-tuned language model, based on the AfriqueQwen3.5-4B-50Langs-Instruct-v1 architecture by McGill-NLP. This model is fine-tuned using LoRA on the AfriGuard-inst dataset, specifically designed for safety system instructions and user prompts. Its primary differentiation lies in its specialized training for generating safety-aligned responses, making it suitable for applications requiring content moderation and responsible AI interactions.

Loading preview...

Model Overview

This model, adzcai/AfriGuardInst-AfriqueQwen3.5-4B-50Langs-Instruct-v1-lora-merged, is a 4.5 billion parameter instruction-tuned variant of the McGill-NLP/AfriqueQwen3.5-4B-50Langs-Instruct-v1 base model. It has been fine-tuned using LoRA (rank 16, applied to all linear layers) with a focus on safety-related instructions.

Key Capabilities

  • Safety-Oriented Responses: Specifically trained on the israel/AfriGuard-inst dataset, which includes safety system instructions, user prompts, and structured outputs for safety, category, and response.
  • Instruction Following: Designed to process and respond to instructions, particularly those related to content safety and moderation.
  • LoRA Fine-tuning: Utilizes Low-Rank Adaptation for efficient fine-tuning, merging the LoRA weights into the base model.

Training Details

The model was trained for 1 epoch on a single GPU, using the base model's chat template. The training objective focused on minimizing loss specifically on the response portion of the output. Key hyperparameters included a learning rate of 0.0001, a batch size of 1 (with 2 gradient accumulation steps), and an AdamW optimizer. This setup serves as a matched baseline for comparison with models trained on the adzcai/AfriGuard-plain dataset.

Ideal Use Cases

This model is particularly well-suited for applications requiring:

  • Content Moderation: Identifying and categorizing potentially unsafe content based on provided instructions.
  • Safety System Integration: Generating responses that adhere to predefined safety guidelines.
  • Responsible AI Development: Assisting in the creation of AI systems that prioritize safety and ethical considerations.

Finetunes
1 models