adzcai/AfriGuardPlain-AfriqueQwen3.5-4B-50Langs-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 AfriGuardPlain-AfriqueQwen3.5-4B-50Langs-lora-merged model, developed by adzcai, is a 4.5 billion parameter language model with a 32768 token context length. It is a fine-tuned version of McGill-NLP/AfriqueQwen3.5-4B-50Langs, utilizing LoRA, and merged into the base weights. This model is specifically trained on the AfriGuard-plain dataset, which consists of user prompts and plain replies without safety system instructions or tags, making it suitable for applications requiring direct, unfiltered responses.

Loading preview...

AfriGuardPlain-AfriqueQwen3.5-4B-50Langs-lora-merged Overview

This model is a 4.5 billion parameter language model, fine-tuned by adzcai from the McGill-NLP/AfriqueQwen3.5-4B-50Langs base model. It leverages LoRA (rank 16, applied to all linear layers) which has been merged into the base weights. The model was trained on the adzcai/AfriGuard-plain dataset, which is characterized by its direct user prompts and plain replies, intentionally omitting safety system instructions and specific tagging like <safety>, <category>, or <response>.

Key Capabilities

  • Direct Response Generation: Designed to provide straightforward replies to user prompts without additional safety or categorization layers.
  • LoRA Fine-tuning: Benefits from efficient fine-tuning using LoRA, integrated directly into the base model.
  • Multilingual Base: Built upon the AfriqueQwen3.5-4B-50Langs base, suggesting potential for multilingual understanding and generation, though specific language performance is not detailed for this fine-tuned version.

Good For

  • Unfiltered Text Generation: Use cases where direct, unprompted responses are preferred, without the influence of explicit safety instructions.
  • Research on Response Modalities: Exploring model behavior when safety system instructions are deliberately excluded from training.
  • Comparative Studies: Can be used in conjunction with its instruction-prompted counterpart, adzcai/AfriGuardInst-AfriqueQwen3.5-4B-50Langs-lora-merged, to analyze the impact of instruction-tuning on model outputs.

Finetunes
1 models