adzcai/AfriGuardPlain-AfriqueQwen3.5-4B-50Langs
AfriGuardPlain-AfriqueQwen3.5-4B-50Langs is a 4.5 billion parameter language model fine-tuned by adzcai from McGill-NLP's AfriqueQwen3.5-4B-50Langs. This model is specifically trained on the AfriGuard-plain dataset to provide direct answers to safe prompts and brief refusals to unsafe ones, without emitting explicit safety labels. It features a 32768 token context length and is designed for direct, unprompted safety responses.
Loading preview...
Model Overview
AfriGuardPlain-AfriqueQwen3.5-4B-50Langs is a 4.5 billion parameter language model developed by adzcai, based on the McGill-NLP/AfriqueQwen3.5-4B-50Langs architecture. It has been fine-tuned using DeepSpeed ZeRO-3 on the adzcai/AfriGuard-plain dataset.
Key Characteristics
- Direct Response Mechanism: Unlike instruction-prompted safety models, this version is trained to provide a plain reply directly. For safe prompts, it offers a helpful answer; for unsafe prompts, it gives a brief refusal.
- No Explicit Safety Tags: The model operates without safety system instructions or
<safety>/<category>/<response>tags, meaning it does not emit a safety label but rather answers or refuses directly. - Base Model: Built upon the AfriqueQwen3.5-4B-50Langs, which supports 50 languages.
- Training Data: Fine-tuned on the non-instruction-prompted version of the AfriGuard dataset.
- Context Length: Supports a context length of 32768 tokens.
Training Details
The model was trained for 1 epoch on a single GPU, utilizing a qwen3_nothink chat template. The loss was calculated exclusively on the response portion of the training examples. Key hyperparameters included a learning rate of 1e-05 and a total training batch size of 2.