20-language African adaptation
According to the model card, the model was adapted for 20 African languages through continued pretraining on about 26B tokens, retaining English, French, Portuguese, and Arabic for forgetting mitigation.
Open Source Model Profile · McGill-NLP
AfriqueGemma-12B
AfriqueGemma-12B is a 12.19B-parameter Gemma 3 text-generation model from McGill-NLP. According to the model card, it continues pretraining of Gemma 3 12B PT for 20 African languages.
- Publisher
- McGill-NLP
- Task
- text-generation
- Model type
- gemma3
- License
- cc-by-4.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
AfriqueGemma-12B is published by McGill-NLP as a gemma3-based text-generation model. The captured configuration identifies Gemma3ForConditionalGeneration and Safetensors metadata reports 12187325040 parameters. According to the model card, it continues pretraining of google/gemma-3-12b-pt on about 26B tokens for 20 African languages.
Recorded capabilities
Gemma 3 12B PT base
According to the model card and hub tags, the base is google/gemma-3-12b-pt, described as a causal base/pretrained model with 12B parameters.
Documented multilingual corpus and stack
According to the model card, the corpus combines about 22.8B African monolingual tokens with code, math, and synthetic data, trained with LLaMA-Factory, DeepSpeed ZeRO, Flash Attention 3, and BF16 mixed precision.
Gemma 3 12.19B Transformers record
Captured config identifies Gemma3ForConditionalGeneration and gemma3, with Safetensors metadata reporting 12187325040 parameters and transformers support.
CC-BY-4.0 licensing
Card data records cc-by-4.0, and the model card states the model is released under the CC BY 4.0 License.
Use cases in the source record
- Multilingual text generation across the documented 20 African languages plus English, French, Portuguese, and Arabic, using the card's Transformers example.
- Self-hosted deployment experiments with vLLM or SGLang OpenAI-compatible endpoints as documented in the model card.
Limitations and unknowns
- No evaluation results for AfriqueGemma-12B itself were extracted; the card's score table covers Qwen-based variants.
- Provider state is historical snapshot data, not independently refreshed current availability.
- Training, corpus, and context figures come from the publisher model card and were not independently verified by Ethen.
Source and provenance
Source: McGill-NLP/AfriqueGemma-12B
Captured: Unknown. Processed: 2026-09-07T19:35:41.788850+00:00.
AfriqueGemma-12B Model Overview AfriqueGemma-12B is part of the AfriqueLLM suite, a collection of open language models adapted to 20 African languages through continued pre-training (CPT) on ~26B tokens . This model is based on google/gemma-3-12b-pt and has been specifically adapted for improved performance on African languages while maintaining strong capabilities in high-resource languages. Key Features Type : Causal Language Model (Base/Pre-trained) Base Model : Gemma 3 12B PT Parameters : 12B Context Length : 8,192 tokens (native) Training Tokens : ~26B tokens of carefully curated multilingual data Supported Languages AfriqueGem…
F001F002F003F004F005F006F007F008F009F010F011F012F013F014F015F016F018