African-language continued pretraining
According to the model card, the model is part of the AfriqueLLM suite adapted to 20 African languages through continued pretraining on about 26B tokens.
Open Source Model Profile · McGill-NLP
AfriqueQwen-4B
AfriqueQwen-4B is a 4.02B-parameter Qwen3 causal model from McGill-NLP, continued-pretrained on about 26B tokens for 20 African languages.
- Publisher
- McGill-NLP
- Task
- text-generation
- Model type
- qwen3
- License
- cc-by-4.0
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
AfriqueQwen-4B is published by McGill-NLP as a Qwen3 text-generation model in the AfriqueLLM suite. The captured configuration identifies Qwen3ForCausalLM and Safetensors metadata reports 4,022,468,096 parameters. According to the model card, it continues Qwen/Qwen3-4B-Base on about 26B tokens for 20 African languages, with card data recording cc-by-4.0.
Recorded capabilities
Qwen3-4B-Base lineage
According to the model card, the model is based on Qwen/Qwen3-4B-Base and is typed as a causal base/pretrained language model with 4B parameters.
Documented 32K native context
According to the model card, native context length is 32,768 tokens, with 16,384-token context and BF16 mixed precision reported for training.
Transformers and vLLM deployment notes
According to the model card, the publisher documents Transformers loading and vLLM or SGLang deployment for an OpenAI-compatible endpoint.
Use cases in the source record
- Multilingual text generation across the 20 documented African languages plus English, French, Portuguese, and Arabic.
- Document-level translation and long-context experiments, which the model card associates with Qwen 3 continued pretraining.
- Local or self-hosted deployment using the publisher's documented Transformers, vLLM, or SGLang setup.
Limitations and unknowns
- According to the model card, the extracted evaluation table covers related AfriqueQwen3.5 variants rather than directly reported AfriqueQwen-4B scores; no standalone AfriqueQwen-4B benchmark is extracted.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
- No pricing, VRAM, or current availability figures were extracted.
Source and provenance
Source: McGill-NLP/AfriqueQwen-4B
Captured: Unknown. Processed: 2026-09-07T19:35:41.889888+00:00.
AfriqueQwen-4B Model Overview AfriqueQwen-4B 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 Qwen/Qwen3-4B-Base and has been specifically adapted for improved performance on African languages while maintaining strong capabilities in high-resource languages. Our experiments show that Qwen 3 models achieve the best performance among all base models tested, better preserving performance in high-resource languages after CPT and achieving strong results on long-context tasks such as document-level translation…
F001F002F003F004F005F006F007F010F011F012F013F014F016F017F018