CamemBERT

The reference French encoder line from Inria ALMAnaCH. The original CamemBERT (2019, with FAIR and Sorbonne) still sees ~1.05M downloads a month; CamemBERT 2.0 (2024, RoBERTa + DeBERTaV3 variants on 275B tokens) and ModernCamemBERT (2025, a ModernBERT port on 1T high-quality French tokens with a controlled architecture study) refresh it. The encoder foundation beneath the generative Gaperon line.