Introducing Timestral [LAB]
New small open model family with CPU-efficient sizes, relevant to local serving and lightweight model evaluation. https://mistral.ai/news/timestral/ (mistral.ai)
Scaling Indic Pretraining: 2.8 Trillion Tokens and an Open Data Generation Pipeline [LAB]
Open pretraining data pipeline and large-scale training details useful for fine-tuning and data-curation work. https://huggingface.co/blog/nvidia/scaling-indic-pretraining (huggingface.co)