Mistral 发布 Timestral 时间序列基础模型系列
原标题：Introducing TimestralState-of-the-art probabilistic forecasts with just 209M parameters, outperforming larger models on CPU.
AI 摘要
Mistral AI 发布首个时间序列基础模型系列 Timestral，最大模型仅 209M 参数，却在 GIFT-Eval、TIME 和 fev-bench 三个主要公开基准上排名第一，超越参数量更大的模型。该系列包含 5M、28M、123M、209M 四个尺寸，支持概率化多变量预测及过去/未来协变量，无需 GPU 即可在 CPU 上高效推理。模型结合合成预训练、协变量感知注意力和长上下文学习，并可通过 Mistral Forge 进行领域微调。
正文节选
Thinking Summary Timestral, Mistral’s first family of time-series foundation models, delivers state-of-the-art probabilistic, multivariate forecasts with just 209M parameters, outperforming larger models on key benchmarks while supporting past and future covariates. Available in four sizes (5M to 209M parameters), it balances accuracy and efficiency, running efficiently on CPUs and excelling in zero-shot forecasting for diverse domains. The model combines synthetic pre-training, covariate-aware