Tag
论文提出MixLID，一个支持111种语言-文字组合的token级混合语言识别模型，采用两阶段训练（先单语再混合语言数据），在单语和混合语言识别基准上均超越现有基线，模型已在Hugging Face开源。
Rampart is a 14.7 MB ONNX token-classification model for client-side PII detection, designed to redact personal information before it leaves the user's device, supporting multiple languages and running via ONNX Runtime Web.
OpenAI releases Privacy Filter, a 1.5B parameter bidirectional token classification model for PII detection and masking, featuring an Apache 2.0 license and long-context support for high-throughput data sanitization.