针对你的问题“please use model.hf_quantizer.is_trainable instead”,以下是详细的回答,包含相关代码片段: 确认model.hf_quantizer.is_trainable属性的存在: 在最新版本的Hugging Face Transformers库中,model.hf_quantizer.is_trainable属性被用来替代之前的_is_quantized_training_enabled方法,以检查模型是否处于量化训练...
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX. - FEAT : Adding BitNet quantization method to HFQuantizer (#33410) · huggingface/transformers@36d410d
This pull request introduces a new quantization method:BitNetquantization at 1.58 bits. It enables users to load and utilize quantized & packed models with ternary weights directly in Transformers, providing out-of-the-box inference. if device_map is None: logger.warning_once( dacorvoSep 13, 20...