wolfe
wolfe
alpaca-zh等通用数据集? 这是chatglm之前预训练过的语料吗?怎么知道模型训练过哪些语料?
以为要保留原来的对话等通用能力,是加入chatglm训练时用过的语料 -- 所以...,其实类似alpaca-zh等通用数据集,也是可以避免遗忘问题
同问... 先做关系分类,再做实体识别。这是找到最早的论文了 Extracting contextualized complex biological events with rich graph based feature sets, Bj¨orne, J.; Heimonen, J.; Ginter, F.; Airola, A.; Pahikkala, T.; and Salakoski, T. 2011. Computational Intelligence. Identifying...
V100 does support 8bit and fp16?
just keep the these keys make it works: text, label, error_type
{"example":1, text:"您好", “label”:[], "serialno":"zx23432", "time":12324234, "other":[], "error_type":"漏召回", "predict":[]}
感谢解答!
@AndrewZhe 另外, 1. S5在CAM评测上,指标达到82.14,怎么解释?训练S5的动机是什么? 2. 在CAM评测任务上,“For the classification task about marriage (CAM), we select four kinds of causes of action, whose meanings have no overlapping to each other.”(意味着分类任务的样本之间区分度较高) 如果下游任务中,存在语义容易混淆的情况,如何解决回答不准的问题? 3....
zero shot能力从哪里来?是与指令微调的任务有关吗? #24