Cloud-Edge Collaborative Chinese Black Industry Jargon Processing Method Based on Lightweight Transfer Learning

In order to solve the problem that pre-trained language models (PLMs) cannot be run and trained on edge devices due to the large number of parameters, lightweighting is introduced and the migration module and the large pre-trained language model are deployed separately to achieve an efficient cloud-edge collaborative transfer learning framework. Make use of the box mentioned It is possible to migrate the representation of large pre-trained language models to downstream tasks with only a few parameters fine-tuned, and can also perform cross-domain cloud edges. Collaborative reasoning. Downstream tasks in multiple domains can share the same pre-trained language model collaboratively, which can effectively save computing overhead. Tasks can be divided efficiently. Separate deployment on different devices to achieve separate deployment of multiple tasks and sharing of pre-trained models. Experiments were made on 4 datasets of public natural language tasks. The results show that the performance of the framework can reach more than 95% of that of the fully fine-tuned BERT method.