ELECTRA Model Summary

The following table summarizes the currently supported ELECTRA models and their corresponding pretrained weights in PaddleNLP. For model details, please refer to the corresponding links.

| | Trained on Chinese text. |

Note: Both models follow ELECTRA-style architecture. The discriminator is trained for sequence classification tasks, while the generator is used for token-level predictions in the ELECTRA framework. | Chinese | Generator, 12-layer, 64-hidden, | | | 1-heads, 3M parameters. | | | Trained on Chinese legal corpus. | | | | | | Please refer to: | | | hfl/chinese-legal-electra-small-generator .. _hfl/chinese-electra-180g-base-discriminator: https://huggingface.co/hfl/chinese-electra-180g-base-discriminator .. _hfl/chinese-electra-180g-small-ex-discriminator: https://huggingface.co/hfl/chinese-electra-180g-small-ex-discriminator .. _hfl/chinese-legal-electra-small-generator: https://huggingface.co/hfl/chinese-legal-electra-small-generator