Guo Sheng

### PR types New features ### PR changes APIs ### Description Make BERT support past_key_value_caches. ```python import numpy as np import paddle import paddlenlp from paddlenlp.transformers import AutoModel, AutoTokenizer, AutoModelForTokenClassification,...

Fix book-nmt for Fluid 1.3

1

Fix book-nmt for Fluid 1.3

Electra 模型增加对输入 `past_key_values` 和 `inputs_embeds` 的支持

4

### Feature request Electra 模型增加对输入 `past_key_values` 和 `inputs_embeds` 的支持。可以参考 [ERNIE 中对应的代码](https://github.com/PaddlePaddle/PaddleNLP/blob/develop/paddlenlp/transformers/ernie/modeling.py#L890) ### Motivation 通过 `past_key_values` 和 `inputs_embeds` 输入可以支持 Prefix-Tuning、P-Tuning这样一些工作。

Add Pipeline Parallel for PPO training and support generation with InferenceModel

1

### PR types New features ### PR changes Others ### Description Add Pipeline Parallel for PPO training and support generation with InferenceModel

Fix LLM prediction with sharded ckpts using FuseMT in dygraph mode

4

### PR types Bug fixes ### PR changes Others ### Description Fix LLM prediction with sharded ckpts using FuseMT in dygraph mode

Add unit tests and more apis for text.py Currently, text.py includes apis as follows: ```python __all__ = [ 'RNNCell', 'BasicLSTMCell', 'BasicGRUCell', 'RNN', 'BidirectionalRNN', 'StackedRNNCell', 'StackedLSTMCell', 'LSTM', 'BidirectionalLSTM', 'StackedGRUCell', 'GRU', 'BidirectionalGRU',...

[Bug] MBPP evaluator cannot extract the correct anwser

4

### Prerequisite - [X] I have searched [Issues](https://github.com/open-compass/opencompass/issues/) and [Discussions](https://github.com/open-compass/opencompass/discussions) but cannot get the expected help. - [X] The bug has not been fixed in the [latest version](https://github.com/open-compass/opencompass). ### Type...

cherry-pick from #4394

1

cherry-pick from #4394

Add ParallelExecutor for Transformer

1

Add ParallelExecutor for Transformer

Add label smoothing in Transformer

1

Add label smoothing in Transformer.