Merge branch 'rename_diffuser_paddle' of https://github.com/JunnYu/PaddleNLP into rename_diffuser_paddle

Author: JunnYu

README_cn.md

@@ -30,6 +30,8 @@
 **PaddleNLP**是一款**简单易用**且**功能强大**的自然语言处理开发库。聚合业界**优质预训练模型**并提供**开箱即用**的开发体验，覆盖NLP多场景的模型库搭配**产业实践范例**可满足开发者**灵活定制**的需求。
 
 ## News 📢
+* 🔥 **2022.10.27 发布 [PaddleNLP v2.4.2](https://github.com/PaddlePaddle/PaddleNLP/releases/tag/v2.4.2)**
+  * NLG能力扩充：📄 新增[**基于Pegasus的中文文本摘要方案**](https://github.com/PaddlePaddle/PaddleNLP/tree/develop/applications/text_summarization/pegasus)，效果领先；❓新增[**问题生成解决方案**](./examples/question_generation)，提供基于业界领先模型UNIMO-Text和大规模多领域问题生成数据集训练的**通用问题生成预训练模型**。均支持Taskflow一键调用，支持FasterGeneration高性能推理，训练推理部署全流程打通。
 
 * 🔥 **2022.10.14 发布 [PaddleNLP v2.4.1](https://github.com/PaddlePaddle/PaddleNLP/releases/tag/v2.4.1)**
   * 🧾 发布多语言跨模态布局增强文档智能大模型 [**ERNIE-Layout**](./model_zoo/ernie-layout/)，刷新11项任务SOTA。同步发布基于ERNIE-Layout的**文档抽取问答模型DocPrompt** 🔖，精准理解文档图片布局与语义信息，轻松应对各类业务场景。

README_en.md

@@ -37,6 +37,8 @@
     <div align="center">
     <img src="https://user-images.githubusercontent.com/11793384/197535273-75db1fb3-3a34-4bd2-bac3-bcacb1b986a8.jpg" width="150" height="150" />
     </div>
+* 🔥 **2022.10.27 [PaddleNLP v2.4.2](https://github.com/PaddlePaddle/PaddleNLP/releases/tag/v2.4.2) Released!**
+  * NLG Upgrade: 📄 Release [**Solution of Text Summarization**](https://github.com/PaddlePaddle/PaddleNLP/tree/develop/applications/text_summarization/pegasus) based on Pegasus；❓ Release [**Solution of Problem Generation**](./examples/question_generation), providing **general problem generation pre-trained model** based on Baidu's UNIMO Text and large-scale multi domain problem generation dataset。Supporting high-performance inference ability based on FasterGeneration , and covering the whole process of training , inference and deployment.
 
 * 🔥 **2022.10.14 [PaddleNLP v2.4.1](https://github.com/PaddlePaddle/PaddleNLP/releases/tag/v2.4.1) Released!**
   * 🧾 Release multilingual/cross-lingual pre-trained models [**ERNIE-Layout**](./model_zoo/ernie-layout/) which achieves new SOTA results in 11 downstream tasks. **DocPrompt** 🔖 based on ERNIE-Layout is also released which has the ability for multilingual document information extraction and question ansering.

examples/language_model/gpt-3/dygraph/run_pretrain.py

@@ -266,6 +266,13 @@ def do_train(args):
             # and add it to 'paddle.amp.decorate'
             multi_precision=args.use_pure_fp16)
 
+    # decorate @to_static for benchmark, skip it by default.
+    if args.to_static:
+        specs = None
+        model = paddle.jit.to_static(model, input_spec=specs)
+        logger.info(
+            "Successfully to apply @to_static with specs: {}".format(specs))
+
     if args.use_pure_fp16:
         scaler = paddle.amp.GradScaler(init_loss_scaling=args.scale_loss)
         # level O2 means converting the network to FP16
@@ -301,13 +308,6 @@ def do_train(args):
             logger.warning("No optimizer checkpoint file found in %s." %
                            opt_path)
 
-    # decorate @to_static for benchmark, skip it by default.
-    if args.to_static:
-        specs = None
-        model = paddle.jit.to_static(model, input_spec=specs)
-        logger.info(
-            "Successfully to apply @to_static with specs: {}".format(specs))
-
     global_step = 0
     tic_train = time.time()
     for epoch in range(args.num_train_epochs):

examples/language_model/moe/dygraph/framework/__init__.py

@@ -15,5 +15,5 @@
 from .storage_process import assign_group_by_size
 from .storage_process import flatten_dense_tensors
 from .storage_process import obtain_storage
-from .adamw import AdamW
+from paddle.optimizer import AdamW
 from .group_sharded import group_sharded_parallel