add ernie-gram framework figure

ernie-gram/README.en.md

 English|[简体中文](./README.zh.md)
 
-`Remind`: *ERNIE-Gram* model has been officially released in [here](??). Our reproduction codes will be released to [repro branch](https://github.com/PaddlePaddle/ERNIE/tree/repro) soon.
+`Remind`: *ERNIE-Gram* model has been officially released in [here](#3-download-pretrained-models-optional). Our reproduction codes will be released to [repro branch](https://github.com/PaddlePaddle/ERNIE/tree/repro) soon.
 
 
 ## _ERNIE-Gram_: Pre-Training with Explicitly N-Gram Masked Language Modeling for Natural Language Understanding
 
+![ERNIE-Gram](.meta/ernie-gram.jpeg)
+
 - [Framework](#ernie-gram-framework)
 - [Quick Tour](#quick-tour)
 - [Setup](#setup)
@@ -17,8 +19,6 @@ English|[简体中文](./README.zh.md)
 
 ### ERNIE-Gram Framework
 
-整图
-
 Since **ERNIE 1.0**, Baidu researchers have introduced **knowledge-enhanced representation learning** in pre-training to achieve better pre-training learning by masking consecutive words, phrases, named entities, and other semantic knowledge units. Furthermore, we propose **ERNIE-Gram**, an explicitly n-gram masking language model to enhance the integration of coarse-grained information for pre-training. In **ERNIE-Gram**, **n-grams** are masked and predicted directly using **explicit** n-gram identities rather than contiguous sequences of tokens.
 
 In downstream tasks, **ERNIE-gram** uses a `bert-style` fine-tuning approach, thus maintaining the same parameter size and computational complexity.
@@ -56,8 +56,8 @@ pip install -e .
 
 | Model                                              | Description                                                  |abbreviation|
 | :------------------------------------------------- | :----------------------------------------------------------- |:-----------|
-| [ERNIE-Gram Base for Chinese](补充链接) | Layer:12, Hidden:768, Heads:12 | ernie-gram|
-| [ERNIE-Gram Base for English](补充链接) | Layer:12, Hidden:768, Heads:12 | ernie-gram-en |
+| [ERNIE-Gram Base for Chinese](https://ernie-github.cdn.bcebos.com/model-ernie-gram-zh.1.tar.gz) | Layer:12, Hidden:768, Heads:12 | ernie-gram|
+| [ERNIE-Gram Base for English](https://ernie-github.cdn.bcebos.com/model-ernie-gram-en.1.tar.gz) | Layer:12, Hidden:768, Heads:12 | ernie-gram-en |
 
 ##### 4. Download datasets
 

ernie-gram/README.zh.md

 [English](./README.en.md)|简体中文
 
-`提醒`: *ERNIE-Gram* 中/英文模型已经[正式开源](??)，paper 复现代码也即将开源至 [repro分支](https://github.com/PaddlePaddle/ERNIE/tree/repro)。现在您可以使用基于 Paddle 2.0 全新升级、基于动静结合的新版 ERNIE 套件体验 *ERNIE-Gram* 中/英文开源模型。
+`提醒`: *ERNIE-Gram* 中/英文模型已经[正式开源](#3-下载预训练模型可选)，paper 复现代码也即将开源至 [repro分支](https://github.com/PaddlePaddle/ERNIE/tree/repro)。现在您可以使用基于 Paddle 2.0 全新升级、基于动静结合的新版 ERNIE 套件体验 *ERNIE-Gram* 中/英文开源模型。
 
 
 ## _ERNIE-Gram_: Pre-Training with Explicitly N-Gram Masked Language Modeling for Natural Language Understanding
 
+![ERNIE-Gram](.meta/ernie-gram.jpeg)
+
 - [模型框架](#模型框架)
 - [快速上手](#快速上手)
 - [安装& 使用](#安装)
@@ -17,8 +19,6 @@
 
 ### 模型框架
 
-整图
-
 从 **ERNIE 1.0** 起，百度研究者们就在预训练中引入**知识增强**学习，通过掩码连续的词、phrase、named entity 等语义知识单元，实现更好的预训练学习。本次开源的通用语义理解模型 **ERNIE-Gram** 更进一步，提出的**显式**、**完备**的 n-gram 掩码语言模型，实现了显式的 n-gram 语义单元知识建模。
 
 #### ERNIE 多粒度预训练语义理解技术
@@ -66,8 +66,8 @@ export PYTHONPATH=$PWD:$PYTHONPATH
 
 | Model                                              | 细节参数                                                                  |下载简写|
 | :------------------------------------------------- |:------------------------------------------------------------------------- |:-------|
-| [ERNIE-Gram 中文](补充链接)           | Layer:12, Hidden:768, Heads:12  |ernie-gram|
-| [ERNIE-Gram 英文](补充链接)                  | Layer:3, Hdden:1024, Heads:16   |ernie-gram-en|
+| [ERNIE-Gram 中文](https://ernie-github.cdn.bcebos.com/model-ernie-gram-zh.1.tar.gz)           | Layer:12, Hidden:768, Heads:12  |ernie-gram|
+| [ERNIE-Gram 英文](https://ernie-github.cdn.bcebos.com/model-ernie-gram-en.1.tar.gz)                  | Layer:3, Hdden:1024, Heads:16   |ernie-gram-en|
 
 ##### 4. 下载数据集
 
@@ -99,8 +99,7 @@ data/xnli
 使用 `动态图` 模型进行finetune:
   
   - [句对分类](./demo/finetune_classifier_distributed.py)
-  - [语义匹配](./demo/finetune_classifier.py)
-  - [命名实体识别(NER)](./demo/finetune_ner.py)
+  - [语义匹配](./demo/finetune_classifier_distributed.py)
   - [机器阅读理解](./demo/finetune_mrc.py)
 
 
@@ -108,13 +107,14 @@ data/xnli
 
 |任务|batch size|learning rate|
 |--|--|--|
-| XNLI         | 512             | 1e-4 |
-| LCQMC        | 32              | 2e-5 |
+| XNLI         | 256             | 1.5e-4 |
+| LCQMC        | 16              | 4e-5 |
 | DRCD         | 64              | 5e-5 |
-| CMRC2018     | 64              | 3e-5 |
-| DuReader     | 64              | 3e-5 |
+| CMRC2018     | 64              | 1.5e-4 |
+| DuReader     | 64              | 1.5e-5 |
 | MSRA-NER(SIGHAN2006)  | 16      | 5e-5 |
 
+若希望复现 paper 中的所有实验，请切换至本 repo 的 `repro` 分支。
 
 ### 文献引用
 
@@ -127,9 +127,6 @@ data/xnli
 }
 ```
 
-
-若希望复现 paper 中的所有实验，请切换至本 repo 的 `repro` 分支。
-
 ### 讨论组
 - [ERNIE官方主页](https://wenxin.baidu.com/)
 - [Github Issues](https://github.com/PaddlePaddle/ERNIE/issues): bug reports, feature requests, install issues, usage issues, etc.