Merge pull request !2303 from gziyan/modify_lars_interface

Merge pull request !2441 from chenzhongming/master

Merge pull request !2140 from ShidaHe/debugger_dev

Merge pull request !237 from zhaozhenlong/fix-issue-ssim-ksize-check-incu

Merge pull request !2180 from chenweifeng/gelu-fp16

Merge pull request !236 from wuxuejian/incu_embedding

Merge pull request !2297 from jiangjinsheng/vm_matrixdiag

Merge pull request !2305 from 李鸿章/callback

Merge pull request !2465 from chentingting/gcn_rectification

Merge pull request !2418 from caojian05/ms_master_dev

move embeddinglookup to the internal

Merge pull request !2241 from zhangzheng/accumulate

Merge pull request !2306 from luoyang/pylint

Merge pull request !2405 from chenzhongming/master

Merge pull request !2413 from Xiaoda/7-auto-parallel-check-optimizer-ops

Merge pull request !2424 from xutianchun/ctcloss

Merge pull request !2369 from baihuawei/reduce

Merge pull request !2398 from h.farahat/slice_bug

Merge pull request !2404 from xulei/input_output

Merge pull request !2375 from h.farahat/python_tokenizer

- It provides a unified style to express graph and kernel for user.
- It provides a unified IR to represent graph and kernel for developer.
- It breaks the boundary between graph and kernel.
- It provides more opportunities to do compile optimization.

Merge pull request !2397 from zhaoting/ResNet50_st

Merge pull request !2380 from xiefangqi/xfq_fix_coco_issue_01

Merge pull request !2384 from YuJianfeng/split

Merge pull request !2378 from flywind/fix_yolov3_bug

Merge pull request !2056 from zhoufeng/enable-new-control-sink

Merge pull request !2157 from hangq/master

Merge pull request !2388 from VectorSL/bert-eval

Merge pull request !2143 from ms_yan/del_storage_c++