Brendan Kenny <******@****.***> Brendan Kenny <******@****.***>
Connor Clark <******@****.***> cjamcl <******@****.***>
Connor Clark <******@****.***> Connor Clark <******@****.***>
Patrick Hulce <******@****.***> Patrick Hulce <******@****.***>
Paul Irish <******@****.***> Paul Irish <******@****.***>

## 特性

### 核心算法重写

这次算法重写带来的主要特性是异步渲染。（**注意**：在 16.0 中尚不支持，但是在未来的 16.x 版本中将会做为可选特性）。另外，新的重写删除了一些不成熟的、妨碍了内部变化的抽象。

> 这些多来自于 [Lin Clark 的演讲](https://www.youtube.com/watch?v=ZCuYPiUIONs)，所以你可以看看这个演讲，再在 twitter 上 [关注并点赞 Clark](https://twitter.com/linclark) 来支持她这个视角独特的概述。

异步渲染的意义在于能够将渲染任务划分为多块。浏览器的渲染引擎是单线程的，这意味着几乎所有的行为都是同步发生的。React 16 使用原生的浏览器 API 来间歇性地检查当前是否还有其他任务需要完成，从而实现了对主线程和渲染过程的管理。在 Firefox 中，一个浏览器主线程的例子很简单：

Use with caution. I have no downstream tasks in Norwegian to test on so I have no idea of its performance yet.
# Model
## Electra: Pre-training Text Encoders as Discriminators Rather Than Generators
Kevin Clark and Minh-Thang Luong and Quoc V. Le and Christopher D. Manning
- https://openreview.net/pdf?id=r1xMH1BtvB
- https://github.com/google-research/electra
# Acknowledgments
### TensorFlow Research Cloud

  propTablesExclude: [Item, Header, React.Fragment],
})(() => {
  const user = {
    name: 'Chrissy',
    email: 'chris.clark@sentry.io',
  };
  const team = {
    slug: 'team-slug',
  };
  const project = {
    slug: 'project-slug',
  };
  const organization = {
    slug: 'organization-slug',
  };
  const member = {

  https://arxiv.org/abs/1905.05950
* Are Sixteen Heads Really Better than One? by Paul Michel, Omer Levy, Graham Neubig: https://arxiv.org/abs/1905.10650
* What Does BERT Look At? An Analysis of BERT's Attention by Kevin Clark, Urvashi Khandelwal, Omer Levy, Christopher D.
  Manning: https://arxiv.org/abs/1906.04341

In order to help this new field develop, we have included a few additional features in the BERT/GPT/GPT-2 models to

# http://yaml.org/type/map.html -----------------------------------------------#

map:
  # Unordered set of key: value pairs.
  Block style: !!map
    Clark : Evans
    Ingy  : döt Net
    Oren  : Ben-Kiki
  Flow style: !!map { Clark: Evans, Ingy: döt Net, Oren: Ben-Kiki }

# http://yaml.org/type/omap.html ----------------------------------------------#

omap:
  # Explicitly typed ordered map (dictionary).

Bryan Bess <******@****.***>
C.T. Lin <******@****.***>
Chiahao Lin <******@****.***>
Chris Tarquini <******@****.***>
Christoffer Hallas <******@****.***>
Clark Du <clark******@****.***>
Darren Cauthon <******@****.***>
Debjeet Biswas <******@****.***>
Dmitry Mazuro <******@****.***>
Douglas Christopher Wilson <******@****.***>

- Alex Wood
- Anatoliy Yastreb
- Andrew Artajos
- Andrew Munsell
- AndrewCz
- Ankush Menat
- Anthony Smith
- Ben Balter
- Brian Jones
- Brint O'Hearn
- Chayoung You
- Chris Wells
- chrisfinazzo
- Clark Winkelmann
- Dan Allen
- David Von Lehman
- David Zhang
- Derek Gottlieb
- Enes Gönültaş
- EricH
- Erick Sasse
- Eugênio Cabral
- Florian Thomas
- Frank Taillandier
- Henry Goodman
- Henry Wright

 - they use an embedding projection layer, but Bert doesn't have one

## Pretraining ttask
![electra task diagram](https://github.com/shoarora/lmtuners/raw/master/assets/electra.png)
(figure from [Clark et al. 2020](https://openreview.net/pdf?id=r1xMH1BtvB))

ELECTRA uses discriminative LM / replaced-token-detection for pretraining.
This involves a generator (a Masked LM model) creating examples for a discriminator

> * 原文地址：[A cartoon intro to WebAssembly](https://hacks.mozilla.org/2017/02/a-cartoon-intro-to-webassembly/)
> * 原文作者：本文已获作者 [Lin Clark](https://code-cartoons.com/@linclark) 授权
> * 译文出自：[掘金翻译计划](https://github.com/xitu/gold-miner)
> * 译者： [根号三](https://github.com/sqrthree)
> * 校对者：[Reid](https://github.com/reid3290)、[Tina92](https://github.com/Tina92)

# 看漫画，学 WebAssembly

WebAssembly 运行得很快，你可能已经听说过这个了。但是是什么让 WebAssembly 这么快呢？