Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

docs(zh-cn): Reviewed 66_the-post-processing-step-in-question-answering-(pytorch).srt #442

Closed
wants to merge 2 commits into from
Closed

docs(zh-cn): Reviewed 66_the-post-processing-step-in-question-answering-(pytorch).srt #442

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
305618d
Update 66_the-post-processing-step-in-question-answering-(pytorch).srt
tyisme614 Dec 27, 2022
a0786d0
finished review
tyisme614 Dec 30, 2022
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
102 changes: 51 additions & 51 deletions subtitles/zh-CN/66_the-post-processing-step-in-question-answering-(pytorch).srt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,12 +5,12 @@

2
00:00:05,940 --> 00:00:08,913
- 问答任务中的后处理步骤。
- 问答任务中的后处理操作。
- The post-processing step in a question answering task.

3
00:00:10,440 --> 00:00:12,180
在做答题时，
在做问答任务时，
When doing question answering,

4
Expand All @@ -20,7 +20,7 @@ the processing of the initial dataset

5
00:00:14,550 --> 00:00:17,370
意味着将示例拆分为多个功能，
意味着以多个特征拆分示例，
implies splitting examples in several features,

6
Expand All @@ -30,37 +30,37 @@ which may or may not contain the answer.

7
00:00:21,000 --> 00:00:22,740
通过模型传递这些特征
由于我们的标签是词元的索引
Passing those features through the model

8
00:00:22,740 --> 00:00:25,830
将为我们提供开始和结束位置的 logits，
其对应于答案的起始和结束，
will give us logits for the start and end positions,

9
00:00:25,830 --> 00:00:28,650
因为我们的标签是令牌的索引
通过模型传递这些特征
since our labels are the indices of the token

10
00:00:28,650 --> 00:00:31,050
对应于开始和结束的答案。
将为我们提供开始和结束位置的对数值。
that correspond to the start and end the answer.

11
00:00:32,664 --> 00:00:35,490
然后我们必须以某种方式将这些 logits 转换为答案，
然后我们必须以某种方式将这些对数值转换为答案，
We must then somehow convert those logits into an answer,

12
00:00:35,490 --> 00:00:38,610
然后从每个功能给出的各种答案中选择一个
然后从每个特征给出的各种答案中选择一个
and then pick one of the various answers each feature gives

13
00:00:38,610 --> 00:00:40,893
成为给定示例的答案。
作为给定示例的答案。
to be the answer for a given example.

14
Expand All @@ -70,12 +70,12 @@ For the processing step,

15
00:00:43,500 --> 00:00:45,750
你应该参考下面链接的视频。
你可以参考下面链接的视频。
you should refer to the video linked below.

16
00:00:45,750 --> 00:00:47,820
验证并没有太大的不同，
验证方面也没有太大的变化，
It's not very different for validation,

17
Expand All @@ -85,42 +85,42 @@ we just need to add a few lines to keep track of two things.

18
00:00:51,660 --> 00:00:54,960
我们保留它们，而不是丢弃偏移映射，
我们保留 offset mapping，而不是丢弃它们，
Instead of discarding the offset mappings, we keep them,

19
00:00:54,960 --> 00:00:55,793
也包括在其中
并且通过设置特殊词元的偏移量
and also include in them

20
00:00:55,793 --> 00:00:58,350
上下文在哪里的信息
以及将问题设置为 None
the information of where the context is

21
00:00:58,350 --> 00:01:00,690
通过设置特殊标记的偏移量
将所保留的 offset mapping
by setting the offsets of the special tokens

22
00:01:00,690 --> 00:01:02,253
和无的问题。
包含在上下文的信息里。
and the question to None.

23
00:01:03,480 --> 00:01:06,630
然后我们还跟踪每个功能的示例 ID，
然后我们还跟踪每个特征的示例 ID，
Then we also keep track of the example ID for each feature,

24
00:01:06,630 --> 00:01:08,280
能够映射回特征
能够将特征映射回
to be able to map back feature

25
00:01:08,280 --> 00:01:10,503
他们起源的例子。
他们初始的例子。
to the examples that they originated from.

26
Expand All @@ -130,22 +130,22 @@ If you don't want to compute the validation loss,

27
00:01:14,100 --> 00:01:15,990
你不需要包含所有特殊代码
你不需要包含所有这些
you won't need to include all the special code

28
00:01:15,990 --> 00:01:18,420
我们用来创建标签的。
用来创建标签的特殊代码。
that we used to create the labels.

29
00:01:18,420 --> 00:01:21,090
完成后，我们可以应用该预处理功能
完成后，我们可以调用 map 方法
With this done, we can apply that preprocessing function

30
00:01:21,090 --> 00:01:22,890
使用映射方法。
应用该预处理功能。
using the map method.

31
Expand All @@ -155,7 +155,7 @@ We take the SQUAD dataset

32
00:01:24,090 --> 00:01:26,840
比如问答视频的预处理。
就和问答视频的预处理中所用到的一样。
like in the preprocessing for question-answering video.

33
Expand All @@ -165,12 +165,12 @@ Once this is done, the next step is to create our model.

34
00:01:30,540 --> 00:01:31,710
我们使用默认模型
我们在这里的问答管道背后
We use the default model

35
00:01:31,710 --> 00:01:33,930
在这里的问答管道背后，
使用默认模型，
behind the question-answering pipeline here,

36
Expand All @@ -190,17 +190,17 @@ so we create a PyTorch DataLoader with our features.

39
00:01:42,657 --> 00:01:44,520
有了它，我们可以计算和收集
有了它，我们可以使用标准的 PyTorch 评估循环
With it, we can compute and gather

40
00:01:44,520 --> 00:01:46,650
所有的开始和结束都是这样的，
来计算和收集所有
all the start and end logits like this,

41
00:01:46,650 --> 00:01:49,653
使用标准的 PyTorch 评估循环。
像这样的开始和结束的对数值。
with a standard PyTorch evaluation loop.

42
Expand All @@ -215,7 +215,7 @@ First, we'll need a map from example to features,

44
00:01:56,340 --> 00:01:57,873
我们可以这样创建。
我们像这样创建。
which we can create like this.

45
Expand All @@ -225,22 +225,22 @@ Now, for the main part of the post-processing,

46
00:02:00,810 --> 00:02:04,230
让我们看看如何从 logits 中提取答案。
让我们看看如何从对数中提取答案。
let's see how to extract an answer from the logits.

47
00:02:04,230 --> 00:02:05,760
我们可以只取最好的索引
我们可以针对开始和结束对数值
We could just take the best index

48
00:02:05,760 --> 00:02:07,980
对于开始和结束登录并完成，
只取最好的索引，
for the start and end logits and be done,

49
00:02:07,980 --> 00:02:10,380
但如果我们的模型预测了一些不可能的事情，
但如果我们的模型预测得出了不可思议的结果，
but if our model predicts something impossible,

50
Expand All @@ -250,7 +250,7 @@ like tokens in the question,

51
00:02:12,150 --> 00:02:13,940
我们将查看更多的 logits。
我们将查看更多的对数值。
we'll look at more of the logits.

52
Expand All @@ -260,12 +260,12 @@ Note that in the question-answering pipeline,

53
00:02:17,070 --> 00:02:18,870
我们将分数归因于每个答案
我们基于概率将分数归因
we attributed score to each answer

54
00:02:18,870 --> 00:02:20,430
基于概率，
于每个答案，
based on the probabilities,

55
Expand All @@ -275,7 +275,7 @@ which we did not compute here.

56
00:02:22,350 --> 00:02:25,560
就逻辑而言，我们在分数中的乘法
就对数而言，我们在分数中的乘法
In terms of logits, the multiplication we had in the scores

57
Expand All @@ -285,27 +285,27 @@ becomes an addition.

58
00:02:28,110 --> 00:02:29,010
要走得快，
要快速完成，
To go fast,

59
00:02:29,010 --> 00:02:31,800
我们不会查看所有可能的开始和结束日志，
我们不会查看所有可能的开始和结束的对数，
we don't look at all possible start and end logits,

60
00:02:31,800 --> 00:02:34,050
但是最好的 20 个就足够了。
仅需最好的 20 个就足够了。
but the 20 best one is enough.

61
00:02:34,050 --> 00:02:36,570
我们忽略产生不可能答案的逻辑
我们忽略产生不可能答案和答案太长
We ignore the logits that spawn impossible answers

62
00:02:36,570 --> 00:02:38,550
或回答太长。
的对数值。
or answer that are too long.

63
Expand All @@ -315,7 +315,7 @@ As we saw in the preprocessing, the labels 0,0

64
00:02:41,430 --> 00:02:43,230
对应不回答。
对应无答案。
correspond to a no answer.

65
Expand All @@ -330,12 +330,12 @@ to get the answer inside the context.

67
00:02:47,910 --> 00:02:49,107
我们来看看预测答案
我们来看看对于第一个特征
Let's have a look at the predicted answer

68
00:02:49,107 --> 00:02:50,370
对于第一个功能，
预测的答案，
for the first feature,

69
Expand All @@ -345,7 +345,7 @@ which is the answer with the best score

70
00:02:51,930 --> 00:02:53,640
或最好的逻辑分数
或最好的对数分数
or the best logit score

71
Expand All @@ -355,7 +355,7 @@ since the SoftMax is an increasing function.

72
00:02:56,280 --> 00:02:58,230
模型做对了。
这个模型就是合适的。
The model got it right.

73
Expand All @@ -365,12 +365,12 @@ Next we just have to loop this for every example,

74
00:03:00,690 --> 00:03:03,720
为每个选择具有最佳 logit 分数的答案
在示例生成的所有特征中
picking for each the answer with the best logit score

75
00:03:03,720 --> 00:03:06,750
在示例生成的所有功能中。
选择具有最佳对数分数的答案。
in all the features the example generated.

76
Expand Down