-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathindex.xml
245 lines (193 loc) · 10.8 KB
/
index.xml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
<channel>
<title>Triloon</title>
<link>https://triloon.space/</link>
<description>Recent content on Triloon</description>
<generator>Hugo -- gohugo.io</generator>
<language>en-us</language>
<lastBuildDate>Tue, 18 Jan 2022 19:07:51 +0800</lastBuildDate><atom:link href="https://triloon.space/index.xml" rel="self" type="application/rss+xml" />
<item>
<title>Triplet Loss 与在线难例挖掘（译）</title>
<link>https://triloon.space/posts/triplet-loss-0/</link>
<pubDate>Tue, 18 Jan 2022 19:07:51 +0800</pubDate>
<guid>https://triloon.space/posts/triplet-loss-0/</guid>
<description><p>虽然 triplet loss 实现非常简单，但是简单的 loss 要想用好也是需要更细致的分析与调试。</p></description>
</item>
<item>
<title>Torch all_gather 的梯度问题</title>
<link>https://triloon.space/posts/torch-allgather/</link>
<pubDate>Sun, 16 Jan 2022 18:43:13 +0800</pubDate>
<guid>https://triloon.space/posts/torch-allgather/</guid>
<description><p>pytorch all_gather 计算结果是叶子节点，也就是不会继续向后传递梯度了。</p></description>
</item>
<item>
<title>Deepspeed Zero论文</title>
<link>https://triloon.space/posts/deepspeed-zero/</link>
<pubDate>Thu, 25 Nov 2021 10:41:35 +0800</pubDate>
<guid>https://triloon.space/posts/deepspeed-zero/</guid>
<description><p>DeepSpeed的开山之作 - ZeRO: Memory Optimizations Toward Training Trillion Parameter Models.</p></description>
</item>
<item>
<title>Swin Transformer V2</title>
<link>https://triloon.space/posts/swin-transformer-v2/</link>
<pubDate>Tue, 23 Nov 2021 12:07:18 +0800</pubDate>
<guid>https://triloon.space/posts/swin-transformer-v2/</guid>
<description><p>本文围绕如何有效的增加模型参数量、弥补不同任务输入图像尺寸不同时的Windows大小不同导致的相对位置编码变化问题这两个任务提出了解决方案，总的来说，方案简单有效，值得学习。</p></description>
</item>
<item>
<title>DALI简单实用案例</title>
<link>https://triloon.space/posts/dali-intro/</link>
<pubDate>Mon, 22 Nov 2021 17:20:46 +0800</pubDate>
<guid>https://triloon.space/posts/dali-intro/</guid>
<description><p>DALI(NVIDIA Data Loading Library)库是NVIDIA提供的用于加速数据加载过程的代码库，支持在GPU上完成一些数据处理，从而提高加载速度；另一方面是方便多种源数据文件格式的加载，包括MXNet RecordIO / TFRrecord / LMDB 或者 文件目录等形式的数据集加载；第三就是支持多种数据格式，包括图片、视频、音频等。</p></description>
</item>
<item>
<title>常见掩码生成方式 2</title>
<link>https://triloon.space/posts/mlm-related-2/</link>
<pubDate>Wed, 27 Oct 2021 22:08:15 +0800</pubDate>
<guid>https://triloon.space/posts/mlm-related-2/</guid>
<description><p>这是接着上一篇掩码生成方式写的，主要仅包含SpanBERT &amp; MacBERT的原理与实现。</p></description>
</item>
<item>
<title>MoE 论文阅读：Sparsely-Gated MoE</title>
<link>https://triloon.space/posts/moe-original-paper/</link>
<pubDate>Wed, 27 Oct 2021 11:15:57 +0800</pubDate>
<guid>https://triloon.space/posts/moe-original-paper/</guid>
<description><p>本文主要是一点 Mixture of Experts (MoE) 网络结构的介绍。</p></description>
</item>
<item>
<title>分词算法基础</title>
<link>https://triloon.space/posts/tokenizers/</link>
<pubDate>Fri, 22 Oct 2021 19:51:11 +0800</pubDate>
<guid>https://triloon.space/posts/tokenizers/</guid>
<description><p>常见的几种分词算法小结，包括BERT用到WordPiece以及Albert用到的Byte-Pair-Encoding。</p></description>
</item>
<item>
<title>常见掩码生成方式</title>
<link>https://triloon.space/posts/mlm-related/</link>
<pubDate>Tue, 19 Oct 2021 19:51:11 +0800</pubDate>
<guid>https://triloon.space/posts/mlm-related/</guid>
<description><p>主要是几种常见的MLM的改进以及对应的代码实现，包括WWM, SpanBERT, ERNIE这三种。</p></description>
</item>
<item>
<title>ICCV2021 Best Paper - Swin Transformer</title>
<link>https://triloon.space/posts/swin-transformer/</link>
<pubDate>Fri, 15 Oct 2021 22:07:45 +0800</pubDate>
<guid>https://triloon.space/posts/swin-transformer/</guid>
<description><p>重读论文之后，发现还真是一个非常精巧的模型。</p></description>
</item>
<item>
<title>Adversarial Training 2</title>
<link>https://triloon.space/posts/adversarial-training-2/</link>
<pubDate>Wed, 13 Oct 2021 14:30:10 +0800</pubDate>
<guid>https://triloon.space/posts/adversarial-training-2/</guid>
<description><p>对PGD算法的改进，包括FreeAT, FreeLB, SMART等。</p></description>
</item>
<item>
<title>Adversarial Training</title>
<link>https://triloon.space/posts/adversarial-training/</link>
<pubDate>Tue, 05 Oct 2021 11:21:33 +0800</pubDate>
<guid>https://triloon.space/posts/adversarial-training/</guid>
<description><p>几个基础的常见的对抗样本的生成算法，包括FGM/FGSM, PGD等。</p></description>
</item>
<item>
<title>Torch的一些使用方法记录</title>
<link>https://triloon.space/posts/torch-usage/</link>
<pubDate>Fri, 01 Oct 2021 16:06:21 +0800</pubDate>
<guid>https://triloon.space/posts/torch-usage/</guid>
<description><p>记录一些Torch使用过程中会用到的小知识点。</p></description>
</item>
<item>
<title>基于Transformer结构的图像自监督模型及训练</title>
<link>https://triloon.space/posts/img-transform-ssl/</link>
<pubDate>Tue, 28 Sep 2021 14:07:50 +0800</pubDate>
<guid>https://triloon.space/posts/img-transform-ssl/</guid>
<description><p>一些基于Transformer结构的图像自监督模型以及训练过程中遇到的问题。</p></description>
</item>
<item>
<title>Torch实现原理分析积累</title>
<link>https://triloon.space/posts/torch-impl-0/</link>
<pubDate>Sat, 18 Sep 2021 11:08:11 +0800</pubDate>
<guid>https://triloon.space/posts/torch-impl-0/</guid>
<description><p>Pytorch 实现学习积累。</p></description>
</item>
<item>
<title>Model Visualization</title>
<link>https://triloon.space/posts/model-visualization/</link>
<pubDate>Sat, 11 Sep 2021 15:16:07 +0800</pubDate>
<guid>https://triloon.space/posts/model-visualization/</guid>
<description><p>深度学习中的一些可视化技术。</p></description>
</item>
<item>
<title>Origin Transformer</title>
<link>https://triloon.space/posts/origin-transformer/</link>
<pubDate>Mon, 06 Sep 2021 20:04:32 +0800</pubDate>
<guid>https://triloon.space/posts/origin-transformer/</guid>
<description><p>Attention is all your need.</p></description>
</item>
<item>
<title>Cnn Transformer 系列之 Volo</title>
<link>https://triloon.space/posts/cnn-transformer-volo/</link>
<pubDate>Mon, 06 Sep 2021 16:26:11 +0800</pubDate>
<guid>https://triloon.space/posts/cnn-transformer-volo/</guid>
<description><p>VOLO论文。</p></description>
</item>
<item>
<title>从Adam到AdamW</title>
<link>https://triloon.space/posts/adam-adamw/</link>
<pubDate>Fri, 03 Sep 2021 20:30:59 +0800</pubDate>
<guid>https://triloon.space/posts/adam-adamw/</guid>
<description><p>Adam算法的实现以及一个主要改进AdamW的原理与实现。</p></description>
</item>
<item>
<title>Resnet Series</title>
<link>https://triloon.space/posts/resnet-series/</link>
<pubDate>Thu, 02 Sep 2021 14:12:59 +0800</pubDate>
<guid>https://triloon.space/posts/resnet-series/</guid>
<description><p>Residual Connection以及后续发展。</p></description>
</item>
<item>
<title>Binary Search Tree</title>
<link>https://triloon.space/posts/binary-search-tree/</link>
<pubDate>Tue, 31 Aug 2021 15:00:22 +0800</pubDate>
<guid>https://triloon.space/posts/binary-search-tree/</guid>
<description><p>二叉搜索树相关笔记</p></description>
</item>
<item>
<title>图像表征算法中的自监督学习方法</title>
<link>https://triloon.space/posts/ssl-image-representation/</link>
<pubDate>Tue, 31 Aug 2021 14:15:48 +0800</pubDate>
<guid>https://triloon.space/posts/ssl-image-representation/</guid>
<description><p>经典自监督模型，包括MoCo / SimCLR / SwAV / BYOL / SimSiam 等。</p></description>
</item>
<item>
<title>Efficient Transformer系列</title>
<link>https://triloon.space/posts/efficient-transformer/</link>
<pubDate>Tue, 31 Aug 2021 14:02:36 +0800</pubDate>
<guid>https://triloon.space/posts/efficient-transformer/</guid>
<description><p>一些以提高 Transformer 计算性能为目的的 Xformer 方案整理。</p></description>
</item>
<item>
<title>My First Post</title>
<link>https://triloon.space/posts/my-first-post/</link>
<pubDate>Mon, 30 Aug 2021 20:03:11 +0800</pubDate>
<guid>https://triloon.space/posts/my-first-post/</guid>
<description><p>怕什么真理无穷，进一寸有一寸的欢喜。</p></description>
</item>
<item>
<title>Hugo Notes</title>
<link>https://triloon.space/posts/hugo-notes/</link>
<pubDate>Mon, 30 Aug 2021 14:37:40 +0800</pubDate>
<guid>https://triloon.space/posts/hugo-notes/</guid>
<description><p>一些hugo blog搭建过程中的记录。</p></description>
</item>
<item>
<title>About</title>
<link>https://triloon.space/about/about/</link>
<pubDate>Mon, 30 Aug 2021 00:00:00 +0000</pubDate>
<guid>https://triloon.space/about/about/</guid>
<description>my github url is: Triloon</description>
</item>
</channel>
</rss>