owned this note
owned this note
Published
Linked with GitHub
<div class="jumbotron" style="text-align:center">
<h1>
T1進度回報報
</h1>
<iframe width="400px" height="200ㄌㄣ" src="https://www.youtube.com/embed/-k56tk-qHwM" frameborder="0" allow="accelerometer; autoplay; encrypted-media; gyroscope; picture-in-picture" allowfullscreen></iframe>
<br>
請搭配BGM服用
</div>
## 2/19
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
裝zabbix, 筆記在樓下的連結
</li>
<li>
快完成重構python shell script及寫dockerfile
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
重灌11 12
</li>
<li>
完成德國賽測驗題目
</li>
<li>
樓下下們上課
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
成功裝impi，hpl，hpcc（搞懂所有路徑）明天用11 12跨機
</li>
<li>
聽樓下上AI課
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>AI教學，除了hpcg gpu跨機皆完成
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>練習用GPU跑Changa，跨機和單機<br>AI小教室
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過2008人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/18
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
裝zabbix, 筆記在樓下的連結
</li>
<li>
快完成重構python shell script及寫dockerfile
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
ChaNGa成功
</li>
<li>
發現Elmer需要多裝HDF5、NETCDF、MMG，目前卡在NETCDF的include
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
在12裡的container找centos openmpi的LAlib
</li>
<li>
複習intelmpi
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>開始跑BERT在wiki和bookcorpus(部分)的pre-trianing<br>在大機器用centos7的container裡裝HPCC、changa、elmer (皆完成)
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>開始跑BERT在wiki和bookcorpus(部分)的pre-trianing<br>在11用centos7的container裡開始練習裝HPCC(包含FFT的部分)、changa、elmer，筆記準備中
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/17
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
裝zabbix, 筆記在樓下的連結
</li>
<li>
快完成重構python shell script及寫dockerfile
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
在12用centos7的container裡開始練習hp系列(ok)、ChaNGa(卡載滿後面的)、Elmer/ice(比之前多一點進展)
</li>
<li>
加了所有CPU的dashboard
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
優化模型，找更好的激勵函數!
</li>
<li>
在n21裝zabbix(最後被n21單殺),明天重灌
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>繼續等待資料處理<br>在大G器用centos7的container裡開始練習裝Intel MPI、module load、HPL、linkpack、HPCG、HPCC(還差一點)
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>繼續等待資料處理<br>在11用centos7的container裡開始練習裝Intel MPI、module load、HPL、linkpack、HPCG、HPCC(剛完成，還沒跑benchmark)
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/15
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
裝zabbix, 筆記在樓下的連結
</li>
<li>
快完成重構python shell script及寫dockerfile
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
N11~N22及AI1全數上線zabbix，
<a href="http://192.168.1.12/zabbix">東西在這</a>
</li>
<li>
目前的dashboard
<a href="https://i.imgur.com/Ivi8dBn.png"><img src="https://i.imgur.com/Ivi8dBn.png" width=400px></a>
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
zabbix安裝成功，有遇到連不到電腦的問題，改port解決
</li>
<li>
重灌n21，測試隊長的python一鍵安裝（還是有bug）
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>參數調整<img src="https://i.imgur.com/Rx4vUKJ.png"><br>dataset轉換底bug，large測試(沒成功)，讀code
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>昨天的gradients checkpoints似乎不能在Nvidia的bert裡用，因為它使用的是自訂函數，回傳的值不一樣<br>看程式碼找到error 403的問題，是由於爬蟲太頻繁導致ip被封鎖，目前已嘗試過代理ip和user-agent均無效
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 1/7(2/14，不能約會只能約分) <- 好笑嗎?
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
裝zabbix, 筆記在樓下的連結
</li>
<li>
快完成重構python shell script及寫dockerfile
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
zabbix 裝好囉，
<a href="https://hackmd.io/msX5f6g9QH-3cc9u36ByQA">筆記點我</a>
</li>
<li>
grafana 放置中
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
優化模型，找更好的激勵函數!
</li>
<li>
在n21裝zabbix(最後被n21單殺),明天重灌
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>測不同的參數執行sst2
，想辦法運行large model，就算batch size設1，seq_len設2還是會out of memory<br>找了gradients accumulate與
gradients checkpoints兩種方法可以嘗試<br>正在理解並改code
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>繼續在大機器上測試：<br>跑large的fine-tunning，失敗<br>已嘗試：改變可利用的gpu memory、使用已經有的gradients accumulate、參考加入gradients checkpoints (https://github.com/cybertronai/gradient-checkpointing)
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/13
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
繼續重構python shell script, <br>觀摩很多github上的python大型專案
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
可以用snmp接到排插的資訊
</li>
<li>
裝好grafana，尋找好裝的 data source
</li>
<li>
graphite 有夠難裝，放棄
</li>
<li>
prometheus 裝好了，還不會用
</li>
<li>
跟 michael 討論後，明天裝 zabbix
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
試著寫數字辨識模型（基礎版本成功，研究如何把input變成自己的照片）最後成功!<img src = https://i.imgur.com/uMBew8c.png><br>
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>測不同的參數執行sst2
<img src="https://i.imgur.com/qvVeLcu.png">
<img src="https://i.imgur.com/F7uHSjY.png">
<img src="https://i.imgur.com/HfCwgRB.png">
<img src ="https://i.imgur.com/sLt7ZPf.png"><img src="https://i.imgur.com/681jPjM.png"><br>large目前就算batch size設1 ,max seq設2還是會out of memory，還要看一下是不是有其他錯誤<img src="https://i.imgur.com/leqBSRJ.png">
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>繼續在大機器上測試：<br>嘗試修改optimization.py，結果：<br>https://docs.google.com/spreadsheets/d/1kSCf28cJUiF4oeYhVHHRc52yMmvqrIFGcgfmHIJlYx0/edit?usp=sharing<br>似乎比原來更好<br>念《Statistics, Data Mining, and Machine Learning in Astronomy》
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/12
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
佈置好docker network
</li>
<li>
重構python shell script
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
把偵測GPU溫度的python做成service<br><a href="https://i.imgur.com/p1rnwGk.png"><img src="https://i.imgur.com/aKHPi9y.png
" width=550px></a>
</li>
<li>
找某個監控軟體
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
研究Mnist數字辨識example<br>
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>實測用同樣的顯卡後發現更改速度並沒有明顯變快，跑MNLI with 不同參數<br>下載NER的資料，確認除了NER以外，其他用來fine-tune的dataset都已經正常運作(MNLI,SST2,SQUAD)
<br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>繼續下載dataset今天外加ner的data<br>在大機器上測試：<br>嘗試修改optimization.py，結果：<br>https://docs.google.com/spreadsheets/d/1kSCf28cJUiF4oeYhVHHRc52yMmvqrIFGcgfmHIJlYx0/edit?usp=sharing
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/11
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
研究docker network
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
寫出偵測大機器GPU溫度的python(原型)<br><a href="https://i.imgur.com/p1rnwGk.png"><img src="https://i.imgur.com/p1rnwGk.png" width=400px></a>
</li>
<li>
電腦硬體小教室
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
寫出第一個tf小程式，跑成功<br>
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td><img src="https://i.imgur.com/05p8W8q.png">
<br>顯卡P100<br>run mnli mrpc squad<br>重新下載資料與作資料轉換
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>debug Nvidia 的 BERT 無法進行pretraining->原因資料沒有下載完成<br>在大機器上測試：<br>嘗試修改程式碼讓gpu利用率更好一些->在mnli上效果不怎麼好，結果跟昨天差不多<br>聽樓上上硬體的教學
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過1017人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/10
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
跟t2交流, 合作開發python專案<br>以下是UML以下是UML class diagram:
<img src="https://i.imgur.com/ImBgZkq.png">
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
成功從container裡面ping出來
</li>
<li>
樹梅派buzzer測試
</li>
<li>
T1-T2合作開發python專案
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
開始看一些tensorflow的資料<br>
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>驗證昨天的code修改，在較小的mrpc上只差一秒，mnli已完成未改code的training<br>已完成的code還沒測<br>大機器環境配置(完成)，fine-tuning BERT on MRPC and MNLI<br>result<br>acc=0.857,time=6m9s<br>train mnli，結果如下面樓下顯示，改shell script進行花費時間計算ㄖㄞ
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>Nvidia 的 BERT dockerfile配置完成<br>在大機器上測試：<br>1.跑MNLI，參數與google的相同，額外參數use_xla=False<br>2.同1.，use_xla=True<br>3.同2.，batch_size=64(原32)<br>1.<img src="https://i.imgur.com/gYaWHSY.png" width="400px" ><br> 2.<img src="https://i.imgur.com/0jmTUgm.png" width="400px"><br>3.<img src="https://i.imgur.com/lFNqlye.png" width="400px"><br>整體而言比colab好
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過638人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/9
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
繼續擴充python3 shell script, 等著有朝一日被做成docker image
</li>
<li>
研究docker
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
開 大 機 器
</li>
<li>
看geeksforgeeks學python
</li>
<li>
研究docker跨機
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
在n12裝nvidia driver跟cuda,用GPU跑hpcg成功！<br>
</li>
<li>
看Geeksforgeeks學拍森
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>優化google-bert的code，code的convert_one_example含式會在每次iteration的時後呼叫，我把它改成用iteration收集成list後一次呼叫<br>環境配置<br>搜尋state-of-art nlp看別人除了參數以外，如何修改模型
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>看google的bert的optimization.py，小筆記：<br>https://hackmd.io/NktzSsTnQmykFfRidFgfxA?view<br>並完成關於優化器的筆記：<br>https://hackmd.io/r_Z6SaNmSnajh8baK5NHEw?view#%E5%84%AA%E5%8C%96%E5%8F%83%E6%95%B8<br>在大機器上開始下載dataset、遠端(非docker)使用spyder</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過638人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/8
<table>
<thead>
<tr style="background-image: linear-gradient( 135deg, #92FFC0 10%, #002661 100%); color:#ffffff "
>
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
研究mpi的send/recv operation
</li>
<li>
完成python shell script, 包含單元測試的code <a href="https://github.com/rickywu0421/OneForAll">連結點我</a>
</li>
<li>
架構:<br>
<img src="https://i.imgur.com/36OuoK8.png" width=300px>
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
CESM簡報分割
</li>
<li>
看geeksforgeeks學python
</li>
<li>
大睡特睡
</li>
</ol>
</td>
</tr>
</tr>
<tr>
<td>趙胤筌<img src="https://media.giphy.com/media/3SKNDFV2pQyt2/giphy.gif" style="width:70px"></td>
<td>
<ol>
<li>
cesm報告版 畫圖<br><br><img src="https://imgur.com/s3kHlit.png" width=200px >
</li>
<li>
排版，美化簡報
</li>
<li>
學一些html,css語法
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>github閱讀，環境配置，nvidia-docker<br>資料下載(有503跟404,但官網說正常)<br>tfrecord理解，理解sst2 class processor(google github梅內建)<br>看別人寫的ner class
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>開始使用https://github.com/NVIDIA/DeepLearningExamples/tree/master/TensorFlow/LanguageModeling/BERT<br>成功build裡面的dockerfile，正在下載dataset，部分出現error 503<br>在先前的BERT上跑SST-2數據集，調整參數<br>參數調整(1)->batch size=16<br>參數調整(2)->optimization中tf.train.polynomial_decay的cycle=True<br>參數調整(3)->optimization中tf.train.polynomial_decay的power=2.0<br>(1)<img src="https://i.imgur.com/lNHrMdf.png" width="400px"><br>(2)<img src="https://i.imgur.com/xDHOsfb.png" width="400px"><br>(3)<img src="https://i.imgur.com/ijCZQJF.png" width="400px">
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過638人請小心 <br>
<img src="https://imgur.com/0dczO2Z.png" width="300px">
</td>
</tr>
</table>
<br>
## 2/7
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
研究mpi的send/recv operation
</li>
<li>
完成python shell script, 包含單元測試的code <a href="https://github.com/rickywu0421/OneForAll">連結點我</a>
</li>
<li>
架構:<br>
<img src="https://i.imgur.com/36OuoK8.png" width=300px>
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
systemctl enable CESM-CLM簡報，開始看巨量的document
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>看懂run_classify的processor class，並能根據文件定義class，run mnli<img src="https://i.imgur.com/Xi49fQI.png">
<br>參數--max_seq_length=128 \
--train_batch_size=8 \
--learning_rate=2e-5 \
--num_train_epochs=1.0 \
時間約8小時
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>看BERT的run_classify的processor，參考網上資料並加入原本沒有的ssf2processor<br>初步理解tfrecord</br>在先前的BERT上跑SST-2數據集，>在先前的BERT上跑SST-2數據集， 參數
--max_seq_length=128 \
--train_batch_size=8 \
--learning_rate=2e-5 \
--num_train_epochs=3.0 \ <br><img src="https://i.imgur.com/3Q09JYp.png">
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過563人請小心<img src="https://i.imgur.com/gorcjTm.png" width="70px"><img src="https://i.imgur.com/ET4txpe.jpg" width="70px">
</td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看RRRRRR
</td>
</td>
</tr>
</tbody>
</table>
<br>
## 2/6
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
研究mpi的send/recv operation
</li>
<li>
完成python shell script, 包含單元測試的code <a href="https://github.com/rickywu0421/OneForAll">連結點我</a>
</li>
<li>
架構:<br>
<img src="https://i.imgur.com/36OuoK8.png" width=300px>
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
systemctl enable CESM-CLM簡報，開始看巨量的document
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>看github的code(class mrpc，tokenize)，train squad dataset(約6小時)<br>result-><br>https://colab.research.google.com/drive/1R5DoKANmkE-X23UO2oAzgG5zPUEvzeRH#scrollTo=MEtSM2eKPn4x<br></br>--train_batch_size=8 \
--learning_rate=3e-5 \
--num_train_epochs=1.0 \
--max_seq_length=128\ \
--doc_stride=64 \
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>BERT嘗試在MNLI數據集上跑-><br> --max_seq_length=128 \
--train_batch_size=8 \
--learning_rate=2e-5 \
--num_train_epochs=3.0 \<br>總共147263 steps，跑了7小時左右<img src="https://i.imgur.com/xV5tcy6.png"><br>看BERT處理數據的code</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過563人請小心<img src="https://i.imgur.com/gorcjTm.png" width="70px"><img src="https://i.imgur.com/ET4txpe.jpg" width="70px">
</td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看RRRRRR
</td>
</td>
</tr>
</tbody>
</table>
<br>
## 2/5
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
完成python shell script, 包含單元測試的code <a href="https://github.com/rickywu0421/OneForAll">連結點我</a>
</li>
<li>
架構:<br>
<img src="https://i.imgur.com/36OuoK8.png" width=300px>
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
看geeksforgeeks學python
</li>
<li>
重灌四台，協助測試樓上的shell script
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>看github的code，train squad dataset(今天沒train完，有保留權重)<br>result-><br>https://colab.research.google.com/drive/1R5DoKANmkE-X23UO2oAzgG5zPUEvzeRH#scrollTo=MEtSM2eKPn4x<br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>更改BERT程式碼，嘗試在SST-2數據集上跑-><br><img src='https://i.imgur.com/3AAT0ew.png'></td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過491人請小心<img src="https://i.imgur.com/gorcjTm.png" width="70px"><img src="https://i.imgur.com/ET4txpe.jpg" width="70px">
</td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看RRRRR
</td>
</td>
</tr>
</tbody>
</table>
<br>
## 2/4
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
繼續完成python shell script, 包含單元測試的code
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
看geeksforgeeks學python，差不多先告一段落
</li>
<li>
開始看樓上的 python shell script，希望明天能寫出一點東西
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>BERT筆記，BERT試跑(MRPC)<br>result-><br>https://colab.research.google.com/drive/1R5DoKANmkE-X23UO2oAzgG5zPUEvzeRH#scrollTo=MEtSM2eKPn4x<br>eval_acc=0.86<br>mnli跑太久斷線
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>寫BERT的架構筆記：https://hackmd.io/xyEw7HcfQd6ZX73ibRuy3w<br>開始在MRPC數據集上跑BERT，準確率 acc 0.87，嘗試在MNLI數據集上跑->colab突然有bug...</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過425人請小心<img src="https://i.imgur.com/gorcjTm.png" width="70px"></td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看RRRR
</td>
</td>
</tr>
</tbody>
</table>
<br>
## 2/3
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
繼續完成python shell script, 包含單元測試的code
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
簡報有點沒動力，耍廢狀態...
</li>
<li>
看geeksforgeeks學python
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>看BERT的github，github筆記，看BERT影片<br></br>https://www.youtube.com/watch?v=UYPa347-DdE
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>看BERT的架構(未完成)：https://hackmd.io/xyEw7HcfQd6ZX73ibRuy3w</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已消失遠超過362人請小心<img src="https://i.imgur.com/gorcjTm.png" width="70px"></td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看RRR
</td>
</td>
</tr>
</tbody>
</table>
<br>
## 2/2
<table>
<thead>
<tr style="background-color:#ffa000" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
研究python project的目錄架構
</li>
<li>
繼續寫python shell
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
做CESM-MOSART簡報(曼寧公式、參數)，暫停中
</li>
<li>
做CESM-WW3簡報(找了一堆莫名奇妙的論文，最後發現還是只能回去看六百多頁的manual)
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>做CESM簡報(land ice，流體公式),處理dockerfile
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>還是繼續Dockerfile，解決昨天的問題—更改docker的主目錄位置，最新報錯：<img src="https://i.imgur.com/Kvzc4j6.png">
<br></br>繼續幫忙做CESM簡報(component sets、網格...) __(:_」∠)_</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已往生遠超過304人請小心<img src="https://i.imgur.com/gorcjTm.png" width="70px"></td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看RR
</td>
</tr>
</tbody>
</table>
<br>
## 2/1
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
yum remove 智齒(左)
</li>
<li>
del 智齒(右)
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
做CESM-MOSART簡報(找到一個不錯的網站，翻譯中)
</li>
<li>
看geeksforgeeks學python
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>處理ISC-19提供的Dockerfile，處理bazel的錯誤，最後卡在no space left on device，做CESM簡報
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>繼續昨天的Dockerfile，目前最後報錯：<img src="https://i.imgur.com/VT3N7mj.png">
<br></br>幫忙做CESM簡報</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已往生遠超過259人請小心<img src="https://i.imgur.com/WCnGllx.gif" width="70px"></td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">肛科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg" width="70px"></td>
<td>
再不來試試看R
</td>
</td>
</tr>
</tbody>
</table>
<br>
## 1/31
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
用python改寫bash一鍵安裝包(python is better!)<br>架構圖:<br><img src="https://i.imgur.com/U8bpd5j.png" height="300px"><br>
</li>
<li>
寫好自動架beegfs的script
</li>
<li>
本日收穫: bash的功能為下載更好的語言
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
在N11(CentOS)裝ChaNGa
</li>
<li>
重灌n22(CentOS7)
</li>
<li>
做CESM-WW3簡報(只有一點點...)
</li>
<li>
看geeksforgeeks學python
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>處理ISC-19提供的Dockerfile<br></br>
nccl cudnn tensorflowRT <br></br>
只差最後一步了<br></br>
FATAL: Failed to extract embedded binaries: Failed to write zipped file '/root/.cache/bazel/_bazel_root/install/8c9f2b2c4ae1e429ee70976939e025c5.tmp.21/A-server.jar': (error: 2): No such file or directory<br></br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>無限debug Tiramisu的Dockerfile，繼樓上的嘗試使用原來的配置，尚未出現之前的bug，同樣只差最後一步，但卡在Cloning into 'tensorflow'...(已知問題待解決)<br></br>重新下載有問題的壓縮檔，失敗，網路過慢自動斷線...</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已過世遠超過213人請小心</td>
</tr>
<thead>
<tr style="background-color:#FFB7DD; color:#ffffff" >
<th colspan=2 style="text-align:center">月工科🈺️2/2~2/8快解放了</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://twgreatdaily.com/images/elastic/9--/9--0AG4BMH2_cNUg-qHV.jpg"></td>
<<td>
<ol>
<li>
做CESM簡報(WW3)(幫其他的美化＋補充)
</li>
<li>
看ＷＷ3論文
</li>
<li>
在N21(centos)裝 changa(成功！)
</li>
</ol>
</td>
</tr>
</tbody>
</table>
<br>
## 1/30
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
真的成功build出執行檔(exe檔)
</li>
<li>
11
</li>
<li>
陪樓下被買耳機
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
做CESM-CLM5簡報(只有一點點...)
</li>
<li>
看莫煩、geeksforgeeks學python，剛學完一小段
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>ISC-19 github環境配置 -><br></br>nvidia-docker<br></br>cuda降版本<br></br>tensorflow<br></br>cudnn<br></br>horovod<br></br>training報錯-><br></br>壓縮檔不完全(FUCK，環境應該沒問題了)
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>開始試著執行Tiramisu->有很多問題解決中...(查看程式碼中)<br></br>裝CUDA 10.0和cuDNN並調整至程式所需的配置、nvidia-docker(Dockerfile build起來空間不夠)<br></br>還是沒辦法用的壓縮檔</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>_(:_」∠)_防疫模式<i class="fa fa-toggle-on fa-fw"></i> 已過世遠超過169人請小心</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">工科🈺️2/2~2/8快解放了</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<<td>
<ol>
<li>
做CESM簡報(CAM-chem)
</li>
<li>
看CAM-chem論文
</li>
<li>
學C語言(快告一段，準備進入莫煩)
</li>
<li>
報名駕訓班，以後當早起的鳥兒
</li>
</ol>
</td>
</tr>
</tbody>
</table>
<br>
## 1/21
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
真的成功build出執行檔(exe檔)
</li>
<li>
./case.submit(跑程式的指令)報錯:
<a href="https://i.imgur.com/YFDXKp0.png" target="_blank"><img src="https://i.imgur.com/YFDXKp0.png" width="500px"></a>
</li>
<li>
陪樓下被買耳機
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
處理該死的config_compiler.xml
</li>
<li>
<img src="https://i.imgur.com/a8PfhH4.png"></img>成功啦啦啦
</li>
<li>
陪樓上買耳機
</li>
<li>
n11過夜./case.submit
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>dataset底bug<br></br>->雲端解壓資料集會有問題<br></br>練習dataset處理<br></br>https://github.com/0bserver07/One-Hundred-Layers-Tiramisu/blob/master/camvid_data_loader.py<br></br>CESM簡報<br></br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>繼續處理Tiramisu要使用的dataset(雲端大問題)<br></br>了解Tiramisu100處理data的過程，https://github.com/0bserver07/One-Hundred-Layers-Tiramisu/blob/master/camvid_data_loader.py<br></br>做一些CESM的簡報，https://docs.google.com/presentation/d/1krKSoIZkTQNttN02EMYVCby28-Mv8nbh99Fui6RnX9E/edit?usp=sharing</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRRRRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/20
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
成功build出執行檔(exe檔)
</li>
<li>
為了跑執行檔, 又需要載一堆model, 我要下班了還沒載完
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
安裝ScaLAPACK
</li>
<li>
Elmer/Ice放置，重灌n11後再次安裝cesm，build快成功了，有執行檔
</li>
<li>
發現這兩天的錯誤都出在一個"$"上
</li>
<li>
協助開發T1共筆.project
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>dataset下載<br></br>問題-><br></br>1.github給的雲端為空<br></br>2.dataset合併有問題>dataset合併有問題<br></br>hdf5檔案處理<br></br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>處理Tiramisu要使用的dataset(大問題)<br></br>查怎麼使用hdf5檔案<br></br>!cat segm_h5_v3_new_split.tar.parta* > segm_h5_v3_new_split.tar</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRRRRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/19
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
在家弄專題
</li>
<li>
專題弄hadoop
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
Elmer/Ice放置，重灌n11後再次安裝cesm，build快成功了
</li>
<li>
發現這兩天的錯誤都出在一個"$"上
</li>
<li>
協助開發T1共筆.project
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>train 新的tiramisu<br></br>前天使用的github壞了<br></br>原github:<br></br>https://github.com/giovanniguidi/Tiramisu-keras.git<br></br>新github:<br></br>https://github.com/0bserver07/One-Hundred-Layers-Tiramisu.git<br></br>notebook: <br></br>
https://colab.research.google.com/drive/1j_MH5E5BYjTOzzR0IGPbg0CgTMmbXg88<br></br>result -> <br></br>150 epoch RMSprop acc:0.702
<br></br>52 epoch Adam acc:0.28<br></br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>托福 ITP<br></br>Deeplab 無使用權重訓練->頗慘<br></br>Tiramisu100用不同優化器訓練，目前嘗試SGD、Adam、RMSprop，SGD -> acc 58.63%，其他尚未訓練完<br></br>慢慢補充筆記中優化器的部分</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRRRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/19
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
在家弄專題
</li>
<li>
專題弄hadoop
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
對cesm進行放置play，嘗試裝elmer
</li>
<li>
用intel的各種東西編cmake、Hypre、mumps、LAPACK(make卡住中)
</li>
<li>
<a href="https://hackmd.io/93VIzy0TQBaV1S-UXeJt_w?view#%E5%AE%89%E8%A3%9Delmerice">elmer/ice筆記點這邊</a>
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>train 新的tiramisu<br></br>前天使用的github壞了<br></br>原github:<br></br>https://github.com/giovanniguidi/Tiramisu-keras.git<br></br>新github:<br></br>https://github.com/0bserver07/One-Hundred-Layers-Tiramisu.git<br></br>notebook: <br></br>
https://colab.research.google.com/drive/1j_MH5E5BYjTOzzR0IGPbg0CgTMmbXg88<br></br>result -> <br></br>150 epoch RMSprop acc:0.702
<br></br>52 epoch Adam acc:0.28<br></br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>托福 ITP<br></br>Deeplab 無使用權重訓練->頗慘<br></br>Tiramisu100用不同優化器訓練，目前嘗試SGD、Adam、RMSprop，SGD -> acc 58.63%，其他尚未訓練完<br></br>慢慢補充筆記中優化器的部分</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRRRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/18
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
去看牙醫(痛)+感冒, 已哭
</li>
<li>
下載不同的compset, 但還是build失敗
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
重灌n11兩次，都在編譯NCetCDF-Fortran失敗
</li>
<li>
寫安裝cesm前置的一鍵安裝包，<a href="http://192.168.1.13/something_about_cesm/before-cesm.sh">點這邊</a>
</li>
<li>
檢查make check，發現在NetCDF-C就出錯了
</li>
<li>
繼續崩潰RRRRRR<img src="https://i.imgur.com/LEGaxaE.gif" style="width:60px">
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>複習之前的練習<br></br>在n21嘗試裝CESM<br></br>ai筆記(mask-rcnn)><br></br>不使用pretrain weight <br></br>
number of step 30000 ,batch size 4,img size = 513,513<br></br>result -> <br></br><img src="https://i.imgur.com/HuKHwtX.png" >
<br></br>時間約6個半小時<br></br>但結果頗差<br></br>
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>看有關目標檢測和實例分割的模型(RNN系列與YOLO)並完善之前的筆記<br></br>Deeplab在Cityscapes上進行22108+3668+16474 steps(中間colab有斷連，但有保留check points)的測試，有使用pretrain weight：miou_1.0[0.770881474](好像沒有比較好)。https://drive.google.com/open?id=13qCGKLMcH-PLvcJA8qfGLNsJHO0xgdUJ (好像有比較好)<br></br>看一點CSEM
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/17
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
成功編譯netcdf-c, netcdf-fortran, pnetcdf, pio
</li>
<li>
自行編譯curl, 解決curl問題
</li>
<li>
cesm build在buildlib.csm_share時報錯
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
成功編譯netcdf-c, netcdf-fortran, pnetcdf, pio
</li>
<li>
手動裝curl，解決昨天3.的錯誤
</li>
<li>
嘗試在n12安裝及編譯pio，發生錯誤:cannot find -lcurl
<br>推測是LD_LIBRARY_PATH出問題
</li>
<li>
cesm的case.build有新的錯誤訊息:
<br>某個fortran檔裡面:
<br>Procedure name in MODULE PROCEDURE statement must be the name of accessible module procedure
</li>
<li>
繼續崩潰<img src="https://i.imgur.com/LEGaxaE.gif" style="width:60px">
</li>
<li>
學BASH
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>ai筆記(找了些物件偵測model)<br></br>成功訓練deeplab<br></br>使用pretrain weight <br></br>
number of step 5000 ,batch size 4,img size = 513,513<br></br>result -> <br></br><img src="https://i.imgur.com/Evby0Cq.png" >
<br></br>出乎意料的優秀<br></br>不使用pretrain weight 參數一樣-><br></br><img src="https://i.imgur.com/MOIDucv.png" ><br></br>結果圖片:<br></br>https://drive.google.com/open?id=1-7Mv7n2gwLyPOizgECns2E-FK0varUiI
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>看有關目標檢測和實例分割的模型(RNN系列與YOLO)並完善之前的筆記<br></br>Deeplab在Cityscapes上進行22108+3668+16474 steps(中間colab有斷連，但有保留check points)的測試，有使用pretrain weight：miou_1.0[0.770881474](好像沒有比較好)。https://drive.google.com/open?id=13qCGKLMcH-PLvcJA8qfGLNsJHO0xgdUJ (好像有比較好)<br></br>看一點CSEM
</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/16
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
寫出一鍵重灌的shell script: <a href="http://192.168.1.13/2020-01-16.sh">點我看code</a>
</li>
<li>
編譯netcdf-c時報錯, 尋找解決方法中,<br>目前測試centos-7, centos-8, rhel-7均會報錯<br>(原本都不會,突然就這樣了, 原因貌似是python版本問題)
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
重灌n21:兩次CentOS7、一次CentOS8
</li>
<li>
找出造成icc的python版本及I_MPI_ROOT錯誤的元兇：Development Tools
</li>
<li>
CentOS7:NetCDF-C configure: error: curl (makefile)equired
<br>CentOS8:HDF5 configure: error: C compiler cannot create executables
<br>推測(CentOS8)是MPI問題
</li>
<li>
更崩潰<img src="https://i.imgur.com/LEGaxaE.gif" style="width:60px">
</li>
<li>
學BASH
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>ai筆記<br></br>解決無限迴圈bug<br></br>->使用shell script執行py檔就不會無限迴圈了<br></br>
ipynb:<br></br>https://colab.research.google.com/drive/1h4BX43oSPkHfhfAzUSQ7viCqYJxwM3mw<br></br>result(3000 step numbers, batch size = 8, img_size = 299) -> <br></br><img src="https://i.imgur.com/VB05mN9.png" >
<br></br>頗慘，還得確定是執行上出了問題還是只是train不夠久
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>嘗試跑deeplab，仍在嘗試中並尋找解決方法。使用程式碼：https://github.com/tensorflow/models/tree/master/research/deeplab<br></br>另外100層的Tiramisu似乎效果不太好</td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/15
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
利用env2生成icc的modulefile
</li>
<li>
安裝及編譯HDF5、NetCDF、PNetCDF、PIO
<li>
改config_compiler.xml
</li>
<li>
變色龍超連結
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:70px"></td>
<td>
<ol>
<li>
重灌n21、n22各兩次，n11一次，目前四台都是CentOS7.7
</li>
<li>
在11、21、22重新嘗試build昨天的PnetCDF，在configure HDF5和NetCDF發生的錯誤。
</li>
<li>
回頭檢查用env2生成icc的modulefile，發現很大的問題，用不同的電腦和intel mpi load前後有差，整理如下：
<br><img src="https://i.imgur.com/G5a7G2P.png" style="width:450px"></img>
</li>
<li>
崩潰<img src="https://i.imgur.com/LEGaxaE.gif" style="width:60px">
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>ai筆記<br></br>使用github上的deeplabv3+<br></br>github:<br></br>https://github.com/tensorflow/models/tree/master/research/deeplab<br></br>
ipynb:<br></br>https://colab.research.google.com/drive/1h4BX43oSPkHfhfAzUSQ7viCqYJxwM3mw<br></br>deeplab架構&&筆記<br></br>
trainging遇到的bug -> 訓練權重無法儲存(fixed)<br></br>待解決問題 -> 測試結果可視化程式碼無限迴圈
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>完成Deeplab V3+筆記：<br>https://hackmd.io/r_Z6SaNmSnajh8baK5NHEw#DeepLab</br><br>使用"The One Hundred Layers Tiramisu"的100層程式碼並運行，由於中途colab斷連，導致結果怪怪的(跑了快8個小時...)：https://colab.research.google.com/drive/1dlDPmfluplrQ-mXJPHYgHeKrN1EihPrg</br><br>accuracy: 0.61</br></td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C#, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RRR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/14
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td>吳恩緯<img src="https://img.icons8.com/color/48/000000/ruby-programming-language.png"></td>
<td>
<ol>
<li>
利用env2生成icc的modulefile
</li>
<li>
安裝及編譯HDF5、NetCDF、PNetCDF、PIO
<li>
改config_compiler.xml
</li>
<li>
變色龍超連結
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴<img src="https://i.imgur.com/5nJ4Lwt.gif" style="width:75px"></td>
<td>
<ol>
<li>
利用env2生成icc的modulefile
</li>
<li>
安裝及編譯HDF5、NetCDF、PNetCDF、PIO
<li>
改config_compiler.xml
</li>
<li>
./case.build錯誤訊息: CIME models require NETCDF in PIO build
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源<img src="https://i.imgur.com/yreyYYM.png" style="width:50px"></td>
<td>ai筆記<br></br>使用tiramisu model train VOC2012 dataset result <br></br>https://colab.research.google.com/drive/1NNe3l9-i_670EguXB1bx6BH4wJMxB_Or<br></br>
最好結果 <br></br>
pixel accuracy: 0.77<br></br>
mean accuracy: 0.14<br></br>
mean IoU: 0.11<br></br>
freq weighted mean IoU: 0.6<br></br> DeepLabV3+(未完成)<br></br>https://github.com/tensorflow/models
</td>
</tr>
<tr>
<td>陳韋安<img src="https://img.icons8.com/color/48/000000/python.png"></td>
<td>完成Tiramisu筆記：<br>https://hackmd.io/r_Z6SaNmSnajh8baK5NHEw?both#Tiramisu</br><br>使用"The One Hundred Layers Tiramisu"的67層程式碼並運行，結果(最下面，並附上GPU訊息)：https://colab.research.google.com/drive/1dlDPmfluplrQ-mXJPHYgHeKrN1EihPrg</br><br>accuracy: 0.65</br></td>
</tr>
<thead>
<tr style="background-color:red; color:#FFFF00" >
<th colspan=2 style="text-align:center">回祖國否放假</th>
</tr>
</tbody>
</thead>
<tr>
<td>丁浩然<img src="https://img.icons8.com/color/48/000000/swift.png"></td>
<td>用手機學C++, ASCII Code</td>
</tr>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌<img src="https://img.icons8.com/color/48/000000/c-programming.png"></td>
<td>工科營 覺得變色龍很帥！<span style="font-size:5px">再不來試試看RR</span></td>
</tr>
</tbody>
</table>
<br>
## 1/13
<table>
<thead>
<tr style="background-color:#ffa000; color:" >
<th colspan=2 style="text-align:center">工人</th>
</tr>
</thead>
<tbody>
<tr>
<td >吳恩緯</td>
<td>
<ol>
<li>
重灌centos7.7到n11、n12
</li>
<li>
./create_newcase成功
</li>
<li>
./check_case，inputdata下載中(過夜下載，約10幾G)
</li>
</ol>
</td>
</tr>
<tr>
<td>薛瑾澴</td>
<td>
<ol>
<li>
重灌centos7.7到n11、n12
</li>
<li>
./create_newcase成功
</li>
<li>
./check_case，inputdata下載中(過夜下載，約10幾G)
</li>
</ol>
</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#00aa00; color:#c0c0c0" >
<th colspan=2 style="text-align:center">人工</th>
</tr>
</thead>
<tbody>
<tr>
<td>黃有源</td>
<td>看tiramisu架構，找可以用的github
https://github.com/Tiramisu-Compiler/tiramisu.git(失敗)
https://github.com/giovanniguidi/Tiramisu-keras.git(keras 成功)
weight訓練完畢，待測試</td>
</tr>
<tr>
<td>陳韋安</td>
<td>找keras的Tiramisu模型，找到論文"The One Hundred Layers Tiramisu"和其代碼並運行，結果：https://colab.research.google.com/drive/1dlDPmfluplrQ-mXJPHYgHeKrN1EihPrg</td>
</tr>
<tr>
<td>丁浩然</td>
<td>用手機學C，尋找白卡</td>
</tr>
</tbody>
<thead>
<tr style="background-color:#0000FF; color:#c0c0c0" >
<th colspan=2 style="text-align:center">月工科營</th>
</tr>
</thead>
<tbody>
<tr>
<td>趙胤筌</td>
<td>工科營<span style="font-size:5px">再不來試試看R</span></td>
</tr>
</tbody>
</table>
## 12/28
**吳恩緯**：成功build cesm及build dockerfile
**薛瑾澴**：從模板修改config_machines.xml，再次./create_newcase，n22失敗。看CESM教學。
**黃有源**：看Tiramisu論文，找github上的Tiramisu模型測試，遇到問題->pytorch版本不符
**陳韋安**：試用看看Xception模型(意外發現相當穩定，準確率常保0.8左右)，看Tiramisu論文，在github上找keras的Tiramisu使用->被colab搞瘋，先暫停
**丁浩然**：繼續自己搭建模型進行訓練 $\&\&$ 看Tiramisu論文 $\&\&$ 嘗試train一個Tiramisu模型（失敗了
**趙胤筌**：
## 12/27
**吳恩緯**：
**薛瑾澴**：(1)./case.build報錯，發現singularity內的ubuntu有 LibXML.pm 
嘗試複製到@INC內後，發生相似錯誤：Can't locate loadable object for module XML::LibXML in @INC，暫停
(2)在singularity的ubuntu內安裝CESM(嘗試中)
**黃有源**：看物件偵測模型(會框出框框的)，今日參數測試最優結果:inceptionV3 SGD lr=1e-2 decay 0.005 batch_size 128 使用pretrain weight accuracy 0.886
**陳韋安**：(請假，fourier series)
**丁浩然**：模型優化暫時結束InceptionV3的準確率上0.92了！！！嘗試自己搭建模型並train（可惜暫時效果 emmmmm 有待改進
**趙胤筌**：在N11(RHEL)安裝singilarity 複習singularity操作 繼續學Ｃ
## 12/26
**吳恩緯**：看CESM doc, 用lustre跑io500
**薛瑾澴**：安裝CESM，./case.build似乎成功，setsid下載中...
**黃有源**：看pytorch多機教學，用pytorch測試inceptionV3，調整learing rate(最好 10個epoch內 lr = $10^{-1}$ 0.708，lr = $10^{-4}$速度太慢)。
**陳韋安**：繼續用VGG16跑caltech256並記錄結果(改變img_size)，嘗試不同的優化器(Adamax、Nadam)，突破0.75。
**丁浩然**：繼續調整inceptionV3在Caltech256 Dataset上的表現，測試準確率穩定在0.9以上。
**趙胤筌**： 練習Ｃ語言 複習cesm安裝
## 12/25
**吳恩緯**：
**薛瑾澴**：研究怎麼裝CESM，修改config_machines.xml，create_newcase後做到case.build報錯:
**黃有源**：抵inceptionV3在pytorch的bug，比較使用pretrain weight的效果，比較keras與pytorch執行差異
**陳韋安**：看一些inceptionV3的架構，繼續用VGG16跑caltech256並記錄結果(改變learning_rate)
> 紀錄：https://docs.google.com/spreadsheets/d/1jhuKokJwgAaO9Hhwjjwv-PbejDNqx06xtnVO_vt93YA/edit?usp=sharing
>
**丁浩然**：嘗試更改InceptionV3的全連接層以優化model；測試ResNet152_V2的表現。
**趙胤筌**：
## 12/24
吳恩緯：安裝CESM, 看document
薛瑾澴：💤回憶docker&singularity，安裝CESM，create失敗，N22開機失敗(無BIOS)
黃有源：裝CUDA，比較keras,pytorch執行差異(keras準確率一直較pytorch高，在找原因)
陳韋安：裝CUDA(紀錄筆記)，用VGG16跑caltech256並記錄結果
丁浩然：成功使caltech256 dataset上的測試準確率5次超過90%.(model: inception V3)
趙胤筌：初探CESM，安裝
## 12/23
**吳恩緯**：<font size=72 color=#********ff0000>**L**</font> | 成功測試lustre跑io500
**薛瑾澴**：**U**|看上面那位教學Lustre，跑io500
**黃有源**：**S**|看Lustre教學，學習keras
**陳韋安**：**T**|看Lustre教學，用Resnet152跑caltech101並記錄結果
**丁浩然**：**R**|看Lustre教學，成功將caltech256 dataset上的測試準確率逼近90%
$\quad\ \quad$**趙<font size=5
color=#********aac006>**e**</font>筌**：
<i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-chain"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i><i class="fa fa-camera"></i>
WI-FI <i class="fa fa-toggle-on fa-fw"></i>
- [ ]
- [ ]
## 12/21
**吳恩緯**: 研讀lustre manual
**薛瑾澴**: 請假
**黃有源**: 看pytorch切割訓練集與驗證集，trainer的寫法，調整參數，把拼圖裱框。
**陳韋安**: 由csv檔導入資料訓練、在Resnet152上修改參數並紀錄其結果、把拼圖裱框。
**丁浩然**: 調inceptionV3參數並記錄結果; 把拼圖裱框。
**趙胤筌**: 太晚起床直接回台北了；沒把拼圖裱框
Google Drive
Gist
Clipboard
Sign in with Wallet
