以下Seq2Seq模型的預期培訓時間是多少？

Question

你好，我想測試下Seq2Seq模型，以獲得一個聊天機器人，我下面這個教程：

http://suriyadeepan.github.io/2016-06-28-easy-seq2seq/

這是主要代碼：

https://github.com/suriyadeepan/easy_seq2seq

我遇到的問題是在訓練時間之後，下載適當的語料庫後，我跑了以下代碼進行訓練：

python execute.py 

按照存儲庫的指示，模型變成了火車，這是輸出，主要問題是我的電腦已經計算了大約2天9小時的結果，使用了所有的處理器，我的電腦的規格有以下內容：

Processors: Intel® Core™ i7-6600U CPU @ 2.60GHz × 4 

Ram: 15.3 GiB 

考慮這個事實，我想感激的人是培養這種模式，如果我有一種錯誤的，或者如果它是正常的，因爲是一個非常複雜的模型，除了要知道反饋如果我的電腦能夠計算這個數據，下面是我得到的輸出，非常感謝支持，

python3 execute.py 

>> Mode : train 

Preparing data in working_dir/ 
Creating 3 layers of 256 units. 
Created model with fresh parameters. 

Reading development and training data (limit: 0). 
global step 300 learning rate 0.5000 step-time 2.58 perplexity 64.59 
    eval: bucket 0 perplexity 75.38 
    eval: bucket 1 perplexity 56.04 
    eval: bucket 2 perplexity 110.91 
    eval: bucket 3 perplexity 92.75 
global step 600 learning rate 0.5000 step-time 2.22 perplexity 8.04 
    eval: bucket 0 perplexity 186.31 
    eval: bucket 1 perplexity 348.11 
    eval: bucket 2 perplexity 286.05 
    eval: bucket 3 perplexity 246.06 
global step 900 learning rate 0.5000 step-time 2.43 perplexity 2.22 
    eval: bucket 0 perplexity 353.47 
    eval: bucket 1 perplexity 851.75 
    eval: bucket 2 perplexity 1252.18 
    eval: bucket 3 perplexity 1092.34 
global step 1200 learning rate 0.5000 step-time 2.51 perplexity 1.27 
    eval: bucket 0 perplexity 2439.94 
    eval: bucket 1 perplexity 4914.90 
    eval: bucket 2 perplexity 4302.70 
    eval: bucket 3 perplexity 4757.61 
global step 1500 learning rate 0.5000 step-time 2.81 perplexity 1.11 
    eval: bucket 0 perplexity 8616.23 
    eval: bucket 1 perplexity 5605.63 
    eval: bucket 2 perplexity 7266.88 
    eval: bucket 3 perplexity 18350.05 
global step 1800 learning rate 0.5000 step-time 2.77 perplexity 1.10 
    eval: bucket 0 perplexity 5445.96 
    eval: bucket 1 perplexity 23896.49 
    eval: bucket 2 perplexity 34089.69 
    eval: bucket 3 perplexity 18601.78 
global step 2100 learning rate 0.5000 step-time 2.66 perplexity 1.01 
    eval: bucket 0 perplexity 13390.66 
    eval: bucket 1 perplexity 14239.79 
    eval: bucket 2 perplexity 62781.50 
    eval: bucket 3 perplexity 66383.43 
global step 2400 learning rate 0.5000 step-time 2.75 perplexity 1.01 
    eval: bucket 0 perplexity 16025.10 
    eval: bucket 1 perplexity 19353.18 
    eval: bucket 2 perplexity 50499.01 
    eval: bucket 3 perplexity 22968.12 
global step 2700 learning rate 0.5000 step-time 2.75 perplexity 1.15 
    eval: bucket 0 perplexity 9214.54 
    eval: bucket 1 perplexity 9529.81 
    eval: bucket 2 perplexity 19052.16 
    eval: bucket 3 perplexity 12740.78 
global step 3000 learning rate 0.4950 step-time 2.54 perplexity 1.03 
    eval: bucket 0 perplexity 18002.15 
    eval: bucket 1 perplexity 48698.23 
    eval: bucket 2 perplexity 56023.42 
    eval: bucket 3 perplexity 43504.27 
global step 3300 learning rate 0.4950 step-time 2.77 perplexity 1.01 
    eval: bucket 0 perplexity 11827.19 
    eval: bucket 1 perplexity 37759.41 
    eval: bucket 2 perplexity 54461.78 
    eval: bucket 3 perplexity 25944.24 
global step 3600 learning rate 0.4950 step-time 2.43 perplexity 1.01 
    eval: bucket 0 perplexity 16221.68 
    eval: bucket 1 perplexity 73671.18 
    eval: bucket 2 perplexity 284799.78 
    eval: bucket 3 perplexity 119904.67 
global step 3900 learning rate 0.4950 step-time 1.88 perplexity 1.01 
    eval: bucket 0 perplexity 24126.39 
    eval: bucket 1 perplexity 65459.55 
    eval: bucket 2 perplexity 42027.96 
    eval: bucket 3 perplexity 73571.20 
global step 4200 learning rate 0.4950 step-time 2.36 perplexity 1.01 
    eval: bucket 0 perplexity 69183.19 
    eval: bucket 1 perplexity 69995.42 
    eval: bucket 2 perplexity 102648.55 
    eval: bucket 3 perplexity 139732.95 
global step 4500 learning rate 0.4950 step-time 2.34 perplexity 1.01 
    eval: bucket 0 perplexity 23524.59 
    eval: bucket 1 perplexity 63201.23 
    eval: bucket 2 perplexity 143448.13 
    eval: bucket 3 perplexity 215924.14 
global step 4800 learning rate 0.4950 step-time 2.32 perplexity 1.21 
    eval: bucket 0 perplexity 14127.02 
    eval: bucket 1 perplexity 22433.28 
    eval: bucket 2 perplexity 56531.84 
    eval: bucket 3 perplexity 24848.56 
global step 5100 learning rate 0.4901 step-time 2.36 perplexity 1.02 
    eval: bucket 0 perplexity 17618.08 
    eval: bucket 1 perplexity 40156.18 
    eval: bucket 2 perplexity 43300.34 
    eval: bucket 3 perplexity 58052.43 
global step 5400 learning rate 0.4901 step-time 3.02 perplexity 1.00 
    eval: bucket 0 perplexity 22818.83 
    eval: bucket 1 perplexity 23717.10 
    eval: bucket 2 perplexity 170402.32 
    eval: bucket 3 perplexity 59760.11 
^Z 
[1]+ Stopped     python3 execute.py 
adolfo@adolfo-Latitude-E5570:~/Downloads/easy_seq2seq-master$ fg 
python3 execute.py 
^Z 
[1]+ Stopped     python3 execute.py 
adolfo@adolfo-Latitude-E5570:~/Downloads/easy_seq2seq-master$ fg 
python3 execute.py 
global step 5700 learning rate 0.4901 step-time 13.76 perplexity 1.00 
    eval: bucket 0 perplexity 19748.73 
    eval: bucket 1 perplexity 62520.70 
    eval: bucket 2 perplexity 49733.03 
    eval: bucket 3 perplexity 97241.32 
global step 6000 learning rate 0.4901 step-time 2.40 perplexity 1.00 
    eval: bucket 0 perplexity 22433.97 
    eval: bucket 1 perplexity 37075.54 
    eval: bucket 2 perplexity 129078.26 
    eval: bucket 3 perplexity 115380.06 
global step 6300 learning rate 0.4901 step-time 2.15 perplexity 1.00 
    eval: bucket 0 perplexity 17475.21 
    eval: bucket 1 perplexity 68835.76 
    eval: bucket 2 perplexity 67453.78 

Answer 1

在CPU上訓練深層模型需要永久。如果你打算真正使用深度學習技術，你必須得到一個gpu或者使用預訓練，然後我會推薦一個gpu，因爲預測速度會更快。