[GitHub] sxjscience commented on a change in pull request #10225: fix word language model script and readme

[GitBox <gi...@apache.org>]

sxjscience commented on a change in pull request #10225: fix word language model script and readme
URL: https://github.com/apache/incubator-mxnet/pull/10225#discussion_r176860732
 
 

 ##########
 File path: example/gluon/word_language_model/README.md
 ##########
 @@ -1,67 +1,59 @@
 # Word-level language modeling RNN
 
-This example trains a multi-layer RNN (Elman, GRU, or LSTM) on Penn Treebank (PTB) language modeling benchmark.
+This example trains a multi-layer RNN (Elman, GRU, or LSTM) on WikiText-2 language modeling benchmark.
 
-The model obtains the state-of-the-art result on PTB using LSTM, getting a test perplexity of ~72.
-And ~97 ppl in WikiText-2, outperform than basic LSTM(99.3) and reach Variational LSTM(96.3).
+The model obtains ~97 ppl in WikiText-2 using LSTM, outperforming basic LSTM(99.3) and reach Variational LSTM(96.3).
 
-The following techniques have been adopted for SOTA results: 
+The following techniques have been adopted for SOTA results:
 - [LSTM for LM](https://arxiv.org/pdf/1409.2329.pdf)
 - [Weight tying](https://arxiv.org/abs/1608.05859) between word vectors and softmax output embeddings
 
 ## Data
 
-### PTB
-
-The PTB data is the processed version from [(Mikolov et al, 2010)](http://www.fit.vutbr.cz/research/groups/speech/publi/2010/mikolov_interspeech2010_IS100722.pdf):
-
-```bash
-bash get_ptb_data.sh
-python data.py
-```
-
 ### Wiki Text
 
-The wikitext-2 data is downloaded from [(The wikitext long term dependency language modeling dataset)](https://www.salesforce.com/products/einstein/ai-research/the-wikitext-dependency-language-modeling-dataset/):
-
-```bash
-bash get_wikitext2_data.sh
-```
+The wikitext-2 data is from [(The wikitext long term dependency language modeling dataset)](https://www.salesforce.com/products/einstein/ai-research/the-wikitext-dependency-language-modeling-dataset/). The training script automatically loads the dataset into `$PWD/data`.
 
 
 ## Usage
 
 Example runs and the results:
 
 ```
-python train.py -data ./data/ptb. --cuda --tied --nhid 650 --emsize 650 --dropout 0.5        # Test ppl of 75.3 in ptb
-python train.py -data ./data/ptb. --cuda --tied --nhid 1500 --emsize 1500 --dropout 0.65      # Test ppl of 72.0 in ptb
 
 Review comment:
   Suggest we should not remove the PTB experiments. I think it's originally included in the gluon example.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services