[onmt] Update README with models; move data to AWS

bellamei · Feb 23, 2017 · c84f7df · c84f7df
1 parent 6be19f9
commit c84f7df
Show file tree

Hide file tree

Showing 15 changed files with 14 additions and 62,949 deletions.
diff --git a/.gitignore b/.gitignore
@@ -1,3 +1,4 @@
 dcgan/data
 data
 *.pyc
+OpenNMT/data
diff --git a/OpenNMT/README.md b/OpenNMT/README.md
@@ -10,6 +10,10 @@ an open-source (MIT) neural machine translation system.
 
 OpenNMT consists of three commands:
 
+0) Download the data.
+
+```wget https://s3.amazonaws.com/pytorch/examples/opennmt/data/onmt-data.tar && tar -xf onmt-data.tar```
+
 1) Preprocess the data.
 
 ```python preprocess.py -train_src data/src-train.txt -train_tgt data/tgt-train.txt -valid_src data/src-val.txt -valid_tgt data/tgt-val.txt -save_data data/demo```
@@ -22,6 +26,13 @@ OpenNMT consists of three commands:
 
 ```python translate.py -cuda -model model_e13_*.pt -src data/src-test.txt -tgt data/tgt-test.txt -replace_unk -verbose```
 
+## Pretrained Models
+
+The following pretrained models can be downloaded and used with translate.py.
+
+- [onmt_model_en_de_200k](https://s3.amazonaws.com/pytorch/examples/opennmt/models/onmt_model_en_de_200k-4783d9c3.pt): An English-German translation model based on the 200k sentence dataset at [OpenNMT/IntegrationTesting](https://github.com/OpenNMT/IntegrationTesting/tree/master/data). Perplexity: 21. 
+- [onmt_model_en_fr_b1M](https://s3.amazonaws.com/pytorch/examples/opennmt/models/onmt_model_en_fr_b1M-261c69a7.pt): An English-French model trained on benchmark-1M. Perplexity: 4.85.
+
 ## Release Notes
 
 The following OpenNMT features are implemented:

diff --git a/OpenNMT/data/src-test-case.txt b/OpenNMT/data/src-test-case.txt
diff --git a/OpenNMT/data/src-test.txt b/OpenNMT/data/src-test.txt
diff --git a/OpenNMT/data/src-train-case.txt b/OpenNMT/data/src-train-case.txt
diff --git a/OpenNMT/data/src-train.txt b/OpenNMT/data/src-train.txt
diff --git a/OpenNMT/data/src-val-case.txt b/OpenNMT/data/src-val-case.txt
diff --git a/OpenNMT/data/src-val.txt b/OpenNMT/data/src-val.txt
diff --git a/OpenNMT/data/tgt-test-case.txt b/OpenNMT/data/tgt-test-case.txt
diff --git a/OpenNMT/data/tgt-test.txt b/OpenNMT/data/tgt-test.txt
diff --git a/OpenNMT/data/tgt-train-case.txt b/OpenNMT/data/tgt-train-case.txt
diff --git a/OpenNMT/data/tgt-train.txt b/OpenNMT/data/tgt-train.txt
diff --git a/OpenNMT/data/tgt-val-case.txt b/OpenNMT/data/tgt-val-case.txt
diff --git a/OpenNMT/data/tgt-val.txt b/OpenNMT/data/tgt-val.txt
diff --git a/OpenNMT/onmt/modules/GlobalAttention.py b/OpenNMT/onmt/modules/GlobalAttention.py
@@ -24,6 +24,7 @@
 import torch.nn as nn
 import math
 
+_INF = float('inf')
 
 class GlobalAttention(nn.Module):
     def __init__(self, dim):
@@ -47,7 +48,7 @@ def forward(self, input, context):
         # Get attention
         attn = torch.bmm(context, targetT).squeeze(2)  # batch x sourceL
         if self.mask is not None:
-            attn.data.masked_fill_(self.mask, -math.inf)
+            attn.data.masked_fill_(self.mask, -_INF)
         attn = self.sm(attn)
         attn3 = attn.view(attn.size(0), 1, attn.size(1))  # batch x 1 x sourceL