remove comments and replace original paths with generic paths
Browse files- bpe-en-gl_emb.yaml +24 -44
bpe-en-gl_emb.yaml
CHANGED
@@ -8,63 +8,48 @@ overwrite: True
|
|
8 |
# Corpus opts:
|
9 |
data:
|
10 |
europarl:
|
11 |
-
path_src:
|
12 |
-
path_tgt:
|
13 |
transforms: [bpe, filtertoolong]
|
14 |
weight: 120
|
15 |
opensub:
|
16 |
-
path_src:
|
17 |
-
path_tgt:
|
18 |
transforms: [bpe, filtertoolong]
|
19 |
weight: 152
|
20 |
opus:
|
21 |
-
path_src:
|
22 |
-
path_tgt:
|
23 |
transforms: [bpe, filtertoolong]
|
24 |
weight: 160
|
25 |
ted2020:
|
26 |
-
path_src:
|
27 |
-
path_tgt:
|
28 |
transforms: [bpe, filtertoolong]
|
29 |
weight: 10
|
30 |
corgaback:
|
31 |
-
path_src:
|
32 |
-
path_tgt:
|
33 |
transforms: [bpe, filtertoolong]
|
34 |
weight: 15
|
35 |
ccmatrix:
|
36 |
-
path_src:
|
37 |
-
path_tgt:
|
38 |
transforms: [bpe, filtertoolong]
|
39 |
-
weight: 380
|
40 |
wikimatrix:
|
41 |
-
path_src:
|
42 |
-
path_tgt:
|
43 |
transforms: [bpe, filtertoolong]
|
44 |
-
weight: 70
|
45 |
cluvi:
|
46 |
-
path_src:
|
47 |
-
path_tgt:
|
48 |
transforms: [bpe, filtertoolong]
|
49 |
-
weight: 70
|
50 |
-
#wikimedia:
|
51 |
-
# path_src: ../DGTcorpora_tokenized/en_gl/wikimedia/en.txt
|
52 |
-
#path_tgt: ../DGTcorpora_tokenized/en_gl/wikimedia/gl.txt
|
53 |
-
#transforms: [bpe, filtertoolong]
|
54 |
-
#weight: 4
|
55 |
-
# xlent:
|
56 |
-
#path_src: ../DGTcorpora_tokenized/en_gl/xlent/en.txt
|
57 |
-
#path_tgt: ../DGTcorpora_tokenized/en_gl/xlent/gl.txt
|
58 |
-
#transforms: [bpe, filtertoolong]
|
59 |
-
#weight: 50 #25000000/1600000=15; 760/15=50
|
60 |
-
#linux:
|
61 |
-
#path_src: ../DGTcorpora_tokenized/en_gl/linux/en.txt
|
62 |
-
#path_tgt: ../DGTcorpora_tokenized/en_gl/linux/gl.txt
|
63 |
-
#transforms: [bpe, filtertoolong]
|
64 |
-
#weight: 20 #25000000/150000=166; 760/166=5 * 5 = 20
|
65 |
valid:
|
66 |
-
path_src:
|
67 |
-
path_tgt:
|
68 |
transforms: [bpe, filtertoolong]
|
69 |
|
70 |
### Transform related opts:
|
@@ -73,8 +58,7 @@ src_subword_model: ./bpe/en.code
|
|
73 |
tgt_subword_model: ./bpe/gl.code
|
74 |
src_subword_vocab: ./run/bpe.vocab.src
|
75 |
tgt_subword_vocab: ./run/bpe.vocab.tgt
|
76 |
-
|
77 |
-
#tgt_subword_model: ../sentencepiece/en-gl/gl.sp.model
|
78 |
src_subword_type: bpe
|
79 |
tgt_subord_type: bpe
|
80 |
|
@@ -97,7 +81,7 @@ tgt_embeddings: ../embeddings/gl.emb.txt
|
|
97 |
embeddings_type: "word2vec"
|
98 |
|
99 |
# word_vec_size need to match with the pretrained embeddings dimensions
|
100 |
-
word_vec_size:
|
101 |
|
102 |
|
103 |
|
@@ -146,13 +130,9 @@ enc_layers: 6
|
|
146 |
dec_layers: 6
|
147 |
heads: 8
|
148 |
rnn_size: 512
|
149 |
-
word_vec_size: 512
|
150 |
transformer_ff: 2048
|
151 |
dropout_steps: [0]
|
152 |
dropout: [0.1]
|
153 |
attention_dropout: [0.1]
|
154 |
share_decoder_embeddings: true
|
155 |
-
share_embeddings: false
|
156 |
-
|
157 |
-
|
158 |
-
|
|
|
8 |
# Corpus opts:
|
9 |
data:
|
10 |
europarl:
|
11 |
+
path_src: corpora/europarl/partitions/en_train.txt
|
12 |
+
path_tgt: corpora/europarl/partitions/gl_train.txt
|
13 |
transforms: [bpe, filtertoolong]
|
14 |
weight: 120
|
15 |
opensub:
|
16 |
+
path_src: corpora/opensub/partitions/en_train.txt
|
17 |
+
path_tgt: corpora/opensub/partitions/gl_train.txt
|
18 |
transforms: [bpe, filtertoolong]
|
19 |
weight: 152
|
20 |
opus:
|
21 |
+
path_src: corpora/opus/partitions/en_train.txt
|
22 |
+
path_tgt: corpora/opus/partitions/gl_train.txt
|
23 |
transforms: [bpe, filtertoolong]
|
24 |
weight: 160
|
25 |
ted2020:
|
26 |
+
path_src: corpora/ted2020/partitions/en_train.txt
|
27 |
+
path_tgt: corpora/ted2020/partitions/gl_train.txt
|
28 |
transforms: [bpe, filtertoolong]
|
29 |
weight: 10
|
30 |
corgaback:
|
31 |
+
path_src: corpora/corgaback/partitions/en_train.txt
|
32 |
+
path_tgt: corpora/corgaback/partitions/gl_train.txt
|
33 |
transforms: [bpe, filtertoolong]
|
34 |
weight: 15
|
35 |
ccmatrix:
|
36 |
+
path_src: corpora/ccmatrix/en_tok_dbo.txt
|
37 |
+
path_tgt: corpora/ccmatrix/gl_tok_dbo.txt
|
38 |
transforms: [bpe, filtertoolong]
|
39 |
+
weight: 380
|
40 |
wikimatrix:
|
41 |
+
path_src: corpora/wikimatrix/en.txt
|
42 |
+
path_tgt: corpora/wikimatrix/gl.txt
|
43 |
transforms: [bpe, filtertoolong]
|
44 |
+
weight: 70
|
45 |
cluvi:
|
46 |
+
path_src: corpora/cluvi/en.txt
|
47 |
+
path_tgt: corpora/cluvi/gl.txt
|
48 |
transforms: [bpe, filtertoolong]
|
49 |
+
weight: 70
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
valid:
|
51 |
+
path_src: corpora/partitions/all-en_valid.txt
|
52 |
+
path_tgt: corpora/partitions/all-gl_valid.txt
|
53 |
transforms: [bpe, filtertoolong]
|
54 |
|
55 |
### Transform related opts:
|
|
|
58 |
tgt_subword_model: ./bpe/gl.code
|
59 |
src_subword_vocab: ./run/bpe.vocab.src
|
60 |
tgt_subword_vocab: ./run/bpe.vocab.tgt
|
61 |
+
|
|
|
62 |
src_subword_type: bpe
|
63 |
tgt_subord_type: bpe
|
64 |
|
|
|
81 |
embeddings_type: "word2vec"
|
82 |
|
83 |
# word_vec_size need to match with the pretrained embeddings dimensions
|
84 |
+
word_vec_size: 512
|
85 |
|
86 |
|
87 |
|
|
|
130 |
dec_layers: 6
|
131 |
heads: 8
|
132 |
rnn_size: 512
|
|
|
133 |
transformer_ff: 2048
|
134 |
dropout_steps: [0]
|
135 |
dropout: [0.1]
|
136 |
attention_dropout: [0.1]
|
137 |
share_decoder_embeddings: true
|
138 |
+
share_embeddings: false
|
|
|
|
|
|