blob: 93d47380d8c30221fa880b1d0c956fe1a5ee1fc1 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
|
#!/bin/bash -x
#####################################################################
# SUMMARY: Create SentencePiece vocabulary for a language model
# AUTHOR: snukky
# TAGS: sentencepiece
#####################################################################
# Exit on error
set -e
# Remove old artifacts and create working directory
rm -rf vocab.lm vocab.lm.*{log,out,diff}
mkdir -p vocab.lm
# Run marian command
$MRT_MARIAN/marian --type lm \
--no-shuffle --seed 1111 --dim-emb 32 --dim-rnn 64 --maxi-batch 1 --maxi-batch-sort none \
-m vocab.lm/model.npz -t $MRT_DATA/europarl.de-en/corpus.small.en.gz \
--dim-vocabs 4000 -v vocab.lm/vocab.en.spm \
--after-batches 1 \
--log vocab.lm.log
# Check if files exist
test -e vocab.lm/vocab.en.spm
test -e vocab.lm.log
# Check logging messages
grep -q "Training SentencePiece vocabulary .*vocab.en.spm" vocab.lm.log
grep -q "Setting vocabulary size .* to 4,\?000" vocab.lm.log
grep -q "Loading SentencePiece vocabulary .*vocab.en.spm" vocab.lm.log
# Extract a textual vocabulary and compare with the expected output
LC_ALL=C $MRT_MARIAN/spm_export_vocab --model vocab.lm/vocab.en.spm | sort > vocab.lm.out
$MRT_TOOLS/diff-nums.py vocab.lm.out vocabs.en.expected -o vocab.lm.diff
# Exit with success code
exit 0
|
