blob: 0e4f52f2b4a9b26fcdcebe4dbf13049fad6b0897 (
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
#####################################################################
# SUMMARY: Decode with a lexical shortlist on GPU via marian-server
# AUTHOR: snukky
# TAGS: server
#####################################################################
# Exit on error
set -e
# Check if marian-server is compiled
test -f $MRT_MARIAN/marian-server || exit $EXIT_CODE_SKIP
clean_up() {
kill $SERVER_PID
}
trap clean_up EXIT
rm -f server.{out,diff}
# Start Marian server
$MRT_MARIAN/marian-server -c $MRT_MODELS/rnn-spm/decode.yml --shortlist $MRT_MODELS/rnn-spm/lex.s2t.gz 100 75 -p 8765 --mini-batch 64 > server.log 2>&1 &
SERVER_PID=$!
# Wait a bit until the model is loaded
sleep 20
# Translate
python3 $MRT_MARIAN/../scripts/server/client_example.py -p 8765 -b 64 < text.in > server.out
kill $SERVER_PID
# Compare the translation with expected output
$MRT_TOOLS/diff.sh server.out rnn_gpu.expected > server.diff
test -e server.log
grep -q "listening on port 8765" server.log
# Exit with success code
exit 0
|
