blob: fcfdcd580c7b30324da49c43cc37058fe444ce45 (
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
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
|
#!/bin/bash -x
#####################################################################
# SUMMARY: Train with Adam optimized on CPU
# TAGS: cpu adam syncsgd
#####################################################################
# Exit on error
set -e
# Skip if no MKL found
if [ ! $MRT_MARIAN_USE_MKL ]; then
exit 100
fi
# Test code goes here
rm -rf adam_sync_cpu adam_sync_cpu_*.log
mkdir -p adam_sync_cpu
# TODO: The weight decaying in the Adam optimizer is enabled for CPU, because
# it gives stable results, in contrary to the GPU version
opts="--no-shuffle --seed 777 --mini-batch 2 --maxi-batch 1 --maxi-batch-sort none --dim-rnn 64 --dim-emb 32 --learn-rate 0.1 --optimizer adam --optimizer-params 0.9 0.98 0.001 --sync-sgd --devices 0 1 --cpu-threads 1"
# Added because default options has changes
opts="$opts --cost-type ce-mean --disp-label-counts false"
# Step 1: Training in one go
$MRT_MARIAN/marian \
-m adam_sync_cpu/model.full.npz -t $MRT_DATA/europarl.de-en/corpus.bpe.{en,de} -v vocab.en.yml vocab.de.yml \
--disp-freq 5 --after-batches 100 $opts \
--log adam_sync_cpu_f.log
test -e adam_sync_cpu/model.full.npz
test -e adam_sync_cpu_f.log
cat adam_sync_cpu_f.log | $MRT_TOOLS/strip-timestamps.sh | grep "Ep\. " | sed 's/ : Time.*//' > adam_sync_cpu.expected
# Step 2: Training from scratch up to the middle of the training in step 1
$MRT_MARIAN/marian \
-m adam_sync_cpu/model.npz -t $MRT_DATA/europarl.de-en/corpus.bpe.{en,de} -v vocab.en.yml vocab.de.yml \
--disp-freq 5 --after-batches 50 $opts \
--log adam_sync_cpu_1.log
test -e adam_sync_cpu/model.npz
test -e adam_sync_cpu_1.log
cat adam_sync_cpu_1.log | $MRT_TOOLS/strip-timestamps.sh | grep "Ep\. " | sed 's/ : Time.*//' > adam_sync_cpu.out
# Step 3: Restore/continue the training from step 2
$MRT_MARIAN/marian \
-m adam_sync_cpu/model.npz -t $MRT_DATA/europarl.de-en/corpus.bpe.{en,de} -v vocab.en.yml vocab.de.yml \
--disp-freq 5 --after-batches 100 $opts \
--log adam_sync_cpu_2.log
test -e adam_sync_cpu/model.npz
test -e adam_sync_cpu_2.log
cat adam_sync_cpu_2.log | $MRT_TOOLS/strip-timestamps.sh | grep "Ep\. " | sed 's/ : Time.*//' >> adam_sync_cpu.out
# Step 4: Compare log outputs between the full training and partial trainings
$MRT_TOOLS/diff-nums.py -p 0.3 adam_sync_cpu.out adam_sync_cpu.expected -o adam_sync_cpu.diff
# Exit with success code
exit 0
|
