transfix-train/scripts/marian-train.sh
2022-01-14 18:45:36 +01:00

54 lines
1.6 KiB
Bash

#!/bin/bash
# arguments
# 1. root of gonito.net challenge-like filestructure
# 2. suffix of source sentences, E.g. en for files like train.en
# 3. suffix of source sentences, E.g. pl for files like train.pl
# 4. number of BPE merge operations, E.g. 32000
# 5. expected number of train epochs
# path to corpus
corpus_path="$1"
source_suffix="$2"
target_suffix="$3"
bpe_merges="$4"
epochs="$5"
source_file="$corpus_path"/train/train."$source_suffix"
source_vocab="$source_file".vocab."$bpe_merges"
target_file="$corpus_path"/train/train."$target_suffix"
target_vocab="$target_file".vocab."$bpe_merges"
./marian/build/marian \
--model model/model.npz --type transformer \
--overwrite
--train-sets "$source_file" "$target_file" \
--max-length 100 \
--vocabs "$source_vocab" "$target_vocab" \
#--vocabs model/vocab.ende.yml model/vocab.ende.yml \
--mini-batch-fit -w 10000 --maxi-batch 1000 \
--after_epochs "$epochs" \
#--early-stopping 10 \
--valid-freq 5000 \
--save-freq 5000 \
--disp-freq 500 \
--beam-size 6 --normalize 0.6 \
--enc-depth 6 --dec-depth 6 \
--transformer-heads 8 \
--transformer-postprocess-emb d \
--transformer-postprocess dan \
--transformer-dropout 0.1 --label-smoothing 0.1 \
--learn-rate 0.0003 --lr-warmup 16000 --lr-decay-inv-sqrt 16000 --lr-report \
--optimizer-params 0.9 0.98 1e-09 --clip-norm 5 \
--tied-embeddings-all \
--exponential-smoothing
--log model/train.log \
#--valid-log model/valid.log \
#--valid-metrics cross-entropy perplexity translation \
#--valid-sets data/valid.bpe.en data/valid.bpe.de \
#--valid-script-path ./scripts/validate.sh \
#--valid-translation-output data/valid.bpe.en.output --quiet-translation \
#--valid-mini-batch 64 \