s470607/transfix-train
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

update readme

Browse Source
This commit is contained in:
jakubknczny 2022-02-01 12:43:26 +01:00
parent a9ccbe0fda
commit ae4ed909d8
3 changed files with 21 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
README.md
View File

@ -1,12 +1,16 @@
# NEEDS TO BE UPDATED # Transfix-train for PLEWI
## PLEWI - clone PLEWI repo into ~/ via following command:
# PLEWI ```
- clone PLEWI repo into ~/ via cd ~ ; git clone https://git.wmi.amu.edu.pl/s434704/PLEWI-polish-errors-correction-challenge.git cd ~ ; git clone https://git.wmi.amu.edu.pl/s434704/PLEWI-polish-errors-correction-challenge.git
```
- run following commands: - run following commands:
- git clone https://git.wmi.amu.edu.pl/s470607/transfix-train.git - ```git clone https://git.wmi.amu.edu.pl/s470607/transfix-train.git```
- cd transfix-train - ```cd transfix-train```
- ./plewi.sh - ```./plewi.sh```
- sudo ./setup.sh (this will take some time) - ```sudo ./setup.sh``` (this will take some time)
- ./train ~/PLEWI-polish-errors-correction-challenge 32000 4 - ```./train ~/PLEWI-polish-errors-correction-challenge model-plewi 4```
where: 32000 is a number of BPE merge operations, 4 is a number of epochs where:
* ~/PLEWI-polish-errors-correction-challenge is a path to gonito repository
* model-plewi is a name for the model and the directory where training file will be stored in
* 4 is a number of training epochs

4
scripts/marian-train.sh
View File

@ -36,13 +36,13 @@ mkdir "$model_name"
--transformer-dropout 0.1 --label-smoothing 0.1 \ --transformer-dropout 0.1 --label-smoothing 0.1 \
--learn-rate 0.0003 --lr-warmup 16000 --lr-decay-inv-sqrt 16000 --lr-report \ --learn-rate 0.0003 --lr-warmup 16000 --lr-decay-inv-sqrt 16000 --lr-report \
--optimizer-params 0.9 0.98 1e-09 --clip-norm 5 \ --optimizer-params 0.9 0.98 1e-09 --clip-norm 5 \
--tied-embeddings \ --tied-embeddings-all \
--exponential-smoothing \ --exponential-smoothing \
--log "$model_name"/train.log \ --log "$model_name"/train.log \
--after-epochs="$epochs" \ --after-epochs="$epochs" \
--vocabs "$model_name"/vocab.in.spm "$model_name"/vocab.expected.spm \ --vocabs "$model_name"/vocab.in.spm "$model_name"/vocab.expected.spm \
--valid-log "$model_name"/valid.log \ --valid-log "$model_name"/valid.log \
--valid-metrics cross-entropy perplexity bleu \ --valid-metrics perplexity bleu \
--valid-mini-batch 64 \ --valid-mini-batch 64 \
--valid-sets "$source_file_valid" "$target_file_valid" \ --valid-sets "$source_file_valid" "$target_file_valid" \
--valid-translation-output "$model_name"/valid.output --quiet-translation --valid-translation-output "$model_name"/valid.output --quiet-translation

8
train.sh
View File

@ -2,10 +2,12 @@
# arguments # arguments
# 1. root of gonito.net challenge-like filestructure # 1. root of gonito.net challenge-like filestructure
# 2. expected number of train epochs # 2. name of the model and therefore a directory that will contain the model
# 3. expected number of train epochs
corpus_path="$1" corpus_path="$1"
epochs="$2" model_name="$2"
epochs="$3"
./scripts/marian-train.sh "$corpus_path" "$epochs" ./scripts/marian-train.sh "$corpus_path" "$model_model_name" "$epochs"