From 336740065d9c23e53e912a1befff18981d9d27ab Mon Sep 17 00:00:00 2001 From: Marcin Junczys-Dowmunt Date: Tue, 27 Nov 2018 11:06:10 -0800 Subject: [PATCH] Update README.md --- training-basics-sentencepiece/README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/training-basics-sentencepiece/README.md b/training-basics-sentencepiece/README.md index ce14f28..3c1d59c 100644 --- a/training-basics-sentencepiece/README.md +++ b/training-basics-sentencepiece/README.md @@ -212,7 +212,7 @@ give you in-training BLEU scores that are very close to sacreBLEU's scores. Diff if unexpected SentencePiece normalization rules are used. You should still report only official sacreBLEU scores for publications. -We are training of four GPUs defined with `--devices 0 1 2 3`. Change this to the required number of GPUs. +We are training using four GPUs defined with `--devices 0 1 2 3`. Change this to the required number of GPUs. ``` # create the model directory @@ -241,7 +241,7 @@ mkdir model ``` The training should stop if cross-entropy on the validation set -stops improving. Depending on the number of and generation of GPUs you are using that +stops improving. Depending on the number and generation of GPUs you are using that may take a while. To inspect the created SentencePiece model `model/vocab.roen.spm`, you can now segment any