diff options
| author | Roman Grundkiewicz <rgrundki@exseed.ed.ac.uk> | 2017-11-09 21:23:43 +0300 |
|---|---|---|
| committer | Roman Grundkiewicz <rgrundki@exseed.ed.ac.uk> | 2017-11-09 21:23:43 +0300 |
| commit | 46eb8be05897e71e42c9d32c3b743d17bb398886 (patch) | |
| tree | 8999e7a18bb76bdfdf68cd7b4ffe38746182f3eb /examples | |
| parent | 07b1a01c9297f85f73c19a60a8988cb8cf31755e (diff) | |
| parent | d99e7451a0caf01b163a0a233a5e7c4250145222 (diff) | |
Merge
Diffstat (limited to 'examples')
| -rw-r--r-- | examples/transformer/README.md | 6 |
1 files changed, 3 insertions, 3 deletions
diff --git a/examples/transformer/README.md b/examples/transformer/README.md index 59746535..22ce1cb0 100644 --- a/examples/transformer/README.md +++ b/examples/transformer/README.md @@ -4,7 +4,7 @@ Files and scripts in this folder shows how to run transformer model ([Vaswani et al, 2017](https://arxiv.org/abs/1706.03762)) on WMT-17 English-German data. The problem-set is adapted from [tensor2tensor](https://github.com/tensorflow/tensor2tensor) repository from -Google, i.e. 32,000 common BPE subwords for both languages. +Google, i.e. 36,000 common BPE subwords for both languages. No back-translations are used. @@ -23,8 +23,8 @@ It executes a training run with `marian` using the following command: --seed 1234 \ --type transformer \ --model model/model.npz \ - --train-sets data/corpus.en data/corpus.de \ - --vocabs model/vocab.en.yml model/vocab.de.yml \ + --train-sets data/corpus.bpe.en data/corpus.bpe.de \ + --vocabs model/vocab.ende.yml model/vocab.ende.yml \ --max-length 100 \ --mini-batch-fit -w 7000 --maxi-batch 1000 \ --early-stopping 10 \ |