Problem with saving model-best

The command spacy ray fails to save the model-best.
Using just spacy it save the model-best.
I get the same issue on Universal Dependencies Treebank example project.

Do you have any suggestion to solve this issue?

Fragment of my project.yml:

```
- name: "train"
help: "Train the request classification model"
script:
- "python -m spacy ray train configs/${vars.config} --output training/ --paths.train corpus/train.spacy --paths.dev corpus/val.spacy --nlp.lang ${vars.lang} --gpu-id ${vars.gpu_id}"
deps:
- "configs/${vars.config}"
- "corpus/train.spacy"
- "corpus/val.spacy"
outputs:
- "training/model-best"
```