run predict function in Colab, get ConfigurationError: Invalid batch_type option.

[raisinbl]

I run predict function with right config in Google Colab but I get ConfigurationError: Invalid `batch_type` option. even when the batch_type in my config.yaml is set to sentence or token (the JoeyNMT version is 2.2.0)

what I want is to get the hypothesis score for each sentence when I pass a dataset to predict function when import from joeynmt.prediction. I pretty sure that my config file is pointed correctly, and also for the config setting. I checked many times, even when run the config with python -m joeynmt test directory/config.yaml .... is okay.

I call

_, _, _, trg_tokens, trg_scores, _ = predict(
            model=model,
            data=active,
            compute_loss=False,
            device='cpu',
            n_gpu=0,
            normalization="none",
            num_workers=4,
            cfg=config,
            fp16=False,
        )

Logged output

---------------------------------------------------------------------------
ConfigurationError                        Traceback (most recent call last)
[<ipython-input-42-4c784ab329a7>](https://localhost:8080/#) in <module>
      2 
      3 print('start here')
----> 4 _, _, _, trg_tokens, trg_scores, _ = predict(
      5             model=model,
      6             data=active,

1 frames
[/usr/local/lib/python3.9/dist-packages/joeynmt/prediction.py](https://localhost:8080/#) in predict(model, data, device, n_gpu, compute_loss, normalization, num_workers, cfg, fp16)
     91         return_prob,
     92         generate_unk,
---> 93         repetition_penalty,
     94         no_repeat_ngram_size,
     95     ) = parse_test_args(cfg)

[/usr/local/lib/python3.9/dist-packages/joeynmt/helpers.py](https://localhost:8080/#) in parse_test_args(cfg)
    329     batch_type: str = cfg.get("batch_type", "sentences")
    330     print(batch_type)
--> 331     if batch_type not in ["sentence", "token"]:
    332         raise ConfigurationError(
    333             "Invalid `batch_type` option. Valid options: {`sentence`, `token`}.")

ConfigurationError: Invalid `batch_type` option. Valid options: {`sentence`, `token`}.

my configuration is

data: ...
training: ...
model: ...
testing:
    n_best: 3
    beam_size: 5
    beam_alpha: 1.0
    batch_size: 20
    batch_type: "sentence"
    max_output_length: 100
    eval_metrics: ["bleu"]
    return_prob: "hyp"
    sacrebleu_cfg:
        tokenize: "13a"