How to generate topK predictions using Levenstein Transformer?

[cyhhhhhh]

❓ Questions and Help

Before asking:

1. search the issues.
2. search the docs.

What is your question?

I want to generate topK translated sentences. I found in the documents that I could set --nbest to achieve this goal. However, I've tried and it still generated only one sentence when translating.

Code

fairseq-generate \ data-bin/Dataset \ --gen-subset test \ --task translation_lev \ --path checkpoints/checkpoint_best.pt \ --iter-decode-max-iter 9 \ --iter-decode-eos-penalty 0 \ --beam 1 --remove-bpe \ --print-step \ --batch-size 400 \ --results-path data-test \ --nbest 10

What have you tried?

I set --nbest to 10 but it didn't work.

What's your environment?

• fairseq Version (e.g., 1.0 or main):
• PyTorch Version (e.g., 1.0)
• OS (e.g., Linux):
• How you installed fairseq (pip, source):
• Build command you used (if compiling from source):
• Python version:
• CUDA/cuDNN version:
• GPU models and configuration:
• Any other relevant information:

[cyhhhhhh]

I spent several days on this and only found out that this modal could not generate multiple predictions so as all the iterative-refine generators, which made me very upset. Why not state more clearly in the documentation or readme file? I think it's too general considering all the models and functions of fairseq, isn't it?