TypeError: 'bool' object is not callable

[thpun]

🐛 Bug

To Reproduce

Steps to reproduce the behavior:

• Followed mBART instructions for preprocessing training data
• Start multilingual fine-tuning (translation_multi_simple_epoch task) with RoundRobin as sampling method

The training starts smoothly if using temperature sampling, but it failed if using RoundRobin.

1. Run cmd

   lang_pairs=<comma-separated list of training lang pairs>
   DATA=<path to training data>
   lang_list=<path to 25 lang id file>
   MODEL=<path to pretrained mode>
   fairseq-train $DATA \
   --finetune-from-model $MODEL \
   --encoder-normalize-before --decoder-normalize-before \
   --arch mbart_large --layernorm-embedding \
   --task translation_multi_simple_epoch \
   --sampling-method "RoundRobin" \
   --encoder-langtok "src" \
   --decoder-langtok \
   --lang-dict "$lang_list" \
   --lang-pairs "$lang_pairs" \
   --criterion label_smoothed_cross_entropy --label-smoothing 0.2 \
   --optimizer adam --adam-eps 1e-06 --adam-betas '(0.9, 0.98)' \
   --lr-scheduler inverse_sqrt --lr 6e-05 --min-lr -1 --warmup-updates 5000 \
   --dropout 0.3 --attention-dropout 0.1 --weight-decay 0.0 \
   --max-tokens 1024 --update-freq 4 --virtual-epoch-size 2000000 \
   --save-interval 1 --save-interval-updates 2500 --no-epoch-checkpoints \
   --seed 222 --log-format simple --log-interval 10 \
   --fp16 --max-update 100000 --save-dir $SAVEDIR

2. See error

   
   Traceback (most recent call last):
   File "/opt/conda/bin/fairseq-train", line 33, in <module>
   sys.exit(load_entry_point('fairseq', 'console_scripts', 'fairseq-train')())
   File "/workspace/fairseq/fairseq_cli/train.py", line 351, in cli_main
   distributed_utils.call_main(args, main)
   File "/workspace/fairseq/fairseq/distributed_utils.py", line 174, in call_main
   args.distributed_world_size,
   File "/opt/conda/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 200, in spawn
   return start_processes(fn, args, nprocs, join, daemon, start_method='spawn')
   File "/opt/conda/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 158, in start_processes
   while not context.join():
   File "/opt/conda/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 119, in join
   raise Exception(msg)
   Exception:

-- Process 3 terminated with the following error: Traceback (most recent call last): File "/opt/conda/lib/python3.6/site-packages/torch/multiprocessing/spawn.py", line 20, in _wrap fn(i, *args) File "/workspace/fairseq/fairseq/distributed_utils.py", line 156, in distributed_main main(args, kwargs) File "/workspace/fairseq/fairseq_cli/train.py", line 106, in main extra_state, epoch_itr = checkpoint_utils.load_checkpoint(args, trainer) File "/workspace/fairseq/fairseq/checkpoint_utils.py", line 188, in load_checkpoint epoch=1, load_dataset=True, passthrough_args File "/workspace/fairseq/fairseq/trainer.py", line 350, in get_train_iterator epoch=epoch File "/workspace/fairseq/fairseq/tasks/translation_multi_simple_epoch.py", line 303, in get_batch_iterator seed=seed, num_shards=num_shards, shard_id=shard_id, num_workers=num_workers, epoch=epoch, File "/workspace/fairseq/fairseq/tasks/fairseq_task.py", line 184, in get_batch_iterator required_batch_size_multiple=required_batch_size_multiple, TypeError: 'bool' object is not callable

### Expected behavior

Training starts smoothly, just like in temperature sampling

### Environment

 - fairseq Version (e.g., 1.0 or master): master, commit 4c55744ec4cb26749cf2cf8dac89942f26ce4bd2
 - PyTorch Version (e.g., 1.0) `1.5.0a0+8f84ded`
 - OS (e.g., Linux): Linux
 - How you installed fairseq (`pip`, source): source
 - Build command you used (if compiling from source): `pip install --editable .`
 - Python version: 3.6.9
 - CUDA/cuDNN version: 10.1
 - GPU models and configuration: V100

[thpun]

cc @tangyuq Does translation_multi_simple_epoch support sampling method other than temperature sampling at this moment?

[tangyuq]

translation_multi_simple_epoch only supports "temperature" sampling right now.

[thpun]

Is there any plan to support the rest of the options in --sampling-method, i.e. uniform, Concat & RoundRobin in near future?