Stephen Roller

@jaseweston do you know?

I've asked @jaseweston to talk to the first author and find the dataset for us.

We've been trying to reach the original author, still no reply.

Confirmed I was able to download. We'll need to add it as a teacher.

Contributions would be welcome. I believe there's an interactive chat script for the ImageChat project, and we could start from there.

``` self = def test_beamdelay(self): """ Test delayedbeam generation. """ # Delayed Beam is inherently stochastic, just ensure no crash. opt = ParlaiParser(True, True).parse_kwargs( model_file='zoo:unittest/transformer_generator2/model', inference='delayedbeam', topk=10, beam_delay=2, beam_min_length=2, )...

``` self = @testing_utils.retry(ntries=3, log_retry=True) def test_crossencoder(self): valid, test = testing_utils.train_model( dict( task='convai2', model='bert_ranker/cross_encoder_ranker', num_epochs=0.002, batchsize=1, candidates="inline", type_optimization="all_encoder_layers", warmup_updates=100, text_truncate=32, label_truncate=32, validation_max_exs=20, short_final_eval=True, ) ) # The cross encoder reaches...

I agree, it would be nice to see this in ParlAI. We don't expect we have resources to prioritize it right now, but we would welcome any contributions toward this...

Thanks for the really thorough bug report, so nice. It looks like that paper assumed that you were always providing the specific teacher. In the other datasets, we designate one...

I still need to submit this pr