Leonard Lausen

``` % python3 question_answering/parse_squad_results.py --dir squad_v2_horovod_fp16 1m 37s ~/src/gluon-nlp/tools/batch master ip-10-20-11-150 name best_f1 best_em best_f1_thresh best_em_thresh time_spent_in_hours 0 albert_base 81.861255 79.112272 -1.671970 -1.742718 1.139900 1 albert_large 84.904438 81.900109 -1.086745 -1.086745...

Looks like an AMP issue or an operator issue causing AMP to continue decreasing the scale.. [finetune_squad2.0.log](https://github.com/dmlc/gluon-nlp/files/5787622/finetune_squad2.0.log)

@szha FYI, after the split in serial / non-serial test execution, only the code coverage of serial tests was uploaded. Fixed by 527b23d

There can be issues when mixing system and conda toolchain. When using conda, we need to be careful to avoid that or we may experience issues such as https://github.com/dmlc/gluon-nlp/issues/1371 again....

@astonzhang FYI The results at https://github.com/dmlc/gluon-nlp/blob/master/scripts/sentiment_analysis/index.rst#textcnn are generated without this change. Could you confirm (on a sample) that the results remain unchanged?

Ping @xiaotinghe

They may have been fixed accidentally, though the tests remain disabled: https://github.com/apache/incubator-mxnet/issues/17782 https://github.com/apache/incubator-mxnet/pull/17969

> Actually I am a little confused about the semantic meaning of defer_init in the numpy mode, because shape with zeros (such as (0, 5)) in the numpy mode will...

17.02 has been released by now. Is there any plan to go forward with merging the job submit wrapper functions in master? Thanks for your efforts in creating this software!

Thanks for the heads up @giovtorres. My university is using version 14.11.11, I'm not sure if your code relies on any APIs released in more recent versions. If not I'll...