Fahim Dalvi comments

Results 23 comments of


                                            Fahim Dalvi

CTranslate does not work with Deep bidirectional encoders

Ah, sorry I missed that! As for results, yes, we do consistently get better results with `dbrnn` compared to `brnn` (we train models involving Arabic, English & German)

Examples for tasks like Classification and Regression?

Hi @Hellisotherpeople, Thanks for the suggestion! We are planning to expand our tutorials/notebooks in the near future, and I think your suggestions are completely aligned with what we have planned!...

Analysing a fairseq transformer model

Hello @aprzez, Currently NeuroX only supports `transformers` (from HuggingFace) models for extraction, because extracting the representations need access and knowledge of the internals of the implementation. There are two possible...

Analysing a fairseq transformer model

Apologies for the delayed response, but I was away and this slipped my mind after coming back. `return_all_hiddens` should change what your _inference_ time outputs look like, not what happens...

Tests/regression tests

@baselmousi Whats v1 and v2 here? Are these testing for different things? We can name them better in that case

Creates an issue in Jupyter notebooks when different models are run in the same session

Should be fixed now, will try to reproduce it locally for good measure just to be sure.

neurox.data.extraction.transformers_extractor.extract_representation too verbose

Agreed, I think it would be a good feature to have.

assertion fails when extracting representations

@ahmed451 can you add the original sentence as well in its raw form?

assertion fails when extracting representations

I think this was fixed with the update to latest neurox? Can you confirm?

assertion fails when extracting representations

Hi @ahmed451, can you confirm that this is no longer an issue? I believe it was fixed with an update right?