Aaron Briel comments

Results 19 comments of


                                            Aaron Briel

Does the multi-label classifier optimize the entire model or it freezes BERT?

I tried adding a freeze_transformers_layer conditional in BertLearner's init function that set requires_grad to false for any param with the model_type in named_parameters but it didn't seem to have any...

Does the multi-label classifier optimize the entire model or it freezes BERT?

https://github.com/kaushaltrivedi/fast-bert/pull/195

Does the multi-label classifier optimize the entire model or it freezes BERT?

That is correct. Also note that I did indeed confirm this by looping through and printing all layer names along with their requires_grad setting (summarized for brevity): bert.embeddings.word_embeddings.weight, requires_grad:False bert.embeddings.position_embeddings.weight,...

Feature/add mlsmote

@SimonErm is this PR still in progress?

Feature/add mlsmote

@SimonErm Thanks for the reply.

{ [Error: No supported authentication method(s) available. Unable to login.] source: 'authentication' }

I'm seeing this issue as well, using the same parameters that are successful with node-imap and also imap-simple. Is it required to use an app token for IMAP_PASS? Thanks

Optimized prompt for multi-class classification contains only a subset of classifiers

@arnavsinghvi11 thank for the quick response! I will try this and let you know the results. Thanks!

Optimized prompt for multi-class classification contains only a subset of classifiers

@arnavsinghvi11 I keep running into the error below. I thought I had resolved it by adding `format=str` to each of the signature InputFields. It progressed a bit further but failed...

Optimized prompt for multi-class classification contains only a subset of classifiers

Using the recommended solution in (1) above, the resulting prompt was still missing 20 intents so that is not a feasible solution for a production release. The "Student must be...