Trjtdtk

Can a medieval gyroplane be built?

Is it true that good novels will automatically sell themselves on Amazon (and so on) and there is no need for one to waste time promoting?

Unfrosted light bulb

HP P840 HDD RAID 5 many strange drive failures

Suggestions on how to spend Shaabath (constructively) alone

Can you move over difficult terrain with only 5 feet of movement?

Am I eligible for the Eurail Youth pass? I am 27.5 years old

Would it be believable to defy demographics in a story?

Light propagating through a sound wave

Help prove this basic trig identity please!

In Aliens, how many people were on LV-426 before the Marines arrived​?

What favor did Moody owe Dumbledore?

World War I as a war of liberals against authoritarians?

A Ri-diddley-iley Riddle

Is honey really a supersaturated solution? Does heating to un-crystalize redissolve it or melt it?

How to define limit operations in general topological spaces? Are nets able to do this?

PTIJ: Do Irish Jews have "the luck of the Irish"?

Does the attack bonus from a Masterwork weapon stack with the attack bonus from Masterwork ammunition?

How to terminate ping <dest> &

What is the term when voters “dishonestly” choose something that they do not want to choose?

Why are there no stars visible in cislunar space?

Recruiter wants very extensive technical details about all of my previous work

What is the significance behind "40 days" that often appears in the Bible?

What does "^L" mean in C?

Incrementally Train BERT with minimum QnA records - to get improved results

Machine Learning models in production environmentClarification wanted for make_step function of Google's deep dream scriptBest approach for image recognition/classification with few training dataMLPClassifier threshold factor to eliminate test samples that are not in match with train dataTools and Techniques for Analyzing German Automotive Discussion Forum PostsHow to train and validate a model continously which affects its own future data?Training dataset decreasing in quality (Google data science blog)Understanding why my binary classification is approaching 50% accuracy using TensorFlow and KerasHyper-parameter tuning when you don't have an access to the test dataIssues with training SSD on own dataset

0

$begingroup$

We are using Google BERT for Question and Answering. We have fine tuned BERT with SQUAD QnA release train data set (https://github.com/google-research/bert , https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v1.1.json)

It generated new checkpoints and BERT is giving good answers for most of questions we asked on our text documents. However, there are some questions which it is answering wrong, so we are trying to further fine tune with our Question and known answer on our text document. We further trained based on last generated checkpoint and got new checkpoint.

With new checkpoint when we are asking the same question, the answer did not got corrected! Previously BERT was giving wrong answer with 99% confidence and now also giving same wrong answer with 95% confidence.

Can someone suggest, if they have same or similar experience, and suggest please.

Following are questions in BERT github Issues, and are unanswered for quite some time:

• BERT accuracy reduced after providing custom training..The answer is
  also not correct : https://github.com/google-research/bert/issues/492
  


• Unable to incrementally train BERT with custom training: https://github.com/google-research/bert/issues/482
  


• Little training has no impact: https://github.com/google-research/bert/issues/481


• Custom Domain Training: https://github.com/google-research/bert/issues/498
  

machine-learning training transformer bert

share|improve this question

edited yesterday

user10557045

asked yesterday

user10557045user10557045

165

$endgroup$

add a comment | 

0

$begingroup$

We are using Google BERT for Question and Answering. We have fine tuned BERT with SQUAD QnA release train data set (https://github.com/google-research/bert , https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v1.1.json)

It generated new checkpoints and BERT is giving good answers for most of questions we asked on our text documents. However, there are some questions which it is answering wrong, so we are trying to further fine tune with our Question and known answer on our text document. We further trained based on last generated checkpoint and got new checkpoint.

With new checkpoint when we are asking the same question, the answer did not got corrected! Previously BERT was giving wrong answer with 99% confidence and now also giving same wrong answer with 95% confidence.

Can someone suggest, if they have same or similar experience, and suggest please.

Following are questions in BERT github Issues, and are unanswered for quite some time:

• BERT accuracy reduced after providing custom training..The answer is
  also not correct : https://github.com/google-research/bert/issues/492
  


• Unable to incrementally train BERT with custom training: https://github.com/google-research/bert/issues/482
  


• Little training has no impact: https://github.com/google-research/bert/issues/481


• Custom Domain Training: https://github.com/google-research/bert/issues/498
  

machine-learning training transformer bert

share|improve this question

edited yesterday

user10557045

asked yesterday

user10557045user10557045

165

$endgroup$

add a comment | 

$begingroup$

We are using Google BERT for Question and Answering. We have fine tuned BERT with SQUAD QnA release train data set (https://github.com/google-research/bert , https://rajpurkar.github.io/SQuAD-explorer/dataset/train-v1.1.json)

It generated new checkpoints and BERT is giving good answers for most of questions we asked on our text documents. However, there are some questions which it is answering wrong, so we are trying to further fine tune with our Question and known answer on our text document. We further trained based on last generated checkpoint and got new checkpoint.

With new checkpoint when we are asking the same question, the answer did not got corrected! Previously BERT was giving wrong answer with 99% confidence and now also giving same wrong answer with 95% confidence.

Can someone suggest, if they have same or similar experience, and suggest please.

Following are questions in BERT github Issues, and are unanswered for quite some time:

• BERT accuracy reduced after providing custom training..The answer is
  also not correct : https://github.com/google-research/bert/issues/492
  


• Unable to incrementally train BERT with custom training: https://github.com/google-research/bert/issues/482
  


• Little training has no impact: https://github.com/google-research/bert/issues/481


• Custom Domain Training: https://github.com/google-research/bert/issues/498
  

machine-learning training transformer bert

share|improve this question

edited yesterday

user10557045

asked yesterday

user10557045user10557045

165

$endgroup$

share|improve this question

edited yesterday

user10557045

asked yesterday

user10557045user10557045

165

share|improve this question

edited yesterday

user10557045

asked yesterday

user10557045user10557045

165

asked yesterday

user10557045user10557045

165

asked yesterday

user10557045user10557045

165