tarrade/proj_multilingual_text_classification

Issues

• Code crashing with TPU on AI Platform while running on Colab TPU

  #60 opened 5 years ago by tarrade
  4

• Add CI step to fix randpm see and compare it to previous runs

  #81 opened 4 years ago by tarrade
  1

• Add CI step to overfit one batch and chekc that the loss is very close to zero

  #80 opened 4 years ago by tarrade
  1

• Model interpretability: test LIP

  #79 opened 4 years ago by tarrade
  1

• run and fix all issues with pylint

  #76 opened 4 years ago by tarrade
  1

• Migration to Tensorflow 2.3.0

  #78 opened 4 years ago by tarrade
  1

• add docstring for all functions and classes

  #77 opened 4 years ago by tarrade
  1

• batch size training versus validation

  #75 opened 4 years ago by tarrade
  2

• Hyperparameter tuning on AI Platform training working with explicit call not default setup

  #61 opened 4 years ago by tarrade
  1

• tf.summary.create_file_writer -> file_cache is unavailable when using oauth2client >= 4.0.0 or google-auth

  #64 opened 4 years ago by tarrade
  2

• new issue

  #71 opened 5 years ago by tarrade
  1

• new issue

  #70 opened 5 years ago by tarrade
  1

• New issue

  #69 opened 5 years ago by tarrade
  1

• New issue

  #68 opened 5 years ago by tarrade
  0

• New issue

  #67 opened 5 years ago by tarrade
  0

• New issue

  #66 opened 5 years ago by tarrade
  0

• Test

  #65 opened 5 years ago by tarrade
  0

• Hyperparameter tuning on GCP: wrong status of the failed job

  #62 opened 5 years ago by tarrade
  1

• Tensorboard Profile: very strange spike in the first bin

  #48 opened 5 years ago by tarrade
  1

• AI Platform training: logging messages missing or duplicated or with the wrong label

  #63 opened 5 years ago by tarrade
  0

• Ideal Architecture for Fine-Tuning

  #53 opened 5 years ago by vluechinger
  0

• Implement a decaying learning rate

  #55 opened 5 years ago by vluechinger
  1

• How to create your own distilled model?

  #50 opened 5 years ago by vluechinger
  1

• Implement LASER and see if it is useful

  #30 opened 5 years ago by vluechinger
  1

• Preprocessing: How to Deal with Text Truncation

  #56 opened 5 years ago by vluechinger
  2

• Reproducability of BERT Model Training Runs

  #59 opened 5 years ago by vluechinger
  1

• Hyperparameter tuning: Is it worth it to change the batch size?

  #54 opened 5 years ago by vluechinger
  2

• Further Pre-training: do-able? Useful?

  #58 opened 5 years ago by vluechinger
  1

• python segfault when using TPU

  #49 opened 5 years ago by tarrade
  2

• switch from gcloud to python client library for model serving

  #37 opened 5 years ago by tarrade
  1

• Transfer Learning in NLP vs. CV: Differences?

  #57 opened 5 years ago by vluechinger
  2

• implement example with python client library for model training instead of gcloud

  #42 opened 5 years ago by tarrade
  1

• Need to store metadata for each job for bookkeeping

  #52 opened 5 years ago by tarrade
  1

• hyperparameter tuning with keras on GCP: failing for no reason

  #40 opened 5 years ago by tarrade
  2

• compute the multilingual BERT model (512 sequence length) with a GPU and compare computation time and costs to the CPU run

  #29 opened 5 years ago by vluechinger
  1

• What is so special about the tfrecord format?

  #51 opened 5 years ago by vluechinger
  0

• region parameter not use in the yaml file when using ai platform training

  #41 opened 5 years ago by tarrade
  1

• TensorBoard don't display properly hyperparameter tuning done on AI Platform training!

  #47 opened 5 years ago by tarrade
  1

• TPU for TensorFlow 2.1 not yet available for AI Platform training

  #43 opened 5 years ago by tarrade
  3

• Your input ran out of data; interrupting -> why repeat is not working here !

  #45 opened 5 years ago by tarrade
  1

• update transformer model in gcs with new version 2.9.0

  #46 opened 5 years ago by tarrade
  1

• SDK issue on AI Platform notebook?

  #44 opened 5 years ago by tarrade
  2

• see how to use TPU and keras with minimal modification

  #39 opened 5 years ago by tarrade
  1

• move the model part in a function for ai platform training

  #38 opened 5 years ago by tarrade
  1

• clean up def batch_translate_text

  #36 opened 5 years ago by tarrade
  1

• Store tf.data after the transformation needed for BERT in TFRecord files (preprocessing done only one)

  #35 opened 5 years ago by tarrade
  1

• Batch Translation on GCP creates problems

  #34 opened 5 years ago by vluechinger
  1

• be able to use any kind of model like xml-roberta

  #33 opened 5 years ago by tarrade
  0

• max length with BERT: 512 or it was only in the original paper from Google ?

  #32 opened 5 years ago by tarrade
  1

• Benchmark classical ML pipeline for the chosen datasets

  #31 opened 5 years ago by vluechinger
  1