Ecosyste.ms: Issues

An open API service for providing issue and pull request metadata for open source projects.

GitHub / google-research/electra issues and pull requests

#100 - Can you share models trained with all weights tied?

Issue - State: open - Opened by YovaKem almost 4 years ago

#100 - Can you share models trained with all weights tied?

Issue - State: open - Opened by YovaKem almost 4 years ago

#99 - about tagging task

Issue - State: open - Opened by LastRyan almost 4 years ago

#98 - Question about expected results

Issue - State: closed - Opened by richarddwang almost 4 years ago - 1 comment

#97 - Using own data to continue pre-training from the released ELECTRA checkpoints

Issue - State: open - Opened by ghost about 4 years ago - 4 comments

#96 - Why RoBERTa-500K has 4.5x more computation than ELECTRA-400K?

Issue - State: closed - Opened by rabbitwayne about 4 years ago - 3 comments

#95 - How to Change Embedding Size of the Model?

Issue - State: open - Opened by FeryET about 4 years ago

#93 - Do you apply wnli trick ? If so, can you open the code ?

Issue - State: open - Opened by RyanHuangNLP about 4 years ago

#92 - Fix F1Scorer in finetune/classification/classification_metrics

Pull Request - State: closed - Opened by jgkimi about 4 years ago - 1 comment

#91 - Fix F1Scorer of classification_metrics

Pull Request - State: closed - Opened by jgkimi about 4 years ago - 4 comments

#89 - Question: Same Batchsize on different TPU sizes

Issue - State: closed - Opened by PhilipMay about 4 years ago - 2 comments

#88 - Add toggle to turn off `strip_accents`.

Pull Request - State: closed - Opened by PhilipMay about 4 years ago - 13 comments

#87 - Sampling step?

Issue - State: open - Opened by anshulsamar about 4 years ago - 2 comments

#86 - How can I make ELECTRA pretraining/dataloading use only one gpu ?

Issue - State: closed - Opened by richarddwang about 4 years ago - 3 comments

#85 - Finetuning loss doesn't converge when using loading weights

Issue - State: closed - Opened by smeaktrobush about 4 years ago

#84 - Improve Description of `--blanks-separate-docs`.

Issue - State: open - Opened by PhilipMay about 4 years ago

#83 - max_predictions_per_seq and TPU training configuration

Issue - State: open - Opened by Mistobaan about 4 years ago

#82 - When will the Chinese model be released？

Issue - State: open - Opened by zsweet about 4 years ago - 1 comment

#81 - Metrics definition

Issue - State: open - Opened by IssaIssa1 about 4 years ago

#80 - build_pretraining_dataset.py

Issue - State: closed - Opened by shinhyeokoh about 4 years ago - 1 comment

#79 - Checkpoint

Issue - State: closed - Opened by Zhengxian-Fan about 4 years ago

#78 - "Device or resource busy" for mounted paths

Issue - State: open - Opened by emirkin about 4 years ago

#77 - Request: pypi package for ELECTRA

Issue - State: open - Opened by mgroovyank about 4 years ago - 1 comment

#76 - keyerror:loss.

Issue - State: closed - Opened by fenfaqingnian over 4 years ago - 2 comments

#75 - Failed to convert object of type <class 'dict'> to Tensor

Issue - State: closed - Opened by lizaigaoge550 over 4 years ago - 1 comment

#74 - add code for continuing pre-training from an ELECTRA checkpoint

Pull Request - State: open - Opened by tuvuumass over 4 years ago - 3 comments

#73 - The difference of reproduced results on electra_small_owt

Issue - State: open - Opened by zheyuye over 4 years ago - 5 comments

#72 - Data loss: truncated record at 10035180

Issue - State: open - Opened by jjkim-zz over 4 years ago - 1 comment

#71 - mask prob in large model

Issue - State: closed - Opened by santaonchair over 4 years ago - 1 comment

#70 - NaN loss during training (again)

Issue - State: open - Opened by gchlodzinski over 4 years ago - 3 comments

#68 - Is `google/electra-small-generator` small or small++ ?

Issue - State: closed - Opened by richarddwang over 4 years ago - 1 comment

#67 - Error when pretraining on TPU: `Malformed device specification`

Issue - State: closed - Opened by danyaljj over 4 years ago - 2 comments

#66 - Training Electra on 2 phases like Bert

Issue - State: closed - Opened by agemagician over 4 years ago - 3 comments

#65 - Calculating ELECTRA infer FLOPs

Issue - State: closed - Opened by asharma20 over 4 years ago - 1 comment

#64 - Question about layerwise learning rate decay

Issue - State: closed - Opened by TianyuZhuuu over 4 years ago - 2 comments

#63 - problem encountered in reproducing Electra-Large

Issue - State: closed - Opened by spectrometerH over 4 years ago - 1 comment

#62 - How to configure tensorflow_gpu 1.15?

Issue - State: closed - Opened by MarkClemens301 over 4 years ago

#61 - Fix deprecated keyword argument in dropout layer.

Pull Request - State: open - Opened by jarednielsen over 4 years ago

#60 - ignore PAD during dynamic masking

Pull Request - State: open - Opened by ccchang0111 over 4 years ago - 2 comments

#59 - Should dynamic masking also ignore ['PAD']

Issue - State: closed - Opened by ccchang0111 over 4 years ago - 2 comments

#58 - [How to create vocab.txt file]

Issue - State: open - Opened by Vietdung113 over 4 years ago - 4 comments

#57 - Token-masking method: whole words or sub-words?

Issue - State: closed - Opened by cbaziotis over 4 years ago - 2 comments

#56 - RFC: List of community provided models

Issue - State: open - Opened by stefan-it over 4 years ago

#54 - Issue with loading weights for eval

Issue - State: closed - Opened by asharma20 over 4 years ago - 2 comments

#53 - [WIP] Define finetuning tasks in command-line hparams

Pull Request - State: open - Opened by mapmeld over 4 years ago

#52 - BasicTokenizer: _run_strip_accents

Issue - State: closed - Opened by Vodolazskyi over 4 years ago - 1 comment

#51 - The implementation of layerwise learning rate decay

Issue - State: closed - Opened by importpandas over 4 years ago - 2 comments

#50 - KeyError: '[SEP]'

Issue - State: closed - Opened by elyesmanai over 4 years ago - 5 comments

#49 - problem on electra's pretraining method

Issue - State: closed - Opened by real-brilliant over 4 years ago - 1 comment

#48 - Low usage of gpu

Issue - State: closed - Opened by amy-hyunji over 4 years ago - 2 comments

#47 - Add keep_checkpoint_max parameter

Pull Request - State: closed - Opened by stefan-it over 4 years ago

#46 - Build Dataset Issue

Issue - State: closed - Opened by qute012 over 4 years ago - 1 comment

#45 - 'adam_m not found in checkpoint ' when further pretraining

Issue - State: closed - Opened by DayuanJiang over 4 years ago - 6 comments

#44 - `num_train_steps` for further pretraining

Issue - State: closed - Opened by DayuanJiang over 4 years ago - 1 comment

#43 - Format of corpus

Issue - State: closed - Opened by mahnerak over 4 years ago - 4 comments

#42 - Bert vs Electra performances

Issue - State: closed - Opened by pretidav over 4 years ago - 1 comment

#41 - Deal with the duplicated positions in generator

Issue - State: closed - Opened by zheyuye over 4 years ago - 2 comments

#40 - Fix path to fine-tuning ELECTRA on MRQA tasks

Pull Request - State: closed - Opened by mrm8488 over 4 years ago - 3 comments

#39 - Model size conflit

Issue - State: closed - Opened by zheyuye over 4 years ago - 3 comments

#38 - Init disc/generator from pre-trained BERT

Issue - State: closed - Opened by volker42maru over 4 years ago - 1 comment

#37 - freeze discriminator and train generator only

Issue - State: closed - Opened by pretidav over 4 years ago - 2 comments

#36 - NaN loss during training

Issue - State: closed - Opened by tomohideshibata over 4 years ago - 6 comments

#35 - Availability on Tensorflow Hub (TFHub)

Issue - State: open - Opened by xhluca over 4 years ago - 2 comments

#34 - Pre-trained SMALL model cannot be loaded

Issue - State: closed - Opened by dreamingjudith over 4 years ago - 1 comment

#33 - SQuAD2 Score ELECTRA-Base

Issue - State: closed - Opened by volker42maru over 4 years ago - 2 comments

#32 - Auto loading in huggingface Transformers is broken

Issue - State: closed - Opened by xhluca over 4 years ago - 7 comments

#31 - eval pretrained model

Issue - State: closed - Opened by pretidav over 4 years ago - 1 comment

#30 - multi-task training

Issue - State: closed - Opened by xiaoxuesheng1234 over 4 years ago - 1 comment

#29 - num_eval_steps

Issue - State: closed - Opened by pretidav over 4 years ago - 2 comments

#28 - Bert tokenization korean decoding problems with lower case.

Pull Request - State: open - Opened by qute012 over 4 years ago

#27 - Bert korean unicode decoding problem.

Pull Request - State: closed - Opened by qute012 over 4 years ago - 1 comment

#26 - Training loss

Issue - State: open - Opened by DevKretov over 4 years ago - 7 comments

#25 - Continue pretraining on custom dataset

Issue - State: closed - Opened by ViktorAlm over 4 years ago - 3 comments

#24 - Question about fine-tuning on squad dataset

Issue - State: closed - Opened by curtis0982 over 4 years ago - 3 comments

#22 - confusing about stop_gradient in the code

Issue - State: closed - Opened by kelvinleen over 4 years ago - 1 comment

#21 - about chinese model

Issue - State: closed - Opened by Fan9 over 4 years ago - 3 comments

#20 - issue about segment in build_pretrain_dataset.py

Issue - State: closed - Opened by kelvinleen over 4 years ago - 2 comments

#19 - Load model in Pytorch.

Issue - State: closed - Opened by loopdigga96 over 4 years ago - 6 comments

#18 - cannot find openwebtext.tar.xz

Issue - State: closed - Opened by Bournet over 4 years ago - 5 comments

#17 - Will the Pre-trained ELECTRA-1.75M be released?

Issue - State: closed - Opened by xf05888 over 4 years ago - 3 comments

#15 - Definition of Loss

Issue - State: closed - Opened by pidahbus over 4 years ago - 2 comments

#14 - How to get the embedding vector or matrix after pre-training

Issue - State: open - Opened by pidahbus over 4 years ago - 4 comments

#13 - Fine Tunned Large model on Squad 2.0 using 8GB GPU

Issue - State: closed - Opened by renatoviolin over 4 years ago - 1 comment

#12 - TPU training

Issue - State: closed - Opened by DevKretov over 4 years ago - 2 comments

#11 - how to predict the masked token ?

Issue - State: closed - Opened by dixonhsiao over 4 years ago - 2 comments

#10 - Is the result based on dev or test set?

Issue - State: closed - Opened by g-jing over 4 years ago - 1 comment

#9 - Multi-GPU training

Issue - State: open - Opened by hamidpalangi over 4 years ago - 7 comments

#8 - Issue while generating pre training data

Issue - State: open - Opened by 008karan over 4 years ago - 15 comments

#7 - pretraining: configure hparms for base and large models

Pull Request - State: closed - Opened by stefan-it over 4 years ago - 1 comment

#6 - using pre training generated for albert for Electra

Issue - State: closed - Opened by 008karan over 4 years ago - 2 comments

#5 - Fix BibTeX entry

Pull Request - State: closed - Opened by michelole over 4 years ago - 3 comments

#4 - TPU training: No matching devices found for

Issue - State: closed - Opened by stefan-it over 4 years ago - 3 comments

#3 - Loss of base and large models

Issue - State: closed - Opened by stefan-it over 4 years ago - 7 comments