modelId
stringlengths 4
112
| sha
stringlengths 40
40
| lastModified
stringlengths 24
24
| tags
sequence | pipeline_tag
stringclasses 29
values | private
bool 1
class | author
stringlengths 2
38
⌀ | config
null | id
stringlengths 4
112
| downloads
float64 0
36.8M
⌀ | likes
float64 0
712
⌀ | library_name
stringclasses 17
values | __index_level_0__
int64 0
38.5k
| readme
stringlengths 0
186k
|
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Laezor/DialoGPT-small-witcher1 | e2b009a46853b2f2b5b0be713c88bf16ebe80d80 | 2021-10-09T11:52:29.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Laezor | null | Laezor/DialoGPT-small-witcher1 | 1 | null | transformers | 28,100 | ---
tags:
- conversational
---
#Witcher1 Geralt DialoGPT small model |
Laezor/DialoGPT-small-yakuza_0 | 0c50448c5507339e4927daad668e66b6c3f4427e | 2021-10-08T15:45:12.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Laezor | null | Laezor/DialoGPT-small-yakuza_0 | 1 | null | transformers | 28,101 | ---
tags:
- conversational
---
#Yakuza 0 DialoGPT Model |
LanPham/wav2vec2-base-jp | 5bfabf136eb4eb141713cb15ac6a3353ba43ad04 | 2022-01-14T09:21:11.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"transformers"
] | automatic-speech-recognition | false | LanPham | null | LanPham/wav2vec2-base-jp | 1 | null | transformers | 28,102 | Entry not found |
Langame/gpt2-waiting | 69d46890b95cb91d0f38d56ffde846e20b39936d | 2021-12-19T09:02:26.000Z | [
"pytorch",
"tensorboard",
"gpt2",
"text-generation",
"en",
"dataset:waiting-messages",
"transformers",
"license:mit"
] | text-generation | false | Langame | null | Langame/gpt2-waiting | 1 | 1 | transformers | 28,103 | ---
language:
- en # Example: en
license: mit # Example: apache-2.0 or any license from https://hf.co/docs/hub/model-repos#list-of-license-identifiers
tags:
- text-generation
datasets:
- waiting-messages # Example: common_voice. Use dataset id from https://hf.co/datasets
widget:
- text: 'List of funny waiting messages:'
example_title: 'Funny waiting messages'
---
# Langame/gpt2-waiting
This fine-tuned model can generate funny waiting messages.
[Langame](https://langa.me) uses these within its platform 😛.
|
LenaSchmidt/distilbert-base-uncased-finetuned-squad | 0d05f72d35f7c1dcb60f2e17e4648d1b68da6c20 | 2022-02-04T19:20:11.000Z | [
"pytorch",
"tensorboard",
"distilbert",
"question-answering",
"transformers",
"generated_from_trainer",
"license:apache-2.0",
"model-index",
"autotrain_compatible"
] | question-answering | false | LenaSchmidt | null | LenaSchmidt/distilbert-base-uncased-finetuned-squad | 1 | null | transformers | 28,104 | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: distilbert-base-uncased-finetuned-squad
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilbert-base-uncased-finetuned-squad
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 1.7713
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 2.0325 | 1.0 | 585 | 1.7520 |
| 1.609 | 2.0 | 1170 | 1.7713 |
### Framework versions
- Transformers 4.16.2
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.11.0
|
Leonel/DialoGPT-small-chandler | db67246e3bb8966316c172a17b1adac280c89774 | 2021-09-01T17:18:41.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Leonel | null | Leonel/DialoGPT-small-chandler | 1 | null | transformers | 28,105 | ---
tags:
- conversational
---
This is Chandler.
Chandler is your friend too. |
LucasS/albertABSA | d23092d8bfe6a194448b8db485401e57475d86c3 | 2021-08-30T08:34:20.000Z | [
"pytorch",
"albert",
"question-answering",
"transformers",
"autotrain_compatible"
] | question-answering | false | LucasS | null | LucasS/albertABSA | 1 | null | transformers | 28,106 | Entry not found |
LucasS/robertaABSA | 41be180120c3ce4464fa10888ee86894ae8f18de | 2021-08-30T08:33:29.000Z | [
"pytorch",
"roberta",
"question-answering",
"transformers",
"autotrain_compatible"
] | question-answering | false | LucasS | null | LucasS/robertaABSA | 1 | null | transformers | 28,107 | Entry not found |
Lucdi90/DialoGPT-medium-XiaoBot | c57896f0ec25046a16ce961b4f96fb9e0798ecd2 | 2021-11-04T08:54:27.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Lucdi90 | null | Lucdi90/DialoGPT-medium-XiaoBot | 1 | null | transformers | 28,108 | ---
tags:
- conversational
---
# XiaoBot for Discord
[Tutorial](https://youtu.be/UjDpW_SOrlw) followed for this model. |
LuckyWill/DialoGPT-small-JakeBot | 2e6a06cadbe516c0649405ce3ed947ad9ff11543 | 2021-11-04T06:13:49.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | LuckyWill | null | LuckyWill/DialoGPT-small-JakeBot | 1 | null | transformers | 28,109 | ---
tags:
- conversational
---
# Jake Peralta B99 DialoGPT Model |
Lurka/DialoGPT-medium-isseibot | 11741a07ea6d7f3b5865b9f77c7b3611a567c621 | 2021-10-07T19:44:37.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Lurka | null | Lurka/DialoGPT-medium-isseibot | 1 | null | transformers | 28,110 | ---
tags:
- conversational
---
# Issei Hyoudou DialoGPT Model |
ML-ass/english_decoder | 739df30f720bf1c0fb79724c3ce622e011396f72 | 2021-07-03T10:17:02.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers"
] | text-generation | false | ML-ass | null | ML-ass/english_decoder | 1 | null | transformers | 28,111 | Entry not found |
ML-ass/g2e_encoder_decoder | d6f1aa9f0108fec5e3c72e5949417a2f962b79c5 | 2021-07-03T16:39:49.000Z | [
"pytorch",
"encoder-decoder",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | ML-ass | null | ML-ass/g2e_encoder_decoder | 1 | null | transformers | 28,112 | Entry not found |
MM98/mt5-small-finetuned-pnsum2 | 6d97e6bf32a76acf91cb9dcbd3aeac9c937282ff | 2022-01-29T18:57:20.000Z | [
"pytorch",
"tensorboard",
"mt5",
"text2text-generation",
"transformers",
"generated_from_trainer",
"license:apache-2.0",
"model-index",
"autotrain_compatible"
] | text2text-generation | false | MM98 | null | MM98/mt5-small-finetuned-pnsum2 | 1 | null | transformers | 28,113 | ---
license: apache-2.0
tags:
- generated_from_trainer
metrics:
- rouge
model-index:
- name: mt5-small-finetuned-pnsum2
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# mt5-small-finetuned-pnsum2
This model is a fine-tuned version of [google/mt5-small](https://huggingface.co/google/mt5-small) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: nan
- Rouge1: 4.3733
- Rouge2: 1.0221
- Rougel: 4.1265
- Rougelsum: 4.1372
- Gen Len: 6.2843
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 1
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Rouge1 | Rouge2 | Rougel | Rougelsum | Gen Len |
|:-------------:|:-----:|:----:|:---------------:|:------:|:------:|:------:|:---------:|:-------:|
| 0.0 | 1.0 | 2500 | nan | 4.3733 | 1.0221 | 4.1265 | 4.1372 | 6.2843 |
### Framework versions
- Transformers 4.16.1
- Pytorch 1.10.0+cu111
- Datasets 1.18.2
- Tokenizers 0.11.0
|
MYX4567/gpt2-wikitext2 | 6e7bad418f2a2390e1d4089e6813034b7942af08 | 2021-07-28T03:42:36.000Z | [
"pytorch",
"tensorboard",
"gpt2",
"text-generation",
"transformers",
"generated_from_trainer"
] | text-generation | false | MYX4567 | null | MYX4567/gpt2-wikitext2 | 1 | null | transformers | 28,114 | ---
tags:
- generated_from_trainer
datasets:
- null
model_index:
- name: gpt2-wikitext2
results:
- task:
name: Causal Language Modeling
type: text-generation
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# gpt2-wikitext2
This model is a fine-tuned version of [](https://huggingface.co/) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 6.3227
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 6.7523 | 1.0 | 2249 | 6.6652 |
| 6.4134 | 2.0 | 4498 | 6.3987 |
| 6.2507 | 3.0 | 6747 | 6.3227 |
### Framework versions
- Transformers 4.9.1
- Pytorch 1.9.0+cu102
- Datasets 1.10.2
- Tokenizers 0.10.3
|
MaCHeaMaRoBi/DialoGPT-small-joshua | 2c91a3a53d388ccf429463689722e20184cd7352 | 2021-09-19T12:17:41.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | MaCHeaMaRoBi | null | MaCHeaMaRoBi/DialoGPT-small-joshua | 1 | null | transformers | 28,115 | ---
tags:
- conversational
---
Rick DialoGPT Model |
MaalK/DialoGPT-small-Petyr | 42ea6b6d01c8fe46d419d4a686110b3158513155 | 2021-06-02T06:34:48.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers"
] | text-generation | false | MaalK | null | MaalK/DialoGPT-small-Petyr | 1 | null | transformers | 28,116 | Entry not found |
MadhanKumar/DialoGPT-small-HarryPotter | c02117b7c3581caf48483853b04b45ae0025f833 | 2021-08-29T05:11:51.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | MadhanKumar | null | MadhanKumar/DialoGPT-small-HarryPotter | 1 | null | transformers | 28,117 | ---
tags:
- conversational
---
#Harry Potter DialoGPT Model |
Mads/wav2vec2-xlsr-large-53-kor-financial-engineering | 5785e67ad3bd0df18e375dc7f493f4d850b7dfab | 2021-07-05T15:07:54.000Z | [
"pytorch",
"jax",
"wav2vec2",
"automatic-speech-recognition",
"transformers"
] | automatic-speech-recognition | false | Mads | null | Mads/wav2vec2-xlsr-large-53-kor-financial-engineering | 1 | null | transformers | 28,118 | # WIP
|
Mads/xlsr-demo-2 | 3121c9f40fef36825e417f6e6e2cbcd352d7d19b | 2021-07-05T15:17:16.000Z | [
"pytorch",
"jax",
"wav2vec2",
"automatic-speech-recognition",
"transformers"
] | automatic-speech-recognition | false | Mads | null | Mads/xlsr-demo-2 | 1 | null | transformers | 28,119 | Entry not found |
Mads/xlsr-demo-eng | 43cff8a792892b32347153ed5403b1d55f488c5f | 2021-07-05T15:24:35.000Z | [
"pytorch",
"jax",
"wav2vec2",
"automatic-speech-recognition",
"transformers"
] | automatic-speech-recognition | false | Mads | null | Mads/xlsr-demo-eng | 1 | null | transformers | 28,120 | Entry not found |
MaggieXM/distilbert-base-uncased-finetuned-squad | cdc2c3c501f50ca0fce95b4a8804ec9d760ac973 | 2022-02-05T04:50:41.000Z | [
"pytorch",
"tensorboard",
"distilbert",
"question-answering",
"dataset:squad",
"transformers",
"generated_from_trainer",
"license:apache-2.0",
"model-index",
"autotrain_compatible"
] | question-answering | false | MaggieXM | null | MaggieXM/distilbert-base-uncased-finetuned-squad | 1 | null | transformers | 28,121 | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- squad
model-index:
- name: distilbert-base-uncased-finetuned-squad
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# distilbert-base-uncased-finetuned-squad
This model is a fine-tuned version of [distilbert-base-uncased](https://huggingface.co/distilbert-base-uncased) on the squad dataset.
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 0.01
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| No log | 0.01 | 56 | 4.8054 |
### Framework versions
- Transformers 4.16.2
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.11.0
|
MaiaMaiaMaia/DialoGPT-medium-PeterParkerBot | 1640b3bf7c8ac625a6aa44328ddb97de2552a48e | 2021-12-13T09:32:35.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers"
] | text-generation | false | MaiaMaiaMaia | null | MaiaMaiaMaia/DialoGPT-medium-PeterParkerBot | 1 | null | transformers | 28,122 | ----
tags:
- conversational
---
#Peter Parker DialoGPT Model |
Makesh/adelectra-base | c4e2da7dfbb86838e12ae4c5c6ee9859f5cdd0e9 | 2021-06-22T13:27:07.000Z | [
"pytorch",
"electra",
"feature-extraction",
"transformers"
] | feature-extraction | false | Makesh | null | Makesh/adelectra-base | 1 | null | transformers | 28,123 | Entry not found |
Makesh/adelectra-large | 2ad505a0b5e2ee7b61393056cea58420eb2ffce7 | 2021-06-21T22:50:41.000Z | [
"pytorch",
"electra",
"feature-extraction",
"transformers"
] | feature-extraction | false | Makesh | null | Makesh/adelectra-large | 1 | null | transformers | 28,124 | Entry not found |
Makesh/adelectra-small | 92acec016bd5439a9fadef4eeec9d25c7ae51090 | 2021-06-21T22:41:10.000Z | [
"pytorch",
"electra",
"feature-extraction",
"transformers"
] | feature-extraction | false | Makesh | null | Makesh/adelectra-small | 1 | null | transformers | 28,125 | Entry not found |
Mapcar/pegasus-samsum | 387967e619fbc3b47be30bfa4fdc2559e038ecab | 2022-02-04T03:27:33.000Z | [
"pytorch",
"tensorboard",
"pegasus",
"text2text-generation",
"dataset:samsum",
"transformers",
"generated_from_trainer",
"model-index",
"autotrain_compatible"
] | text2text-generation | false | Mapcar | null | Mapcar/pegasus-samsum | 1 | null | transformers | 28,126 | ---
tags:
- generated_from_trainer
datasets:
- samsum
model-index:
- name: pegasus-samsum
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# pegasus-samsum
This model is a fine-tuned version of [google/pegasus-cnn_dailymail](https://huggingface.co/google/pegasus-cnn_dailymail) on the samsum dataset.
It achieves the following results on the evaluation set:
- Loss: 1.4844
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- gradient_accumulation_steps: 16
- total_train_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 1
### Training results
| Training Loss | Epoch | Step | Validation Loss |
|:-------------:|:-----:|:----:|:---------------:|
| 1.6936 | 0.54 | 500 | 1.4844 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.16.1
- Tokenizers 0.10.3
|
Martinlabla/bert_cn_finetunning | ce9fbabae129c6a7759d8eb5597f1195a90bc0e2 | 2022-02-11T12:49:15.000Z | [
"pytorch",
"bert",
"fill-mask",
"transformers",
"autotrain_compatible"
] | fill-mask | false | Martinlabla | null | Martinlabla/bert_cn_finetunning | 1 | null | transformers | 28,127 | Entry not found |
MaryaAI/opus-mt-en-ar-finetuned-dummyData-10-10-ar-to-en | 5df607a7fdb88e4ae2d78298d082167f93449b58 | 2021-10-10T06:33:20.000Z | [
"pytorch",
"tensorboard",
"marian",
"text2text-generation",
"dataset:syssr_en_ar",
"transformers",
"generated_from_trainer",
"license:apache-2.0",
"model-index",
"autotrain_compatible"
] | text2text-generation | false | MaryaAI | null | MaryaAI/opus-mt-en-ar-finetuned-dummyData-10-10-ar-to-en | 1 | null | transformers | 28,128 | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- syssr_en_ar
metrics:
- bleu
model-index:
- name: opus-mt-en-ar-finetuned-dummyData-10-10-ar-to-en
results:
- task:
name: Sequence-to-sequence Language Modeling
type: text2text-generation
dataset:
name: syssr_en_ar
type: syssr_en_ar
args: default
metrics:
- name: Bleu
type: bleu
value: 7.9946
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# opus-mt-en-ar-finetuned-dummyData-10-10-ar-to-en
This model is a fine-tuned version of [Helsinki-NLP/opus-mt-en-ar](https://huggingface.co/Helsinki-NLP/opus-mt-en-ar) on the syssr_en_ar dataset.
It achieves the following results on the evaluation set:
- Loss: 1.2046
- Bleu: 7.9946
- Gen Len: 20.0
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 2e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 5
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Gen Len |
|:-------------:|:-----:|:----:|:---------------:|:------:|:-------:|
| No log | 1.0 | 1 | 1.2038 | 7.9946 | 20.0 |
| No log | 2.0 | 2 | 1.2038 | 7.9946 | 20.0 |
| No log | 3.0 | 3 | 1.2038 | 7.9946 | 20.0 |
| No log | 4.0 | 4 | 1.2036 | 7.9946 | 20.0 |
| No log | 5.0 | 5 | 1.2046 | 7.9946 | 20.0 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.9.0+cu111
- Datasets 1.12.1
- Tokenizers 0.10.3
|
MaxW0748/DialoGPT-small-Rick | 22c726459f4cc92b1fb32049b99d5e32efdf2775 | 2021-11-03T02:51:13.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | MaxW0748 | null | MaxW0748/DialoGPT-small-Rick | 1 | null | transformers | 28,129 | ---
tags:
- conversational
---
#Rick and Morty DialoGPT Model |
Maya/essai1 | 4734616f597e7a070d82ce569c8a92395bd3be29 | 2021-04-10T16:32:44.000Z | [
"pytorch",
"marian",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | Maya | null | Maya/essai1 | 1 | null | transformers | 28,130 | hello
|
MehdiHosseiniMoghadam/m | 6d713266829c9be60d8164ba4c8dfff38a8f4b4f | 2021-07-05T15:38:28.000Z | [
"pytorch",
"jax",
"wav2vec2",
"automatic-speech-recognition",
"transformers"
] | automatic-speech-recognition | false | MehdiHosseiniMoghadam | null | MehdiHosseiniMoghadam/m | 1 | null | transformers | 28,131 | Entry not found |
MehdiHosseiniMoghadam/wav2vec2-large-xlsr-53-Dutch | 39ec872bbded01a282e26ec7959f11ddd0e1a332 | 2021-07-05T15:48:59.000Z | [
"pytorch",
"jax",
"wav2vec2",
"automatic-speech-recognition",
"nl",
"dataset:common_voice",
"transformers",
"audio",
"speech",
"xlsr-fine-tuning-week",
"license:apache-2.0",
"model-index"
] | automatic-speech-recognition | false | MehdiHosseiniMoghadam | null | MehdiHosseiniMoghadam/wav2vec2-large-xlsr-53-Dutch | 1 | null | transformers | 28,132 | ---
language: nl
datasets:
- common_voice
tags:
- audio
- automatic-speech-recognition
- speech
- xlsr-fine-tuning-week
license: apache-2.0
model-index:
- name: wav2vec2-large-xlsr-53-Dutch by Mehdi Hosseini Moghadam
results:
- task:
name: Speech Recognition
type: automatic-speech-recognition
dataset:
name: Common Voice nl
type: common_voice
args: nl
metrics:
- name: Test WER
type: wer
value: 26.494162
---
# wav2vec2-large-xlsr-53-Dutch
Fine-tuned [facebook/wav2vec2-large-xlsr-53](https://huggingface.co/facebook/wav2vec2-large-xlsr-53) in Dutch using the [Common Voice](https://huggingface.co/datasets/common_voice)
When using this model, make sure that your speech input is sampled at 16kHz.
## Usage
The model can be used directly (without a language model) as follows:
```python
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "nl", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("MehdiHosseiniMoghadam/wav2vec2-large-xlsr-53-Dutch")
model = Wav2Vec2ForCTC.from_pretrained("MehdiHosseiniMoghadam/wav2vec2-large-xlsr-53-Dutch")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][:2])
```
## Evaluation
The model can be evaluated as follows on the Dutch test data of Common Voice.
```python
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "nl", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("MehdiHosseiniMoghadam/wav2vec2-large-xlsr-53-Dutch")
model = Wav2Vec2ForCTC.from_pretrained("MehdiHosseiniMoghadam/wav2vec2-large-xlsr-53-Dutch")
model.to("cuda")
chars_to_ignore_regex = '[\\\\,\\\\?\\\\.\\\\!\\\\-\\\\;\\\\:\\\\"\\\\“]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def speech_file_to_array_fn(batch):
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
# Preprocessing the datasets.
# We need to read the aduio files as arrays
def evaluate(batch):
inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
pred_ids = torch.argmax(logits, dim=-1)
batch["pred_strings"] = processor.batch_decode(pred_ids)
return batch
result = test_dataset.map(evaluate, batched=True, batch_size=8)
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
```
**Test Result**: 26.494162 %
## Training
The Common Voice `train`, `validation` datasets were used for training. |
MeysamAC/packt_bert_random | 26afd185b59fcf4a0a70dfe6a6706edd44a39b6a | 2021-07-15T13:37:27.000Z | [
"pytorch",
"bert",
"feature-extraction",
"transformers"
] | feature-extraction | false | MeysamAC | null | MeysamAC/packt_bert_random | 1 | null | transformers | 28,133 | Entry not found |
MhF/xlm-roberta-base-finetuned-panx-de-fr | f01a8d5170749cc9af1d7bf4266b9cf8d70978ea | 2022-02-17T03:30:32.000Z | [
"pytorch",
"xlm-roberta",
"token-classification",
"transformers",
"generated_from_trainer",
"license:mit",
"model-index",
"autotrain_compatible"
] | token-classification | false | MhF | null | MhF/xlm-roberta-base-finetuned-panx-de-fr | 1 | null | transformers | 28,134 | ---
license: mit
tags:
- generated_from_trainer
metrics:
- f1
model-index:
- name: xlm-roberta-base-finetuned-panx-de-fr
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# xlm-roberta-base-finetuned-panx-de-fr
This model is a fine-tuned version of [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1576
- F1: 0.8571
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 24
- eval_batch_size: 24
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | F1 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.2924 | 1.0 | 715 | 0.1819 | 0.8286 |
| 0.1503 | 2.0 | 1430 | 0.1580 | 0.8511 |
| 0.0972 | 3.0 | 2145 | 0.1576 | 0.8571 |
### Framework versions
- Transformers 4.16.2
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.11.0
|
MhF/xlm-roberta-base-finetuned-panx-en | df51c6367cc0bc6f2a4b549c049c569a72116a7a | 2022-02-17T03:40:08.000Z | [
"pytorch",
"xlm-roberta",
"token-classification",
"dataset:xtreme",
"transformers",
"generated_from_trainer",
"license:mit",
"model-index",
"autotrain_compatible"
] | token-classification | false | MhF | null | MhF/xlm-roberta-base-finetuned-panx-en | 1 | null | transformers | 28,135 | ---
license: mit
tags:
- generated_from_trainer
datasets:
- xtreme
metrics:
- f1
model-index:
- name: xlm-roberta-base-finetuned-panx-en
results:
- task:
name: Token Classification
type: token-classification
dataset:
name: xtreme
type: xtreme
args: PAN-X.en
metrics:
- name: F1
type: f1
value: 0.6807563959955506
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# xlm-roberta-base-finetuned-panx-en
This model is a fine-tuned version of [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the xtreme dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3856
- F1: 0.6808
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 24
- eval_batch_size: 24
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | F1 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 1.1038 | 1.0 | 50 | 0.5255 | 0.5331 |
| 0.4922 | 2.0 | 100 | 0.4377 | 0.6379 |
| 0.3664 | 3.0 | 150 | 0.3856 | 0.6808 |
### Framework versions
- Transformers 4.16.2
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.11.0
|
MhF/xlm-roberta-base-finetuned-panx-fr | a696a32f2a68bc4e0855b8e7451a9a85545dd3a7 | 2022-02-17T03:34:17.000Z | [
"pytorch",
"xlm-roberta",
"token-classification",
"dataset:xtreme",
"transformers",
"generated_from_trainer",
"license:mit",
"model-index",
"autotrain_compatible"
] | token-classification | false | MhF | null | MhF/xlm-roberta-base-finetuned-panx-fr | 1 | null | transformers | 28,136 | ---
license: mit
tags:
- generated_from_trainer
datasets:
- xtreme
metrics:
- f1
model-index:
- name: xlm-roberta-base-finetuned-panx-fr
results:
- task:
name: Token Classification
type: token-classification
dataset:
name: xtreme
type: xtreme
args: PAN-X.fr
metrics:
- name: F1
type: f1
value: 0.8353494623655915
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# xlm-roberta-base-finetuned-panx-fr
This model is a fine-tuned version of [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the xtreme dataset.
It achieves the following results on the evaluation set:
- Loss: 0.2736
- F1: 0.8353
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 24
- eval_batch_size: 24
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | F1 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.5826 | 1.0 | 191 | 0.3442 | 0.7888 |
| 0.2669 | 2.0 | 382 | 0.2848 | 0.8326 |
| 0.1818 | 3.0 | 573 | 0.2736 | 0.8353 |
### Framework versions
- Transformers 4.16.2
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.11.0
|
MhF/xlm-roberta-base-finetuned-panx-it | b6f8e78a7194af37f279b687edc4989aae39d0d3 | 2022-02-17T03:37:20.000Z | [
"pytorch",
"xlm-roberta",
"token-classification",
"dataset:xtreme",
"transformers",
"generated_from_trainer",
"license:mit",
"model-index",
"autotrain_compatible"
] | token-classification | false | MhF | null | MhF/xlm-roberta-base-finetuned-panx-it | 1 | null | transformers | 28,137 | ---
license: mit
tags:
- generated_from_trainer
datasets:
- xtreme
metrics:
- f1
model-index:
- name: xlm-roberta-base-finetuned-panx-it
results:
- task:
name: Token Classification
type: token-classification
dataset:
name: xtreme
type: xtreme
args: PAN-X.it
metrics:
- name: F1
type: f1
value: 0.8213114754098361
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# xlm-roberta-base-finetuned-panx-it
This model is a fine-tuned version of [xlm-roberta-base](https://huggingface.co/xlm-roberta-base) on the xtreme dataset.
It achieves the following results on the evaluation set:
- Loss: 0.2491
- F1: 0.8213
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 24
- eval_batch_size: 24
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3
### Training results
| Training Loss | Epoch | Step | Validation Loss | F1 |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.8192 | 1.0 | 70 | 0.3300 | 0.7184 |
| 0.2949 | 2.0 | 140 | 0.2817 | 0.7959 |
| 0.189 | 3.0 | 210 | 0.2491 | 0.8213 |
### Framework versions
- Transformers 4.16.2
- Pytorch 1.10.0+cu113
- Datasets 1.18.3
- Tokenizers 0.11.0
|
MichaelTheLearner/DialoGPT-medium-harry | 9b11a723927b007f9bada5123f36abb182590583 | 2021-09-15T23:49:04.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | MichaelTheLearner | null | MichaelTheLearner/DialoGPT-medium-harry | 1 | null | transformers | 28,138 | ---
tags:
- conversational
---
# Harry Potter DialoGPT Model |
MightyCoderX/DialoGPT-medium-EdwardElric | 8dbb0fe316aafc7e3ce555d5937ef0fa6c1f9cea | 2021-09-24T18:20:42.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | MightyCoderX | null | MightyCoderX/DialoGPT-medium-EdwardElric | 1 | null | transformers | 28,139 | ---
tags:
- conversational
---
# Edward Elric DialoGPT Model |
MigyeongYang/kcbert-mlm-finetune | 02e72115dc83d1a01077f45926e65fd521a10f5e | 2021-11-12T02:50:57.000Z | [
"pytorch",
"bert",
"fill-mask",
"transformers",
"autotrain_compatible"
] | fill-mask | false | MigyeongYang | null | MigyeongYang/kcbert-mlm-finetune | 1 | null | transformers | 28,140 | kcbert-mlm-finetune |
MisterFavourite/Genesis_KJV_fine_tuned | 8260023d6ac62a16a3082e37e93209d3a064d516 | 2021-12-01T22:32:46.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers"
] | text-generation | false | MisterFavourite | null | MisterFavourite/Genesis_KJV_fine_tuned | 1 | null | transformers | 28,141 | Entry not found |
MisterFavourite/Sherlock_Holmes_fine_tuned | 8a45fe262f1e3acc9c78316ac0ed3bdda9e3c4c8 | 2021-12-01T22:43:19.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers"
] | text-generation | false | MisterFavourite | null | MisterFavourite/Sherlock_Holmes_fine_tuned | 1 | null | transformers | 28,142 | Entry not found |
ModzabazeR/small-okaberintaro | 0c6ce8c4e3d0a89d685ffe9d34bf691c768e1034 | 2021-09-15T10:38:59.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | ModzabazeR | null | ModzabazeR/small-okaberintaro | 1 | null | transformers | 28,143 | ---
tags:
- conversational
---
# Okabe Rintaro DialoGPT Model |
MohamedHesham/bart_large_subset_cnn | 70cd97980c8d743fcfdcb51f101c97d0414a0e31 | 2021-06-12T14:29:52.000Z | [
"pytorch",
"bart",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | MohamedHesham | null | MohamedHesham/bart_large_subset_cnn | 1 | null | transformers | 28,144 | Entry not found |
Monsia/test-model-lg-data | 87370fbd9db7a391dae6576169cb3bebf079e272 | 2021-12-23T14:03:38.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"dataset:common_voice",
"transformers",
"generated_from_trainer",
"license:apache-2.0",
"model-index"
] | automatic-speech-recognition | false | Monsia | null | Monsia/test-model-lg-data | 1 | null | transformers | 28,145 | ---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- common_voice
model-index:
- name: test-model-lg-data
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# test-model-lg-data
This model is a fine-tuned version of [Monsia/test-model-lg-data](https://huggingface.co/Monsia/test-model-lg-data) on the common_voice dataset.
It achieves the following results on the evaluation set:
- Loss: 0.3354
- Wer: 0.4150
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0003
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 200
- num_epochs: 5
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 0.0236 | 0.67 | 100 | 0.4048 | 0.4222 |
| 0.0304 | 1.35 | 200 | 0.4266 | 0.4809 |
| 0.0545 | 2.03 | 300 | 0.4309 | 0.4735 |
| 0.0415 | 2.7 | 400 | 0.4269 | 0.4595 |
| 0.033 | 3.38 | 500 | 0.4085 | 0.4537 |
| 0.0328 | 4.05 | 600 | 0.3642 | 0.4224 |
| 0.0414 | 4.73 | 700 | 0.3354 | 0.4150 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu113
- Datasets 1.13.3
- Tokenizers 0.10.3
|
Motahar/dummy-model | c06e3fa05bb4416595af0a08e9da006ad2d01c77 | 2021-12-30T16:14:49.000Z | [
"pytorch",
"bert",
"feature-extraction",
"transformers"
] | feature-extraction | false | Motahar | null | Motahar/dummy-model | 1 | null | transformers | 28,146 | Entry not found |
Mousumi/finetuned_bart | 00a8f19a8a27ffddf27bf13265ccd5ac01e4c749 | 2021-10-03T15:37:02.000Z | [
"pytorch",
"bart",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | Mousumi | null | Mousumi/finetuned_bart | 1 | null | transformers | 28,147 | ### Description:
BART Model has been finetuned on CNN/DailyMail Dataset with Sample Size 10000.
### How To Use:
```
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
src_text = [" PG&E stated it scheduled the blackouts in response to forecasts for high winds amid dry conditions. The aim is to reduce the risk of wildfires. Nearly 800 thousand customers were scheduled to be affected by the shutoffs which were expected to last through at least midday tomorrow.", "In the end, it played out like a movie. A tense, heartbreaking story, and then a surprise twist at the end. As eight of Mary Jane Veloso's fellow death row inmates -- mostly foreigners, like her -- were put to death by firing squad early Wednesday in a wooded grove on the Indonesian island of Nusa Kambangan, the Filipina maid and mother of two was spared, at least for now. Her family was returning from what they thought was their final visit to the prison on so-called \"execution island\" when a Philippine TV crew flagged their bus down to tell them of the decision to postpone her execution. Her ecstatic mother, Celia Veloso, told CNN: \"We are so happy, so happy. I thought I had lost my daughter already but God is so good. Thank you to everyone who helped us."]
torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = AutoTokenizer.from_pretrained("Mousumi/finetuned_bart")
model = AutoModelForSeq2SeqLM.from_pretrained("Mousumi/finetuned_bart").to(torch_device)
no_samples = len(src_text)
result = []
for i in range(no_samples):
with tokenizer.as_target_tokenizer():
tokenized_text = tokenizer([src_text[i]], return_tensors='pt', padding=True, truncation=True)
batch = tokenized_text.to(torch_device)
translated = model.generate(**batch)
tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
result.append(tgt_text[0])
print(result)
``` |
Mousumi/finetuned_pegasus | 49228137d360a832674ee09a9ff08af76fb82ded | 2021-10-03T15:36:45.000Z | [
"pytorch",
"pegasus",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | Mousumi | null | Mousumi/finetuned_pegasus | 1 | null | transformers | 28,148 | ### Description:
Pegasus Model has been finetuned on CNN/DailyMail Dataset with Sample Size 10000.
### How To Use:
```
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
src_text = [" PG&E stated it scheduled the blackouts in response to forecasts for high winds amid dry conditions. The aim is to reduce the risk of wildfires. Nearly 800 thousand customers were scheduled to be affected by the shutoffs which were expected to last through at least midday tomorrow.", "In the end, it played out like a movie. A tense, heartbreaking story, and then a surprise twist at the end. As eight of Mary Jane Veloso's fellow death row inmates -- mostly foreigners, like her -- were put to death by firing squad early Wednesday in a wooded grove on the Indonesian island of Nusa Kambangan, the Filipina maid and mother of two was spared, at least for now. Her family was returning from what they thought was their final visit to the prison on so-called \"execution island\" when a Philippine TV crew flagged their bus down to tell them of the decision to postpone her execution. Her ecstatic mother, Celia Veloso, told CNN: \"We are so happy, so happy. I thought I had lost my daughter already but God is so good. Thank you to everyone who helped us."]
torch_device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = AutoTokenizer.from_pretrained("Mousumi/finetuned_pegasus")
model = AutoModelForSeq2SeqLM.from_pretrained("Mousumi/finetuned_pegasus").to(torch_device)
no_samples = len(src_text)
result = []
for i in range(no_samples):
with tokenizer.as_target_tokenizer():
tokenized_text = tokenizer([src_text[i]], return_tensors='pt', padding=True, truncation=True)
batch = tokenized_text.to(torch_device)
translated = model.generate(**batch)
tgt_text = tokenizer.batch_decode(translated, skip_special_tokens=True)
result.append(tgt_text[0])
print(result)
``` |
MrE/DialoGPT-medium-SARGER1 | 9b9470f0b787af4c0815db830a942266ba63debc | 2021-10-18T03:43:32.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | MrE | null | MrE/DialoGPT-medium-SARGER1 | 1 | null | transformers | 28,149 | ---
tags:
- conversational
---
# Sarge |
Muennighoff/SBERT-base-msmarco-bitfit | 0b6465c4501a7761eb8ead8e2b46ec3581d281bd | 2022-02-21T05:58:04.000Z | [
"pytorch",
"bert",
"feature-extraction",
"arxiv:2202.08904",
"sentence-transformers",
"sentence-similarity",
"transformers"
] | sentence-similarity | false | Muennighoff | null | Muennighoff/SBERT-base-msmarco-bitfit | 1 | null | sentence-transformers | 28,150 | ---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
---
# SBERT-base-msmarco-bitfit
## Usage
For usage instructions, refer to our codebase: https://github.com/Muennighoff/sgpt
## Evaluation Results
For eval results, refer to our paper: https://arxiv.org/abs/2202.08904
## Training
The model was trained with the parameters:
**DataLoader**:
`torch.utils.data.dataloader.DataLoader` of length 15600 with parameters:
```
{'batch_size': 32, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}
```
**Loss**:
`sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss` with parameters:
```
{'scale': 20.0, 'similarity_fct': 'cos_sim'}
```
Parameters of the fit()-Method:
```
{
"epochs": 10,
"evaluation_steps": 0,
"evaluator": "NoneType",
"max_grad_norm": 1,
"optimizer_class": "<class 'transformers.optimization.AdamW'>",
"optimizer_params": {
"lr": 0.0002
},
"scheduler": "WarmupLinear",
"steps_per_epoch": null,
"warmup_steps": 1000,
"weight_decay": 0.01
}
```
## Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 300, 'do_lower_case': False}) with Transformer model: BertModel
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
)
```
## Citing & Authors
```bibtex
@article{muennighoff2022sgpt,
title={SGPT: GPT Sentence Embeddings for Semantic Search},
author={Muennighoff, Niklas},
journal={arXiv preprint arXiv:2202.08904},
year={2022}
}
```
|
Muennighoff/SBERT-base-nli-stsb-v2 | 4d5713a1a4207c0ebb259274d7a3ac10b6845c68 | 2021-12-31T07:59:14.000Z | [
"pytorch",
"bert",
"feature-extraction",
"sentence-transformers",
"sentence-similarity",
"transformers",
"license:apache-2.0"
] | sentence-similarity | false | Muennighoff | null | Muennighoff/SBERT-base-nli-stsb-v2 | 1 | null | sentence-transformers | 28,151 | ---
pipeline_tag: sentence-similarity
license: apache-2.0
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
---
This model is used in "TSDAE: Using Transformer-based Sequential Denoising Auto-Encoder for Unsupervised Sentence Embedding Learning".
|
Muennighoff/SGPT-125M-mean-nli-bitfit | 7f4eec6d68aae9464b71120cd4e29e98d6f6590a | 2022-02-21T06:08:15.000Z | [
"pytorch",
"gpt_neo",
"feature-extraction",
"arxiv:2202.08904",
"sentence-transformers",
"sentence-similarity",
"transformers"
] | sentence-similarity | false | Muennighoff | null | Muennighoff/SGPT-125M-mean-nli-bitfit | 1 | 1 | sentence-transformers | 28,152 | ---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
- transformers
---
# SGPT-125M-mean-nli-bitfit
## Usage
For usage instructions, refer to our codebase: https://github.com/Muennighoff/sgpt
## Evaluation Results
For eval results, refer to our paper: https://arxiv.org/abs/2202.08904
## Training
The model was trained with the parameters:
**DataLoader**:
`sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader` of length 8807 with parameters:
```
{'batch_size': 64}
```
**Loss**:
`sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss` with parameters:
```
{'scale': 20.0, 'similarity_fct': 'cos_sim'}
```
Parameters of the fit()-Method:
```
{
"epochs": 1,
"evaluation_steps": 880,
"evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
"max_grad_norm": 1,
"optimizer_class": "<class 'transformers.optimization.AdamW'>",
"optimizer_params": {
"lr": 0.0002
},
"scheduler": "WarmupLinear",
"steps_per_epoch": null,
"warmup_steps": 881,
"weight_decay": 0.01
}
```
## Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 75, 'do_lower_case': False}) with Transformer model: GPTNeoModel
(1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
)
```
## Citing & Authors
```bibtex
@article{muennighoff2022sgpt,
title={SGPT: GPT Sentence Embeddings for Semantic Search},
author={Muennighoff, Niklas},
journal={arXiv preprint arXiv:2202.08904},
year={2022}
}
```
|
Muennighoff/SGPT-125M-mean-nli-linearthenpool5 | dd054e39070dad240e5400b81eaf8f6d495bf564 | 2022-02-21T06:10:51.000Z | [
"pytorch",
"gpt_neo",
"feature-extraction",
"arxiv:2202.08904",
"sentence-transformers",
"sentence-similarity"
] | sentence-similarity | false | Muennighoff | null | Muennighoff/SGPT-125M-mean-nli-linearthenpool5 | 1 | null | sentence-transformers | 28,153 | ---
pipeline_tag: sentence-similarity
tags:
- sentence-transformers
- feature-extraction
- sentence-similarity
---
# SGPT-125M-mean-nli-linearthenpool5
## Usage
For usage instructions, refer to our codebase: https://github.com/Muennighoff/sgpt
## Evaluation Results
For eval results, refer to our paper: https://arxiv.org/abs/2202.08904
## Training
The model was trained with the parameters:
**DataLoader**:
`sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader` of length 8807 with parameters:
```
{'batch_size': 64}
```
**Loss**:
`sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss` with parameters:
```
{'scale': 20.0, 'similarity_fct': 'cos_sim'}
```
Parameters of the fit()-Method:
```
{
"epochs": 1,
"evaluation_steps": 880,
"evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
"max_grad_norm": 1,
"optimizer_class": "<class 'transformers.optimization.AdamW'>",
"optimizer_params": {
"lr": 2e-05
},
"scheduler": "WarmupLinear",
"steps_per_epoch": null,
"warmup_steps": 881,
"weight_decay": 0.01
}
```
## Full Model Architecture
```
SentenceTransformer(
(0): Transformer({'max_seq_length': 75, 'do_lower_case': False}) with Transformer model: GPTNeoModel
(1): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.GELU', 'key_name': 'token_embeddings'})
(2): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.GELU', 'key_name': 'token_embeddings'})
(3): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.GELU', 'key_name': 'token_embeddings'})
(4): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.GELU', 'key_name': 'token_embeddings'})
(5): Dense({'in_features': 768, 'out_features': 768, 'bias': True, 'activation_function': 'torch.nn.modules.activation.GELU', 'key_name': 'token_embeddings'})
(6): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False})
)
```
## Citing & Authors
```bibtex
@article{muennighoff2022sgpt,
title={SGPT: GPT Sentence Embeddings for Semantic Search},
author={Muennighoff, Niklas},
journal={arXiv preprint arXiv:2202.08904},
year={2022}
}
```
|
Muennighoff/tiny-random-bert | 5df856b53cdab85662eea8c499dcca56660e7e2a | 2021-11-19T16:18:11.000Z | [
"pytorch",
"tf",
"bert",
"transformers"
] | null | false | Muennighoff | null | Muennighoff/tiny-random-bert | 1 | null | transformers | 28,154 | Entry not found |
MultiBertGunjanPatrick/multiberts-seed-0-1300k | b1f45b2306ae89c9173480c269120109031d7752 | 2021-10-04T04:57:32.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-0",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-0-1300k | 1 | null | transformers | 28,155 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-0
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 0 Checkpoint 1300k (uncased)
Seed 0 intermediate checkpoint 1300k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-0](https://hf.co/multberts-seed-0). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-0-1300k')
model = BertModel.from_pretrained("multiberts-seed-0-1300k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-1-1400k | 2f96d8ddbdde784dd0fd0ef604e53c0ac0ceff7f | 2021-10-04T05:01:16.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-1",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-1-1400k | 1 | null | transformers | 28,156 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-1
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 1 Checkpoint 1400k (uncased)
Seed 1 intermediate checkpoint 1400k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-1](https://hf.co/multberts-seed-1). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-1-1400k')
model = BertModel.from_pretrained("multiberts-seed-1-1400k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-1-20k | 972d55877265b22864122367e6f8e852a90564b0 | 2021-10-04T04:58:38.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-1",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-1-20k | 1 | null | transformers | 28,157 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-1
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 1 Checkpoint 20k (uncased)
Seed 1 intermediate checkpoint 20k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-1](https://hf.co/multberts-seed-1). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-1-20k')
model = BertModel.from_pretrained("multiberts-seed-1-20k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-1-40k | 7eb6d5e7614b9ef482d22db8a4b88c94cbd5f56a | 2021-10-04T04:58:46.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-1",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-1-40k | 1 | null | transformers | 28,158 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-1
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 1 Checkpoint 40k (uncased)
Seed 1 intermediate checkpoint 40k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-1](https://hf.co/multberts-seed-1). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-1-40k')
model = BertModel.from_pretrained("multiberts-seed-1-40k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-13 | 220a163ac625f009d33fadf744f312e689029eda | 2021-10-04T04:50:46.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-13 | 1 | null | transformers | 28,159 | ---
language: en
tags:
- exbert
- multiberts
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 13 (uncased)
Seed 13 MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-13')
model = BertModel.from_pretrained("multiberts-seed-13")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-100k | b2e3188aedc6a79317e4b19d766483adc1039cad | 2021-10-04T05:02:43.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-100k | 1 | null | transformers | 28,160 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 100k (uncased)
Seed 2 intermediate checkpoint 100k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-100k')
model = BertModel.from_pretrained("multiberts-seed-2-100k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-1400k | c2aad5d1f51e2d36c5820301585748d41f5fad1f | 2021-10-04T05:04:59.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-1400k | 1 | null | transformers | 28,161 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 1400k (uncased)
Seed 2 intermediate checkpoint 1400k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-1400k')
model = BertModel.from_pretrained("multiberts-seed-2-1400k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-1600k | 783b632df2fa6d3e30704e86962807ff073968db | 2021-10-04T05:05:14.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-1600k | 1 | null | transformers | 28,162 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 1600k (uncased)
Seed 2 intermediate checkpoint 1600k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-1600k')
model = BertModel.from_pretrained("multiberts-seed-2-1600k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-180k | 2dab40820581dbfcd59f2f4af7cffad787787773 | 2021-10-04T05:03:12.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-180k | 1 | null | transformers | 28,163 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 180k (uncased)
Seed 2 intermediate checkpoint 180k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-180k')
model = BertModel.from_pretrained("multiberts-seed-2-180k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-20k | 71bad09fcb48bcd1295360cfa93417e0175a9c78 | 2021-10-04T05:02:14.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-20k | 1 | null | transformers | 28,164 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 20k (uncased)
Seed 2 intermediate checkpoint 20k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-20k')
model = BertModel.from_pretrained("multiberts-seed-2-20k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-40k | bb07e027f509b3ee2e9e8899ade4c010366b8b40 | 2021-10-04T05:02:21.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-40k | 1 | null | transformers | 28,165 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 40k (uncased)
Seed 2 intermediate checkpoint 40k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-40k')
model = BertModel.from_pretrained("multiberts-seed-2-40k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-60k | 233c87dc4ef04a5ea0a8315cc1121bab791f6b4e | 2021-10-04T05:02:28.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-60k | 1 | null | transformers | 28,166 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 60k (uncased)
Seed 2 intermediate checkpoint 60k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-60k')
model = BertModel.from_pretrained("multiberts-seed-2-60k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-2-900k | 8188d7930e4480be14e78b7d73a93589f697b7de | 2021-10-04T05:04:18.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-2",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-2-900k | 1 | null | transformers | 28,167 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-2
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 2 Checkpoint 900k (uncased)
Seed 2 intermediate checkpoint 900k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-2](https://hf.co/multberts-seed-2). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-2-900k')
model = BertModel.from_pretrained("multiberts-seed-2-900k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-20 | 7af31622d974e481a4ee3a00d0048308e6891543 | 2021-10-04T04:54:10.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-20 | 1 | null | transformers | 28,168 | ---
language: en
tags:
- exbert
- multiberts
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 20 (uncased)
Seed 20 MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-20')
model = BertModel.from_pretrained("multiberts-seed-20")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-3-100k | 1da99ecfe6a61f21dd9f4981858bc622e45ad0ad | 2021-10-04T05:06:29.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-3",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-3-100k | 1 | null | transformers | 28,169 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-3
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 3 Checkpoint 100k (uncased)
Seed 3 intermediate checkpoint 100k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-3](https://hf.co/multberts-seed-3). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-3-100k')
model = BertModel.from_pretrained("multiberts-seed-3-100k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-3-1400k | 84b806ae35cbdd500dcd93060c192d322c27fa01 | 2021-10-04T05:08:37.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-3",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-3-1400k | 1 | null | transformers | 28,170 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-3
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 3 Checkpoint 1400k (uncased)
Seed 3 intermediate checkpoint 1400k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-3](https://hf.co/multberts-seed-3). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-3-1400k')
model = BertModel.from_pretrained("multiberts-seed-3-1400k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-3-140k | bc75c65c06a18ec7a2eabebf30364d88576cdc02 | 2021-10-04T05:06:44.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-3",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-3-140k | 1 | null | transformers | 28,171 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-3
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 3 Checkpoint 140k (uncased)
Seed 3 intermediate checkpoint 140k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-3](https://hf.co/multberts-seed-3). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-3-140k')
model = BertModel.from_pretrained("multiberts-seed-3-140k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-3-40k | 4f5b7339fbb8d57626a0933987380d6cb1ee3723 | 2021-10-04T05:06:08.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-3",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-3-40k | 1 | null | transformers | 28,172 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-3
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 3 Checkpoint 40k (uncased)
Seed 3 intermediate checkpoint 40k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-3](https://hf.co/multberts-seed-3). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-3-40k')
model = BertModel.from_pretrained("multiberts-seed-3-40k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-3-700k | d87b7bc0e1aafae2c6840efb539e6b257c5ba3cc | 2021-10-04T05:07:46.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-3",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-3-700k | 1 | null | transformers | 28,173 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-3
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 3 Checkpoint 700k (uncased)
Seed 3 intermediate checkpoint 700k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-3](https://hf.co/multberts-seed-3). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-3-700k')
model = BertModel.from_pretrained("multiberts-seed-3-700k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-3-900k | 3679b14602f32d8cea3b020c3b861c92a8ec6958 | 2021-10-04T05:08:00.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-3",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-3-900k | 1 | null | transformers | 28,174 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-3
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 3 Checkpoint 900k (uncased)
Seed 3 intermediate checkpoint 900k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-3](https://hf.co/multberts-seed-3). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-3-900k')
model = BertModel.from_pretrained("multiberts-seed-3-900k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-1200k | 901f7d7f0838148853bbc74c473f6665679584c3 | 2021-10-04T05:12:02.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-1200k | 1 | null | transformers | 28,175 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 1200k (uncased)
Seed 4 intermediate checkpoint 1200k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-1200k')
model = BertModel.from_pretrained("multiberts-seed-4-1200k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-140k | 9ad7e5b2f19d377bf2853ce7b9dd8bd339e109a7 | 2021-10-04T05:10:19.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-140k | 1 | null | transformers | 28,176 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 140k (uncased)
Seed 4 intermediate checkpoint 140k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-140k')
model = BertModel.from_pretrained("multiberts-seed-4-140k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-1500k | 2d0e6982023671750a513b3c2109b0bd4c555870 | 2021-10-04T05:12:23.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-1500k | 1 | null | transformers | 28,177 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 1500k (uncased)
Seed 4 intermediate checkpoint 1500k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-1500k')
model = BertModel.from_pretrained("multiberts-seed-4-1500k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-1600k | dda6ac1a682f880bf00989375d3476777fd37d22 | 2021-10-04T05:12:31.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-1600k | 1 | null | transformers | 28,178 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 1600k (uncased)
Seed 4 intermediate checkpoint 1600k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-1600k')
model = BertModel.from_pretrained("multiberts-seed-4-1600k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-1800k | dad9c83cabb2219d51cb2528ea22496bc6ecf6c4 | 2021-10-04T05:12:44.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-1800k | 1 | null | transformers | 28,179 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 1800k (uncased)
Seed 4 intermediate checkpoint 1800k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-1800k')
model = BertModel.from_pretrained("multiberts-seed-4-1800k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-180k | 5ce93485f88a72aded36ccfe53ecb8b5d5b789c2 | 2021-10-04T05:10:34.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-180k | 1 | null | transformers | 28,180 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 180k (uncased)
Seed 4 intermediate checkpoint 180k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-180k')
model = BertModel.from_pretrained("multiberts-seed-4-180k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-400k | aeb1ade657cf3cc7bab221c18064ac8e6ba9fe80 | 2021-10-04T05:11:03.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-400k | 1 | null | transformers | 28,181 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 400k (uncased)
Seed 4 intermediate checkpoint 400k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-400k')
model = BertModel.from_pretrained("multiberts-seed-4-400k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MultiBertGunjanPatrick/multiberts-seed-4-900k | e2a1e35cb811abc0fc5e375a95cae58339627fd1 | 2021-10-04T05:11:41.000Z | [
"pytorch",
"bert",
"pretraining",
"en",
"dataset:bookcorpus",
"dataset:wikipedia",
"arxiv:2106.16163",
"transformers",
"exbert",
"multiberts",
"multiberts-seed-4",
"license:apache-2.0"
] | null | false | MultiBertGunjanPatrick | null | MultiBertGunjanPatrick/multiberts-seed-4-900k | 1 | null | transformers | 28,182 | ---
language: en
tags:
- exbert
- multiberts
- multiberts-seed-4
license: apache-2.0
datasets:
- bookcorpus
- wikipedia
---
# MultiBERTs Seed 4 Checkpoint 900k (uncased)
Seed 4 intermediate checkpoint 900k MultiBERTs (pretrained BERT) model on English language using a masked language modeling (MLM) objective. It was introduced in
[this paper](https://arxiv.org/pdf/2106.16163.pdf) and first released in
[this repository](https://github.com/google-research/language/tree/master/language/multiberts). This is an intermediate checkpoint.
The final checkpoint can be found at [multiberts-seed-4](https://hf.co/multberts-seed-4). This model is uncased: it does not make a difference
between english and English.
Disclaimer: The team releasing MultiBERTs did not write a model card for this model so this model card has been written by [gchhablani](https://hf.co/gchhablani).
## Model description
MultiBERTs models are transformers model pretrained on a large corpus of English data in a self-supervised fashion. This means it
was pretrained on the raw texts only, with no humans labelling them in any way (which is why it can use lots of
publicly available data) with an automatic process to generate inputs and labels from those texts. More precisely, it
was pretrained with two objectives:
- Masked language modeling (MLM): taking a sentence, the model randomly masks 15% of the words in the input then run
the entire masked sentence through the model and has to predict the masked words. This is different from traditional
recurrent neural networks (RNNs) that usually see the words one after the other, or from autoregressive models like
GPT which internally mask the future tokens. It allows the model to learn a bidirectional representation of the
sentence.
- Next sentence prediction (NSP): the models concatenates two masked sentences as inputs during pretraining. Sometimes
they correspond to sentences that were next to each other in the original text, sometimes not. The model then has to
predict if the two sentences were following each other or not.
This way, the model learns an inner representation of the English language that can then be used to extract features
useful for downstream tasks: if you have a dataset of labeled sentences for instance, you can train a standard
classifier using the features produced by the MultiBERTs model as inputs.
## Intended uses & limitations
You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to
be fine-tuned on a downstream task. See the [model hub](https://huggingface.co/models?filter=multiberts) to look for
fine-tuned versions on a task that interests you.
Note that this model is primarily aimed at being fine-tuned on tasks that use the whole sentence (potentially masked)
to make decisions, such as sequence classification, token classification or question answering. For tasks such as text
generation you should look at model like GPT2.
### How to use
Here is how to use this model to get the features of a given text in PyTorch:
```python
from transformers import BertTokenizer, BertModel
tokenizer = BertTokenizer.from_pretrained('multiberts-seed-4-900k')
model = BertModel.from_pretrained("multiberts-seed-4-900k")
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### Limitations and bias
Even if the training data used for this model could be characterized as fairly neutral, this model can have biased
predictions. This bias will also affect all fine-tuned versions of this model. For an understanding of bias of this particular
checkpoint, please try out this checkpoint with the snippet present in the [Limitation and bias section](https://huggingface.co/bert-base-uncased#limitations-and-bias) of the [bert-base-uncased](https://huggingface.co/bert-base-uncased) checkpoint.
## Training data
The MultiBERTs models were pretrained on [BookCorpus](https://yknzhu.wixsite.com/mbweb), a dataset consisting of 11,038
unpublished books and [English Wikipedia](https://en.wikipedia.org/wiki/English_Wikipedia) (excluding lists, tables and
headers).
## Training procedure
### Preprocessing
The texts are lowercased and tokenized using WordPiece and a vocabulary size of 30,000. The inputs of the model are
then of the form:
```
[CLS] Sentence A [SEP] Sentence B [SEP]
```
With probability 0.5, sentence A and sentence B correspond to two consecutive sentences in the original corpus and in
the other cases, it's another random sentence in the corpus. Note that what is considered a sentence here is a
consecutive span of text usually longer than a single sentence. The only constrain is that the result with the two
"sentences" has a combined length of less than 512 tokens.
The details of the masking procedure for each sentence are the following:
- 15% of the tokens are masked.
- In 80% of the cases, the masked tokens are replaced by `[MASK]`.
- In 10% of the cases, the masked tokens are replaced by a random token (different) from the one they replace.
- In the 10% remaining cases, the masked tokens are left as is.
### Pretraining
The full model was trained on 16 Cloud TPU v2 chips for two million steps with a batch size
of 256. The sequence length was set to 512 throughout. The optimizer
used is Adam with a learning rate of 1e-4, \\(\beta_{1} = 0.9\\) and \\(\beta_{2} = 0.999\\), a weight decay of 0.01,
learning rate warmup for 10,000 steps and linear decay of the learning rate after.
### BibTeX entry and citation info
```bibtex
@article{DBLP:journals/corr/abs-2106-16163,
author = {Thibault Sellam and
Steve Yadlowsky and
Jason Wei and
Naomi Saphra and
Alexander D'Amour and
Tal Linzen and
Jasmijn Bastings and
Iulia Turc and
Jacob Eisenstein and
Dipanjan Das and
Ian Tenney and
Ellie Pavlick},
title = {The MultiBERTs: {BERT} Reproductions for Robustness Analysis},
journal = {CoRR},
volume = {abs/2106.16163},
year = {2021},
url = {https://arxiv.org/abs/2106.16163},
eprinttype = {arXiv},
eprint = {2106.16163},
timestamp = {Mon, 05 Jul 2021 15:15:50 +0200},
biburl = {https://dblp.org/rec/journals/corr/abs-2106-16163.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
```
<a href="https://huggingface.co/exbert/?model=multiberts">
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
</a>
|
MutazYoune/testt2 | 51a3aa1f7337c4d4856bbfa154f0be1cb1b57ca7 | 2021-05-18T21:45:57.000Z | [
"pytorch",
"jax",
"bert",
"fill-mask",
"transformers",
"autotrain_compatible"
] | fill-mask | false | MutazYoune | null | MutazYoune/testt2 | 1 | null | transformers | 28,183 | Entry not found |
NTUYG/SOTitle-java-BART | 4b78e2e9756af0d22f7f81d1f02ec213ed04f4e0 | 2021-01-28T15:12:29.000Z | [
"pytorch",
"bart",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | NTUYG | null | NTUYG/SOTitle-java-BART | 1 | null | transformers | 28,184 | ## How to use
```python
import logging
from simpletransformers.seq2seq import Seq2SeqModel, Seq2SeqArgs
logging.basicConfig(level=logging.INFO)
transformers_logger = logging.getLogger("transformers")
transformers_logger.setLevel(logging.WARNING)
model_args = Seq2SeqArgs()
# 加载本地训练好的模型
model = Seq2SeqModel(
encoder_decoder_type="bart",
encoder_decoder_name="NTUYG/SOTitle-java-BART",
args=model_args,
)
describe = """
I am a beginner at Android Java development but I have a few years of school + uni experience in Java. I am trying to write to a text file in an assets folder in my app using FileOutputStream but it doesn't seem to write to it at all since I am using InputStream to read the file after and there haven't any updates. Here is my code
"""
code = """
private void updateTextFile(String update) {
FileOutputStream fos = null;
try
{
fos = openFileOutput("Questions",MODE_PRIVATE);
fos.write("Testing".getBytes());
}
catch (FileNotFoundException e)
{
e.printStackTrace();
}
catch (IOException e)
{
e.printStackTrace();
}
finally
{
if(fos!=null)
{
try
{
fos.close();
}
catch (IOException e)
{
e.printStackTrace();
}
}
}
String text = "";
try
{
InputStream is = getAssets().open("Questions");
int size = is.available();
byte[] buffer = new byte[size];
is.read(buffer);
is.close();
text = new String(buffer);
}
catch (IOException e)
{
e.printStackTrace();
}
System.out.println("Tesing output " + text);
}
"""
from nltk import word_tokenize
describe = describe.replace('\n',' ').replace('\r',' ')
describe = ' '.join(word_tokenize(describe))
code = code.replace('\n',' ').replace('\r',' ')
code = ' '.join(word_tokenize(code))
# human ： Java Android Cant seem to update text file using FileOutputStream
body = describe + ' <code> ' + code +' </code>'
print(
model.predict(
[
body
]
)
)
```
|
NTUYG/SOTitle-js-BART | 893ccce8bbea372076d88f55d79edbbfb4c3238d | 2021-01-30T11:08:05.000Z | [
"pytorch",
"bart",
"text2text-generation",
"transformers",
"autotrain_compatible"
] | text2text-generation | false | NTUYG | null | NTUYG/SOTitle-js-BART | 1 | null | transformers | 28,185 | Entry not found |
Nabarun/DialoGPT-small-joshua | cca80989c9d28e95ce19f7213a2d47b566c2e234 | 2022-02-12T19:18:08.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Nabarun | null | Nabarun/DialoGPT-small-joshua | 1 | null | transformers | 28,186 | ---
tags:
- conversational
---
# My Awesome Model |
NamPE/DialoGPT-small-satouhina | 3443c8d890c268b62cb02726ed91f855832eff64 | 2021-12-31T03:28:21.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | NamPE | null | NamPE/DialoGPT-small-satouhina | 1 | null | transformers | 28,187 | ---
tags:
- conversational
---
# Satou Hina DialoGPT Model |
Narshion/bert-base-multilingual-cased-mwach | a7c6d059e982598045c544ec382345d0ca9f733a | 2021-10-15T20:25:20.000Z | [
"pytorch"
] | null | false | Narshion | null | Narshion/bert-base-multilingual-cased-mwach | 1 | null | null | 28,188 | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: test-mlm
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# test-mlm
This model is a fine-tuned version of [bert-base-multilingual-cased](https://huggingface.co/bert-base-multilingual-cased) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6481
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3.0
### Training results
### Framework versions
- Transformers 4.11.3
- Pytorch 1.9.0+cu111
- Datasets 1.12.1
- Tokenizers 0.10.3
|
Narsil/fr_pretrained | fab1ab2441ff9b75242a2b7b343e93f4b627c16a | 2020-01-30T08:39:05.000Z | [
"pytorch",
"transformers"
] | null | false | Narsil | null | Narsil/fr_pretrained | 1 | null | transformers | 28,189 | Entry not found |
Naturealbe/DialoGPT-small-harrypotter-2 | 018f1c875db25ed49371630af008767de5b0c644 | 2021-09-19T10:34:11.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Naturealbe | null | Naturealbe/DialoGPT-small-harrypotter-2 | 1 | null | transformers | 28,190 | ---
tags:
- conversational
---
# Harry Potter DialoGPT Model |
NbAiLab/wav2vec2-xls-r-1b-npsc-bokmaal | 3b0fa4c75cd6bbe79b7c4947139002a248a77259 | 2022-03-24T11:57:25.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"nb-NO",
"dataset:NbAiLab/NPSC",
"transformers",
"license:apache-2.0",
"model-index"
] | automatic-speech-recognition | false | NbAiLab | null | NbAiLab/wav2vec2-xls-r-1b-npsc-bokmaal | 1 | null | transformers | 28,191 |
---
license: apache-2.0
tags:
- generated_from_trainer
- automatic-speech-recognition
- NbAiLab/NPSC
- robust-speech-event
- false
- nb-NO
- hf-asr-leaderboard
datasets:
- NbAiLab/NPSC
language:
- nb-NO
model-index:
- name: wav2vec2-xls-r-1b-npsc-bokmaal
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: NPSC
type: NbAiLab/NPSC
args: 16K_mp3_bokmaal
metrics:
- name: "Test (Bokm\xE5l) WER"
type: wer
value: 0.07901700231893541
- name: "Test (Bokm\xE5l) CER"
type: cer
value: 0.029734583252347752
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# wav2vec2-xls-r-1b-npsc
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-1b](https://huggingface.co/facebook/wav2vec2-xls-r-1b) on the [NbAiLab/NPSC (16K_mp3_bokmaal)](https://huggingface.co/datasets/NbAiLab/NPSC/viewer/16K_mp3_bokmaal/train) dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1598
- WER: 0.0966
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 15.0
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 0.8361 | 0.32 | 500 | 0.6304 | 0.4970 |
| 0.5703 | 0.64 | 1000 | 0.3195 | 0.2775 |
| 0.5451 | 0.97 | 1500 | 0.2700 | 0.2246 |
| 0.47 | 1.29 | 2000 | 0.2564 | 0.2329 |
| 0.4063 | 1.61 | 2500 | 0.2459 | 0.2099 |
| 0.374 | 1.93 | 3000 | 0.2175 | 0.1894 |
| 0.3297 | 2.26 | 3500 | 0.2036 | 0.1755 |
| 0.3145 | 2.58 | 4000 | 0.1957 | 0.1757 |
| 0.3989 | 2.9 | 4500 | 0.1923 | 0.1723 |
| 0.271 | 3.22 | 5000 | 0.1889 | 0.1649 |
| 0.2758 | 3.55 | 5500 | 0.1768 | 0.1588 |
| 0.2683 | 3.87 | 6000 | 0.1720 | 0.1534 |
| 0.2341 | 4.19 | 6500 | 0.1689 | 0.1471 |
| 0.2316 | 4.51 | 7000 | 0.1706 | 0.1405 |
| 0.2383 | 4.84 | 7500 | 0.1637 | 0.1426 |
| 0.2148 | 5.16 | 8000 | 0.1584 | 0.1347 |
| 0.2085 | 5.48 | 8500 | 0.1601 | 0.1387 |
| 0.2944 | 5.8 | 9000 | 0.1566 | 0.1294 |
| 0.1944 | 6.13 | 9500 | 0.1494 | 0.1271 |
| 0.1853 | 6.45 | 10000 | 0.1561 | 0.1247 |
| 0.235 | 6.77 | 10500 | 0.1461 | 0.1215 |
| 0.2286 | 7.09 | 11000 | 0.1447 | 0.1167 |
| 0.1781 | 7.41 | 11500 | 0.1502 | 0.1199 |
| 0.1714 | 7.74 | 12000 | 0.1425 | 0.1179 |
| 0.1725 | 8.06 | 12500 | 0.1427 | 0.1173 |
| 0.143 | 8.38 | 13000 | 0.1448 | 0.1142 |
| 0.154 | 8.7 | 13500 | 0.1392 | 0.1104 |
| 0.1447 | 9.03 | 14000 | 0.1404 | 0.1094 |
| 0.1471 | 9.35 | 14500 | 0.1404 | 0.1088 |
| 0.1479 | 9.67 | 15000 | 0.1414 | 0.1133 |
| 0.1607 | 9.99 | 15500 | 0.1458 | 0.1171 |
| 0.166 | 10.32 | 16000 | 0.1652 | 0.1264 |
| 0.188 | 10.64 | 16500 | 0.1713 | 0.1322 |
| 0.1461 | 10.96 | 17000 | 0.1423 | 0.1111 |
| 0.1289 | 11.28 | 17500 | 0.1388 | 0.1097 |
| 0.1273 | 11.61 | 18000 | 0.1438 | 0.1074 |
| 0.1317 | 11.93 | 18500 | 0.1312 | 0.1066 |
| 0.1448 | 12.25 | 19000 | 0.1446 | 0.1042 |
| 0.1424 | 12.57 | 19500 | 0.1386 | 0.1015 |
| 0.1392 | 12.89 | 20000 | 0.1379 | 0.1005 |
| 0.1408 | 13.22 | 20500 | 0.1408 | 0.0992 |
| 0.1239 | 13.54 | 21000 | 0.1338 | 0.0968 |
| 0.1244 | 13.86 | 21500 | 0.1335 | 0.0957 |
| 0.1254 | 14.18 | 22000 | 0.1382 | 0.0950 |
| 0.1597 | 14.51 | 22500 | 0.1544 | 0.0970 |
| 0.1566 | 14.83 | 23000 | 0.1589 | 0.0963 |
### Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.2+cu113
- Datasets 1.18.3.dev0
- Tokenizers 0.11.0
|
NbAiLab/wav2vec2-xlsr-300M-NPSC-OH | df6db091dc686b2ebfd05c4ad39f4709fce32ec6 | 2022-02-02T06:10:42.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"transformers",
"NbAiLab/NPSC",
"generated_from_trainer",
"license:apache-2.0",
"model-index"
] | automatic-speech-recognition | false | NbAiLab | null | NbAiLab/wav2vec2-xlsr-300M-NPSC-OH | 1 | null | transformers | 28,192 | ---
license: apache-2.0
tags:
- automatic-speech-recognition
- NbAiLab/NPSC
- generated_from_trainer
model-index:
- name: wav2vec2-xlsr-300M-NPSC-OH
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# wav2vec2-xlsr-300M-NPSC-OH
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the NBAILAB/NPSC - 16K_MP3 dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1692
- Wer: 0.1663
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 7.5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 13
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2000
- num_epochs: 15.0
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:-----:|:---------------:|:------:|
| 3.1638 | 0.66 | 500 | 3.0686 | 1.0 |
| 2.9311 | 1.31 | 1000 | 2.9208 | 1.0 |
| 2.4175 | 1.97 | 1500 | 1.5009 | 0.9049 |
| 1.4442 | 2.63 | 2000 | 0.4426 | 0.3783 |
| 1.2624 | 3.28 | 2500 | 0.3193 | 0.2998 |
| 1.1889 | 3.94 | 3000 | 0.2867 | 0.2630 |
| 1.1315 | 4.6 | 3500 | 0.2566 | 0.2444 |
| 1.0864 | 5.26 | 4000 | 0.2368 | 0.2294 |
| 1.093 | 5.91 | 4500 | 0.2240 | 0.2151 |
| 1.0368 | 6.57 | 5000 | 0.2117 | 0.2056 |
| 1.0178 | 7.23 | 5500 | 0.2020 | 0.1954 |
| 1.0035 | 7.88 | 6000 | 0.2005 | 0.1924 |
| 0.9759 | 8.54 | 6500 | 0.1971 | 0.1863 |
| 0.9795 | 9.2 | 7000 | 0.1892 | 0.1812 |
| 0.9601 | 9.85 | 7500 | 0.1863 | 0.1795 |
| 0.9673 | 10.51 | 8000 | 0.1809 | 0.1761 |
| 0.9233 | 11.17 | 8500 | 0.1818 | 0.1755 |
| 0.9382 | 11.83 | 9000 | 0.1767 | 0.1741 |
| 0.9242 | 12.48 | 9500 | 0.1743 | 0.1703 |
| 0.9703 | 13.14 | 10000 | 0.1711 | 0.1711 |
| 0.9139 | 13.8 | 10500 | 0.1718 | 0.1672 |
| 0.9073 | 14.45 | 11000 | 0.1700 | 0.1665 |
### Framework versions
- Transformers 4.17.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.18.2.dev0
- Tokenizers 0.11.0
|
NbAiLab/wav2vec2-xlsr-300M-NPSC | 0cae29e61927933c2858160d35aaeb835b5fafd8 | 2022-01-31T14:31:10.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"transformers",
"license:cc"
] | automatic-speech-recognition | false | NbAiLab | null | NbAiLab/wav2vec2-xlsr-300M-NPSC | 1 | null | transformers | 28,193 | ---
license: cc
---
|
NbAiLab/wav2vec2-xlsr-300m-test | b61d0d339e5d2fad9cb274875e08196f508eac72 | 2022-01-29T16:47:01.000Z | [
"pytorch",
"wav2vec2",
"automatic-speech-recognition",
"transformers",
"license:cc"
] | automatic-speech-recognition | false | NbAiLab | null | NbAiLab/wav2vec2-xlsr-300m-test | 1 | null | transformers | 28,194 | ---
license: cc
---
|
NbAiLab/xls-npsc | ac1100cdaeb321693163e0eafdf5ecc5dcec61de | 2022-01-25T15:06:34.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"dataset:npsc",
"transformers",
"NbAiLab/NPSC",
"generated_from_trainer",
"license:apache-2.0",
"model-index"
] | automatic-speech-recognition | false | NbAiLab | null | NbAiLab/xls-npsc | 1 | null | transformers | 28,195 | ---
license: apache-2.0
tags:
- automatic-speech-recognition
- NbAiLab/NPSC
- generated_from_trainer
datasets:
- npsc
model-index:
- name: xls-npsc
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# xls-npsc
This model is a fine-tuned version of [facebook/wav2vec2-xls-r-300m](https://huggingface.co/facebook/wav2vec2-xls-r-300m) on the NBAILAB/NPSC - 48K_MP3 dataset.
It achieves the following results on the evaluation set:
- Loss: 3.5006
- Wer: 1.0
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 7.5e-05
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 10.0
- mixed_precision_training: Native AMP
### Training results
### Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.1+cu102
- Datasets 1.18.1.dev0
- Tokenizers 0.10.3
|
NbAiLab/xls-r-300m-sv2 | 27bec2b73dfecc43f8f53c28199bb2c2b89f98f1 | 2022-01-29T18:53:18.000Z | [
"pytorch",
"wav2vec2",
"automatic-speech-recognition",
"transformers"
] | automatic-speech-recognition | false | NbAiLab | null | NbAiLab/xls-r-300m-sv2 | 1 | null | transformers | 28,196 | Entry not found |
Nekoism/Zhongli-Beta | 25a8e25297af183536606ed032743112b4b672ae | 2021-10-26T01:17:12.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | Nekoism | null | Nekoism/Zhongli-Beta | 1 | null | transformers | 28,197 | ---
tags:
- conversational
---
#zhongli DialoGTP Model |
NickCavarretta/DialoGPT-small-laffy | 8fa09e843414615b9f823cf621a8442e036b77d7 | 2021-06-03T03:10:42.000Z | [
"pytorch",
"gpt2",
"text-generation",
"transformers",
"conversational"
] | conversational | false | NickCavarretta | null | NickCavarretta/DialoGPT-small-laffy | 1 | null | transformers | 28,198 | ---
tags:
- conversational
---
# My Awesome Laffy |
NicoGrageda/wav2vec2-base-timit-demo-colab | 9a9d452d9794a17634152f3909903b619633cbc3 | 2022-02-14T21:18:23.000Z | [
"pytorch",
"tensorboard",
"wav2vec2",
"automatic-speech-recognition",
"transformers",
"generated_from_trainer",
"license:apache-2.0",
"model-index"
] | automatic-speech-recognition | false | NicoGrageda | null | NicoGrageda/wav2vec2-base-timit-demo-colab | 1 | null | transformers | 28,199 | ---
license: apache-2.0
tags:
- generated_from_trainer
model-index:
- name: wav2vec2-base-timit-demo-colab
results: []
---
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# wav2vec2-base-timit-demo-colab
This model is a fine-tuned version of [facebook/wav2vec2-base](https://huggingface.co/facebook/wav2vec2-base) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.4519
- Wer: 0.3375
## Model description
More information needed
## Intended uses & limitations
More information needed
## Training and evaluation data
More information needed
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 32
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 1000
- num_epochs: 30
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 3.4351 | 4.0 | 500 | 1.2740 | 0.8259 |
| 0.5828 | 8.0 | 1000 | 0.4276 | 0.4403 |
| 0.2274 | 12.0 | 1500 | 0.4646 | 0.3739 |
| 0.135 | 16.0 | 2000 | 0.4320 | 0.3662 |
| 0.0962 | 20.0 | 2500 | 0.4831 | 0.3607 |
| 0.0719 | 24.0 | 3000 | 0.4506 | 0.3463 |
| 0.0556 | 28.0 | 3500 | 0.4519 | 0.3375 |
### Framework versions
- Transformers 4.11.3
- Pytorch 1.10.0+cu111
- Datasets 1.18.3
- Tokenizers 0.10.3
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.