Open-weight model · Question answering
distilbert-base-uncased-finetuned-natural-questions
by Arpit Gupta datarpit/distilbert-base-uncased-finetuned-natural-questions
This model is a fine-tuned version of distilbert-base-uncased on the naturalquestions dataset.
Parameters—
Context512
Weights265.5 MB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads2.1k
Model Card
By Arpit Gupta, published under apache-2.0, revision 28400e5824c2.
This model is a fine-tuned version of distilbert-base-uncased on the naturalquestions dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 64 - evalbatchsize: 64 - lrschedulertype: linear - numepochs: 40 - Transformers 4.17.0 - Pytorch 1.10.0 - Datasets 1.18.4 - Tokenizers 0.11.6
Read Arpit Gupta's full model card
This model is a fine-tuned version of distilbert-base-uncased on the natural_questions dataset. It achieves the following results on the evaluation set: - Loss: 0.6267
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: 64 - eval_batch_size: 64 - seed: 42 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: linear - num_epochs: 40
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 2.0532 | 1.0 | 5104 | 0.2393 |
| 1.8912 | 2.0 | 10208 | 0.2284 |
| 1.7854 | 3.0 | 15312 | 0.2357 |
| 1.6856 | 4.0 | 20416 | 0.2487 |
| 1.5918 | 5.0 | 25520 | 0.2743 |
| 1.5067 | 6.0 | 30624 | 0.2586 |
| 1.4323 | 7.0 | 35728 | 0.2763 |
| 1.365 | 8.0 | 40832 | 0.2753 |
| 1.3162 | 9.0 | 45936 | 0.3200 |
| 1.281 | 10.0 | 51040 | 0.3127 |
| 1.308 | 11.0 | 57104 | 0.2947 |
| 1.241 | 12.0 | 62208 | 0.2941 |
| 1.1391 | 13.0 | 67312 | 0.3103 |
| 1.0334 | 14.0 | 72416 | 0.3694 |
| 0.9538 | 15.0 | 77520 | 0.3658 |
| 0.8749 | 16.0 | 82624 | 0.4009 |
| 0.8154 | 17.0 | 87728 | 0.3672 |
| 0.7533 | 18.0 | 92832 | 0.3675 |
| 0.7079 | 19.0 | 97936 | 0.4611 |
| 0.6658 | 20.0 | 103040 | 0.4222 |
| 0.595 | 21.0 | 108144 | 0.4095 |
| 0.5765 | 22.0 | 113248 | 0.4400 |
| 0.5259 | 23.0 | 118352 | 0.5109 |
| 0.4804 | 24.0 | 123456 | 0.4711 |
| 0.4389 | 25.0 | 128560 | 0.5072 |
| 0.4034 | 26.0 | 133664 | 0.5363 |
| 0.374 | 27.0 | 138768 | 0.5460 |
| 0.3434 | 28.0 | 143872 | 0.5627 |
| 0.3181 | 29.0 | 148976 | 0.5657 |
| 0.2971 | 30.0 | 154080 | 0.5819 |
| 0.275 | 31.0 | 159184 | 0.5649 |
| 0.2564 | 32.0 | 164288 | 0.6087 |
| 0.2431 | 33.0 | 169392 | 0.6137 |
| 0.2289 | 34.0 | 174496 | 0.6123 |
| 0.2151 | 35.0 | 179600 | 0.5979 |
| 0.2041 | 36.0 | 184704 | 0.6196 |
| 0.1922 | 37.0 | 189808 | 0.6191 |
| 0.1852 | 38.0 | 194912 | 0.6313 |
| 0.1718 | 39.0 | 200016 | 0.6234 |
| 0.1718 | 39.81 | 204160 | 0.6267 |
Framework versions
- Transformers 4.17.0
- Pytorch 1.10.0
- Datasets 1.18.4
- Tokenizers 0.11.6
Configuration
- Architecture
- DistilBertForQuestionAnswering
- Context length (tokens)
- 512
- Vocabulary size
- 30,522
- Stored precision
- float32
- Model type
- distilbert
Identity and Version
- Repository
- datarpit/distilbert-base-uncased-finetuned-natural-questions
- Publisher
- Arpit Gupta
- Task
- Question answering
- Modality
- Text
- Library
- transformers
- Parameters
- Not stated by the source
- Languages
- Not stated by the source
- Revision
- 28400e5824c250ea3fac5f53da0fee11e03dfd4d
- First published
- 2022-03-08
- Last updated
- 2022-03-16
Files and Weights
10 files, 266.4 MB in total. The weights are 2 files totalling 265.5 MB in bin.
Weights2 files · 265.5 MB
Configuration2 files · 673 B
Tokenizer3 files · 943.2 KB
Documentation1 file · 3.4 KB
Repository2 files · 1.2 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| pytorch_model.bin | Weights | 265.5 MB | 59744bf02b60 |
| training_args.bin | Weights | 3.1 KB | b14ef0c0c627 |
| config.json | Configuration | 561 B | — |
| special_tokens_map.json | Configuration | 112 B | — |
| README.md | Documentation | 3.4 KB | — |
| .gitattributes | Repository | 1.2 KB | — |
| .gitignore | Repository | 13 B | — |
| tokenizer.json | Tokenizer | 711.4 KB | — |
| tokenizer_config.json | Tokenizer | 333 B | — |
| vocab.txt | Tokenizer | 231.5 KB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 265.5 MB
Download from Arpit Gupta
Released by Arpit Gupta through its official repository on Hugging Face. Read the license.
Built From
- Trained on (disclosed) natural_questions
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 265.5 MB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About distilbert-base-uncased-finetuned-natural-questions
Can I use distilbert-base-uncased-finetuned-natural-questions commercially?
Yes. distilbert-base-uncased-finetuned-natural-questions is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
What is distilbert-base-uncased-finetuned-natural-questions's context length?
512 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Splinter-base is the pretrained model discussed in the paper Few-Shot Question Answering by Pretraining Span Selection (at ACL 2021). Its original repository can be found here. The model is case-sensitive. Note: This model doesn't contain the pretrained weights for the QASS layer (see paper for details), and therefore the QASS layer is randomly initialized upon loading it. For the model with those weights, see tau/splinter-base-qass. Splinter is a model that is pretrained in a self-supervised fashion for few-shot question answering. 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…
This model can be used for the task of question answering. The model should not be used to intentionally create hostile or alienating environments for people. Significant research has explored bias and fairness issues with language models (see, e.g., Sheng et al. (2021) and Bender et al. (2021)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. The model creators note in the associated paper: The model creators note in the associated paper: The model…
This model is a fine-tune checkpoint of DistilBERT-base-cased, fine-tuned using (a second step of) knowledge distillation on SQuAD v1.1. This model reaches a F1 score of 87.1 on the dev set (for comparison, BERT bert-base-cased version reaches a F1 score of 88.7).
This model can be used for the task of Question Answering on Legal Documents. Read: An Open Source Contractual Language Understanding Application Using Machine Learning for detailed information on training procedure, dataset preprocessing and evaluation. See CUAD dataset card for more information. See CUAD dataset card for more information. Used V100/P100 from Google Colab Pro Python, Transformers Mohammed Rakib in collaboration with Ezi Ozoani and the Hugging Face team Use the code below to get started with the model.