deepset/roberta-base-squad2-distilled

1---2language: en3license: mit4tags:5- exbert6datasets:7- squad_v28thumbnail: https://thumb.tildacdn.com/tild3433-3637-4830-a533-353833613061/-/resize/720x/-/format/webp/germanquad.jpg9model-index:10- name: deepset/roberta-base-squad2-distilled11  results:12  - task:13      type: question-answering14      name: Question Answering15    dataset:16      name: squad_v217      type: squad_v218      config: squad_v219      split: validation20    metrics:21    - type: exact_match22      value: 80.859323      name: Exact Match24      verified: true25      verifyToken: eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJoYXNoIjoiMzVjNzkxNmNiNDkzNzdiYjJjZGM3ZTViMGJhOGM2ZjFmYjg1MjYxMDM2YzM5NWMwNDIyYzNlN2QwNGYyNDMzZSIsInZlcnNpb24iOjF9.Rgww8tf8D7nF2dh2U_DMrFzmp87k8s7RFibrDXSvQyA66PGWXwjlsd1552lzjHnNV5hvHUM1-h3PTuY_5p64BA26    - type: f127      value: 84.010428      name: F129      verified: true30      verifyToken: eyJhbGciOiJFZERTQSIsInR5cCI6IkpXVCJ9.eyJoYXNoIjoiNTAyZDViNWYzNjA4OWQ5MzgyYmQ2ZDlhNWRhMTIzYTYxYzViMmI4NWE4ZGU5MzVhZTAwNTRlZmRlNWUwMjI0ZSIsInZlcnNpb24iOjF9.Er21BNgJ3jJXLuZtpubTYq9wCwO1i_VLQFwS5ET0e4eAYVVj0aOA40I5FvP5pZac3LjkCnVacxzsFWGCYVmnDA31  - task:32      type: question-answering33      name: Question Answering34    dataset:35      name: squad36      type: squad37      config: plain_text38      split: validation39    metrics:40    - type: exact_match41      value: 86.22542      name: Exact Match43    - type: f144      value: 92.48345      name: F146  - task:47      type: question-answering48      name: Question Answering49    dataset:50      name: adversarial_qa51      type: adversarial_qa52      config: adversarialQA53      split: validation54    metrics:55    - type: exact_match56      value: 29.90057      name: Exact Match58    - type: f159      value: 41.18360      name: F161  - task:62      type: question-answering63      name: Question Answering64    dataset:65      name: squad_adversarial66      type: squad_adversarial67      config: AddOneSent68      split: validation69    metrics:70    - type: exact_match71      value: 79.07172      name: Exact Match73    - type: f174      value: 84.47275      name: F176  - task:77      type: question-answering78      name: Question Answering79    dataset:80      name: squadshifts amazon81      type: squadshifts82      config: amazon83      split: test84    metrics:85    - type: exact_match86      value: 70.73387      name: Exact Match88    - type: f189      value: 83.95890      name: F191  - task:92      type: question-answering93      name: Question Answering94    dataset:95      name: squadshifts new_wiki96      type: squadshifts97      config: new_wiki98      split: test99    metrics:100    - type: exact_match101      value: 82.011102      name: Exact Match103    - type: f1104      value: 91.092105      name: F1106  - task:107      type: question-answering108      name: Question Answering109    dataset:110      name: squadshifts nyt111      type: squadshifts112      config: nyt113      split: test114    metrics:115    - type: exact_match116      value: 84.203117      name: Exact Match118    - type: f1119      value: 91.521120      name: F1121  - task:122      type: question-answering123      name: Question Answering124    dataset:125      name: squadshifts reddit126      type: squadshifts127      config: reddit128      split: test129    metrics:130    - type: exact_match131      value: 72.029132      name: Exact Match133    - type: f1134      value: 83.454135      name: F1136---137 138# roberta-base distilled for Extractive QA 139 140## Overview141**Language model:** deepset/roberta-base-squad2-distilled   142**Language:** English  143**Training data:** SQuAD 2.0 training set   144**Eval data:** SQuAD 2.0 dev set   145**Code:**  See [an example extractive QA pipeline built with Haystack](https://haystack.deepset.ai/tutorials/34_extractive_qa_pipeline)  146**Infrastructure**: 4x V100 GPU  147**Published**: Dec 8th, 2021148 149## Details150- haystack's distillation feature was used for training. deepset/roberta-large-squad2 was used as the teacher model.151 152## Hyperparameters153```154batch_size = 80155n_epochs = 4156max_seq_len = 384157learning_rate = 3e-5158lr_schedule = LinearWarmup159embeds_dropout_prob = 0.1160temperature = 1.5161distillation_loss_weight = 0.75162```163 164## Usage165 166### In Haystack167Haystack is an AI orchestration framework to build customizable, production-ready LLM applications. You can use this model in Haystack to do extractive question answering on documents. 168To load and run the model with [Haystack](https://github.com/deepset-ai/haystack/):169```python170# After running pip install haystack-ai "transformers[torch,sentencepiece]"171 172from haystack import Document173from haystack.components.readers import ExtractiveReader174 175docs = [176    Document(content="Python is a popular programming language"),177    Document(content="python ist eine beliebte Programmiersprache"),178]179 180reader = ExtractiveReader(model="deepset/roberta-base-squad2-distilled")181reader.warm_up()182 183question = "What is a popular programming language?"184result = reader.run(query=question, documents=docs)185# {'answers': [ExtractedAnswer(query='What is a popular programming language?', score=0.5740374326705933, data='python', document=Document(id=..., content: '...'), context=None, document_offset=ExtractedAnswer.Span(start=0, end=6),...)]}186```187For a complete example with an extractive question answering pipeline that scales over many documents, check out the [corresponding Haystack tutorial](https://haystack.deepset.ai/tutorials/34_extractive_qa_pipeline).188 189### In Transformers190```python191from transformers import AutoModelForQuestionAnswering, AutoTokenizer, pipeline192 193model_name = "deepset/roberta-base-squad2-distilled"194 195# a) Get predictions196nlp = pipeline('question-answering', model=model_name, tokenizer=model_name)197QA_input = {198    'question': 'Why is model conversion important?',199    'context': 'The option to convert models between FARM and transformers gives freedom to the user and let people easily switch between frameworks.'200}201res = nlp(QA_input)202 203# b) Load model & tokenizer204model = AutoModelForQuestionAnswering.from_pretrained(model_name)205tokenizer = AutoTokenizer.from_pretrained(model_name)206```207 208## Performance209```210"exact": 79.8366040596311211"f1": 83.916407079888212```213 214## Authors215**Timo Möller:** timo.moeller@deepset.ai    216**Julian Risch:** julian.risch@deepset.ai    217**Malte Pietsch:** malte.pietsch@deepset.ai    218**Michel Bartels:** michel.bartels@deepset.ai    219 220## About us221 222<div class="grid lg:grid-cols-2 gap-x-4 gap-y-3">223    <div class="w-full h-40 object-cover mb-2 rounded-lg flex items-center justify-center">224         <img alt="" src="https://raw.githubusercontent.com/deepset-ai/.github/main/deepset-logo-colored.png" class="w-40"/>225     </div>226     <div class="w-full h-40 object-cover mb-2 rounded-lg flex items-center justify-center">227         <img alt="" src="https://raw.githubusercontent.com/deepset-ai/.github/main/haystack-logo-colored.png" class="w-40"/>228     </div>229</div>230 231[deepset](http://deepset.ai/) is the company behind the production-ready open-source AI framework [Haystack](https://haystack.deepset.ai/).232 233Some of our other work: 234- [Distilled roberta-base-squad2 (aka "tinyroberta-squad2")](https://huggingface.co/deepset/tinyroberta-squad2)235- [German BERT](https://deepset.ai/german-bert), [GermanQuAD and GermanDPR](https://deepset.ai/germanquad), [German embedding model](https://huggingface.co/mixedbread-ai/deepset-mxbai-embed-de-large-v1)236- [deepset Cloud](https://www.deepset.ai/deepset-cloud-product), [deepset Studio](https://www.deepset.ai/deepset-studio)237 238## Get in touch and join the Haystack community239 240<p>For more info on Haystack, visit our <strong><a href="https://github.com/deepset-ai/haystack">GitHub</a></strong> repo and <strong><a href="https://docs.haystack.deepset.ai">Documentation</a></strong>. 241 242We also have a <strong><a class="h-7" href="https://haystack.deepset.ai/community">Discord community open to everyone!</a></strong></p>243 244[Twitter](https://twitter.com/Haystack_AI) | [LinkedIn](https://www.linkedin.com/company/deepset-ai/) | [Discord](https://haystack.deepset.ai/community) | [GitHub Discussions](https://github.com/deepset-ai/haystack/discussions) | [Website](https://haystack.deepset.ai/) | [YouTube](https://www.youtube.com/@deepset_ai)245 246By the way: [we're hiring!](http://www.deepset.ai/jobs)