Search is not available for this dataset
pipeline_tag
stringclasses 48
values | library_name
stringclasses 205
values | text
stringlengths 0
18.3M
| metadata
stringlengths 2
1.07B
| id
stringlengths 5
122
| last_modified
null | tags
listlengths 1
1.84k
| sha
null | created_at
stringlengths 25
25
|
|---|---|---|---|---|---|---|---|---|
feature-extraction | transformers | {} | HyeonSang/kobert-sentiment | null | [
"transformers",
"tf",
"bert",
"feature-extraction",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
feature-extraction | transformers | {} | HyeonSang/test | null | [
"transformers",
"tf",
"bert",
"feature-extraction",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers |
#DwightSchrute DialoGPT-Model
#TheOffice | {"tags": ["conversational"]} | HypNyx/DialoGPT-small-DwightBot | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
#Thanos DialoGPT Model | {"tags": ["conversational"]} | HypNyx/DialoGPT-small-Thanos | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Peter from Your Boyfriend Game.
| {"tags": ["conversational"]} | HypedKid/PeterBot | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | Hyper2310/DialoGPT-small-harrypotter | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Hyunbin/PEGASUS_SKKU | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Hyunwoo/gee | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Hzaal/H | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | transformers | # Erlangshen-MegatronBert-1.3B
- Main Page:[Fengshenbang](https://fengshenbang-lm.com/)
- Github: [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)
## 简介 Brief Introduction
2021登顶FewCLUE和ZeroCLUE，处理NLU任务，开源时最大的中文BERT模型
It topped FewCLUE and ZeroCLUE benchmarks in 2021, solving NLU tasks, was the largest BERT when publicly released.
## 模型分类 Model Taxonomy
| 需求 Demand | 任务 Task | 系列 Series | 模型 Model | 参数 Parameter | 额外 Extra |
| :----: | :----: | :----: | :----: | :----: | :----: |
| 通用 General | 自然语言理解 NLU | 二郎神 Erlangshen | MegatronBERT | 1.3B | 中文 Chinese |
## 模型信息 Model Information
Encoder结构为主的双向语言模型，专注于解决各种自然语言理解任务。
我们跟进了[Megatron-LM](https://github.com/NVIDIA/Megatron-LM)的工作，使用了32张A100，总共耗时14天在悟道语料库（180 GB版本）上训练了十亿级别参数量的BERT。同时，鉴于中文语法和大规模训练的难度，我们使用四种预训练策略来改进BERT：1) 整词掩码, 2) 知识动态遮掩, 3) 句子顺序预测, 4) 层前归一化.
A bidirectional language model based on the Encoder structure, focusing on solving various NLU tasks.
We follow [Megatron-LM](https://github.com/NVIDIA/Megatron-LM), using 32 A100s and spending 14 days training a billion-level BERT on WuDao Corpora (180 GB version). Given Chinese grammar and the difficulty of large-scale training, we use four pre-training procedures to improve BERT: 1) Whole Word Masking (WWM), 2) Knowledge-based Dynamic Masking (KDM), 3) Sentence Order Prediction (SOP), 4) Pre-layer Normalization (Pre-LN).
### 成就 Achievements
1.2021年11月10日，Erlangshen-MegatronBert-1.3B在FewCLUE上取得第一。其中，它在CHIDF(成语填空)和TNEWS(新闻分类)子任务中的表现优于人类表现。此外，它在CHIDF(成语填空), CSLDCP(学科文献分类), OCNLI(自然语言推理)任务中均名列前茅。
2.2022年1月24日，Erlangshen-MegatronBert-1.3B在CLUE基准测试中的ZeroCLUE中取得第一。具体到子任务，我们在CSLDCP(主题文献分类), TNEWS(新闻分类), IFLYTEK(应用描述分类), CSL(抽象关键字识别)和CLUEWSC(参考消歧)任务中取得第一。
3.在2022年7月10日，Erlangshen-MegatronBert-1.3B在CLUE基准的语义匹配任务中取得第一。
1.On November 10, 2021, Erlangshen-MegatronBert-1.3B topped the FewCLUE benchmark. Among them, our Erlangshen outperformed human performance in CHIDF (idiom fill-in-the-blank) and TNEWS (news classification) subtasks. In addition, our Erlangshen ranked the top in CHIDF (idiom fill-in-the-blank), CSLDCP (subject literature classification), and OCNLI (natural language inference) tasks.
2.On January 24, 2022, Erlangshen-MegatronBert-1.3B topped the ZeroCLUE benchmark. For each of these tasks, we rank the top ones in CSLDCP (Subject Literature Classification), TNEWS (News Classification), IFLYTEK (Application Description Classification), CSL (Abstract Keyword Recognition), and CLUEWSC (Referential Disambiguation) tasks.
3.Erlangshen-MegatronBert-1.3B topped the CLUE benchmark semantic matching task on July 10, 2022.
### 下游效果 Performance
| 模型 | afqmc | tnews | iflytek | ocnli | cmnli | wsc | csl |
| :--------: | :-----: | :----: | :-----: | :----: | :----: | :----: | :----: |
| roberta-wwm-ext-large | 0.7514 | 0.5872 | 0.6152 | 0.777 | 0.814 | 0.8914 | 0.86 |
| Erlangshen-MegatronBert-1.3B | 0.7608 | 0.5996 | 0.6234 | 0.7917 | 0.81 | 0.9243 | 0.872 |
## 使用 Usage
``` python
from transformers import MegatronBertConfig, MegatronBertModel
from transformers import BertTokenizer
tokenizer = BertTokenizer.from_pretrained("IDEA-CCNL/Erlangshen-MegatronBert-1.3B")
config = MegatronBertConfig.from_pretrained("IDEA-CCNL/Erlangshen-MegatronBert-1.3B")
model = MegatronBertModel.from_pretrained("IDEA-CCNL/Erlangshen-MegatronBert-1.3B")
```
## 引用 Citation
如果您在您的工作中使用了我们的模型，可以引用我们的[论文](https://arxiv.org/abs/2209.02970)：
If you are using the resource for your work, please cite the our [paper](https://arxiv.org/abs/2209.02970):
```text
@article{fengshenbang,
author = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
title = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
journal = {CoRR},
volume = {abs/2209.02970},
year = {2022}
}
```
也可以引用我们的[网站](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
You can also cite our [website](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
```text
@misc{Fengshenbang-LM,
title={Fengshenbang-LM},
author={IDEA-CCNL},
year={2021},
howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}
``` | {"language": ["zh"], "license": "apache-2.0", "tags": ["bert", "NLU", "FewCLUE", "ZeroCLUE"], "inference": true} | IDEA-CCNL/Erlangshen-MegatronBert-1.3B | null | [
"transformers",
"pytorch",
"megatron-bert",
"bert",
"NLU",
"FewCLUE",
"ZeroCLUE",
"zh",
"arxiv:2209.02970",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text2text-generation | transformers | # Randeng-MegatronT5-770M
- Main Page:[Fengshenbang](https://fengshenbang-lm.com/)
- Github: [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)
## 简介 Brief Introduction
善于处理NLT任务，中文版的T5-large。
Good at solving NLT tasks, Chinese T5-large.
## 模型分类 Model Taxonomy
| 需求 Demand | 任务 Task | 系列 Series | 模型 Model | 参数 Parameter | 额外 Extra |
| :----: | :----: | :----: | :----: | :----: | :----: |
| 通用 General | 自然语言转换 NLT | 燃灯 Randeng | MegatronT5 | 770M | 中文-Chinese |
## 模型信息 Model Information
为了得到一个大规模的中文版的T5，我们使用了Megatron-LM的方法和悟道语料库(180G版本)用于预训练。具体地，我们在预训练阶段中使用了[Megatron-LM](https://github.com/NVIDIA/Megatron-LM) 大概花费了16张A100约14天。
To get a large-scale Chinese T5, we use of [Megatron-LM](https://github.com/NVIDIA/Megatron-LM) and WuDao Corpora (180 GB version) for pre-training. Specifically, in the pre-training phase which cost about 14 days with 16 A100 GPUs.
## 使用 Usage
因为[transformers](https://github.com/huggingface/transformers)库中是没有Randeng-MegatronT5-770M相关的模型结构的，所以你可以在我们的[Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)中找到并且运行代码。
Since there is no structure of Randeng-MegatronT5-770M in [transformers library](https://github.com/huggingface/transformers), you can find the structure of Randeng-MegatronT5-770M and run the codes in [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM).
```shell
git clone https://github.com/IDEA-CCNL/Fengshenbang-LM.git
```
### 加载模型 Loading Models
```python
from fengshen import T5ForConditionalGeneration
from fengshen import T5Config
from fengshen import T5Tokenizer
tokenizer = T5Tokenizer.from_pretrained('IDEA-CCNL/Randeng-MegatronT5-770M')
config = T5Config.from_pretrained('IDEA-CCNL/Randeng-MegatronT5-770M')
model = T5ForConditionalGeneration.from_pretrained('IDEA-CCNL/Randeng-MegatronT5-770M')
```
## 引用 Citation
如果您在您的工作中使用了我们的模型，可以引用我们的[论文](https://arxiv.org/abs/2209.02970)：
If you are using the resource for your work, please cite the our [paper](https://arxiv.org/abs/2209.02970):
```text
@article{fengshenbang,
author = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
title = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
journal = {CoRR},
volume = {abs/2209.02970},
year = {2022}
}
```
也可以引用我们的[网站](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
You can also cite our [website](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
```text
@misc{Fengshenbang-LM,
title={Fengshenbang-LM},
author={IDEA-CCNL},
year={2021},
howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}
```
| {"language": ["zh"], "license": "apache-2.0", "inference": false} | IDEA-CCNL/Randeng-MegatronT5-770M | null | [
"transformers",
"pytorch",
"t5",
"text2text-generation",
"zh",
"arxiv:2209.02970",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Wenzhong-GPT2-3.5B
- Main Page:[Fengshenbang](https://fengshenbang-lm.com/)
- Github: [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)
## 简介 Brief Introduction
善于处理NLG任务，目前最大的，中文版的GPT2
Focused on handling NLG tasks, the current largest, Chinese GPT2.
## 模型分类 Model Taxonomy
| 需求 Demand | 任务 Task | 系列 Series | 模型 Model | 参数 Parameter | 额外 Extra |
| :----: | :----: | :----: | :----: | :----: | :----: |
| 通用 General | 自然语言生成 NLG| 闻仲 Wenzhong | GPT2 | 3.5B | 中文 Chinese |
## 模型信息 Model Information
为了可以获得一个强大的单向语言模型，我们采用GPT模型结构，并且应用于中文语料上。具体地，这个模型拥有30层解码器和35亿参数，这比原本的GPT2-XL还要大。我们在100G的中文语料上预训练，这消耗了32个NVIDIA A100显卡大约28小时。据我们所知，它是目前最大的中文的GPT模型。
To obtain a robust unidirectional language model, we adopt the GPT model structure and apply it to the Chinese corpus. Specifically, this model has 30 decoder layers and 3.5 billion parameters, which is larger than the original GPT2-XL. We pre-train it on 100G of Chinese corpus, which consumes 32 NVIDIA A100 GPUs for about 28 hours. To the best of our knowledge, it is the largest Chinese GPT model currently available.
## 使用 Usage
### 加载模型 Loading Models
```python
from transformers import GPT2Tokenizer, GPT2Model
tokenizer = GPT2Tokenizer.from_pretrained('IDEA-CCNL/Wenzhong-GPT2-3.5B')
model = GPT2Model.from_pretrained('IDEA-CCNL/Wenzhong-GPT2-3.5B')
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### 使用示例 Usage Examples
```python
from transformers import pipeline, set_seed
set_seed(55)
generator = pipeline('text-generation', model='IDEA-CCNL/Wenzhong-GPT2-3.5B')
generator("北京位于", max_length=30, num_return_sequences=1)
```
## 引用 Citation
如果您在您的工作中使用了我们的模型，可以引用我们的[论文](https://arxiv.org/abs/2209.02970)：
If you are using the resource for your work, please cite the our [paper](https://arxiv.org/abs/2209.02970):
```text
@article{fengshenbang,
author = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
title = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
journal = {CoRR},
volume = {abs/2209.02970},
year = {2022}
}
```
也可以引用我们的[网站](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
You can also cite our [website](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
```text
@misc{Fengshenbang-LM,
title={Fengshenbang-LM},
author={IDEA-CCNL},
year={2021},
howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}
``` | {"language": ["zh"], "license": "apache-2.0", "inference": {"parameters": {"max_new_tokens": 128, "do_sample": true}}} | IDEA-CCNL/Wenzhong-GPT2-3.5B | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"zh",
"arxiv:2209.02970",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Yuyuan-GPT2-3.5B
- Main Page:[Fengshenbang](https://fengshenbang-lm.com/)
- Github: [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)
## 简介 Brief Introduction
目前最大的，医疗领域的生成语言模型GPT2。
The currently largest, generative language model GPT2 in the medical domain.
## 模型分类 Model Taxonomy
| 需求 Demand | 任务 Task | 系列 Series | 模型 Model | 参数 Parameter | 额外 Extra |
| :----: | :----: | :----: | :----: | :----: | :----: |
| 特殊 Special | 领域 Domain | 余元 Yuyuan | GPT2 | 3.5B | - |
## 模型信息 Model Information
我们采用与Wenzhong-GPT2-3.5B相同的架构，在50GB的医学(PubMed)语料库上进行预训练。我们使用了32个NVIDIA A100显卡大约7天。我们的Yuyuan-GPT2-3.5B是医疗领域最大的开源的GPT2模型。进一步地，模型可以通过计算困惑度（PPL）来判断事实。为了完成问答功能，我们将短语模式从疑问的形式转换为了陈述句。
We adopt the same architecture as Wenzhong-GPT2-3.5B to be pre-trained on 50 GB medical (PubMed) corpus. We use 32 NVIDIA A100 GPUs for about 7 days. Our Yuyuan-GPT2-3.5B is the largest open-source GPT2 model in the medical domain. We further allow the model to judge facts by computing perplexity (PPL). To accomplish question-and-answer functionality, we transform the phrase pattern from interrogative to declarative.
## 使用 Usage
### 加载模型 Loading Models
```python
from transformers import GPT2Tokenizer, GPT2Model
tokenizer = GPT2Tokenizer.from_pretrained('IDEA-CCNL/Yuyuan-GPT2-3.5B')
model = GPT2Model.from_pretrained('IDEA-CCNL/Yuyuan-GPT2-3.5B')
text = "Replace me by any text you'd like."
encoded_input = tokenizer(text, return_tensors='pt')
output = model(**encoded_input)
```
### 使用示例 Usage Examples
```python
from transformers import pipeline, set_seed
set_seed(55)
generator = pipeline('text-generation', model='IDEA-CCNL/Yuyuan-GPT2-3.5B')
generator("Diabetics should not eat", max_length=30, num_return_sequences=1)
```
## 引用 Citation
如果您在您的工作中使用了我们的模型，可以引用我们的[论文](https://arxiv.org/abs/2209.02970)：
If you are using the resource for your work, please cite the our [paper](https://arxiv.org/abs/2209.02970):
```text
@article{fengshenbang,
author = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
title = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
journal = {CoRR},
volume = {abs/2209.02970},
year = {2022}
}
```
也可以引用我们的[网站](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
You can also cite our [website](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
```text
@misc{Fengshenbang-LM,
title={Fengshenbang-LM},
author={IDEA-CCNL},
year={2021},
howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}
```
| {"language": ["en"], "license": "apache-2.0", "inference": false} | IDEA-CCNL/Yuyuan-GPT2-3.5B | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"en",
"arxiv:2209.02970",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | transformers | # Zhouwenwang-Unified-1.3B
- Main Page:[Fengshenbang](https://fengshenbang-lm.com/)
- Github: [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)
## 简介 Brief Introduction
与追一科技合作探索的中文统一模型，13亿参数的编码器结构模型。
The Chinese unified model explored in cooperation with Zhuiyi Technology, the encoder structure model with 1.3B parameters.
## 模型分类 Model Taxonomy
| 需求 Demand | 任务 Task | 系列 Series | 模型 Model | 参数 Parameter | 额外 Extra |
| :----: | :----: | :----: | :----: | :----: | :----: |
| 特殊 Special | 探索 Exploration | 周文王 Zhouwenwang | 待定 TBD | 1.3B | 中文 Chinese |
## 模型信息 Model Information
IDEA研究院认知计算中心联合追一科技有限公司提出的具有新结构的大模型。该模型在预训练阶段时考虑统一LM和MLM的任务，这让其同时具备生成和理解的能力，并且增加了旋转位置编码技术。目前已有13亿参数的Zhouwenwang-Unified-1.3B大模型，是中文领域中可以同时做LM和MLM任务的最大的模型。我们后续会持续在模型规模、知识融入、监督辅助任务等方向不断优化。
A large-scale model (Zhouwenwang-Unified-1.3B) with a new structure proposed by IDEA CCNL and Zhuiyi Technology. The model considers the task of unifying LM (Language Modeling) and MLM (Masked Language Modeling) during the pre-training phase, which gives it both generative and comprehension capabilities, and applys rotational position encoding. At present, Zhouwenwang-Unified-1.3B with 13B parameters is the largest Chinese model that can do both LM and MLM tasks. In the future, we will continue to optimize it in the direction of model size, knowledge incorporation, and supervisory assistance tasks.
### 下游任务 Performance
下游中文任务的得分（没有做任何数据增强）。
Scores on downstream chinese tasks (without any data augmentation)
| 模型 Model | afqmc | tnews | iflytek | ocnli | cmnli | wsc | csl |
| :--------: | :-----: | :----: | :-----: | :----: | :----: | :----: | :----: |
| roberta-wwm-ext-large | 0.7514 | 0.5872 | 0.6152 | 0.7770 | 0.8140 | 0.8914 | 0.8600 |
| Zhouwenwang-Unified-1.3B | 0.7463 | 0.6036 | 0.6288 | 0.7654 | 0.7741 | 0.8849 | 0. 8777 |
## 使用 Usage
因为[transformers](https://github.com/huggingface/transformers)库中是没有 Zhouwenwang-Unified-1.3B相关的模型结构的，所以你可以在我们的[Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)中找到并且运行代码。
Since there is no structure of Zhouwenwang-Unified-1.3B in [transformers library](https://github.com/huggingface/transformers), you can find the structure of Zhouwenwang-Unified-1.3B and run the codes in [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM).
```shell
git clone https://github.com/IDEA-CCNL/Fengshenbang-LM.git
```
### 加载模型 Loading Models
```python
from fengshen import RoFormerModel
from fengshen import RoFormerConfig
from transformers import BertTokenizer
tokenizer = BertTokenizer.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-1.3B")
config = RoFormerConfig.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-1.3B")
model = RoFormerModel.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-1.3B")
```
### 使用示例 Usage Examples
你可以使用该模型进行续写任务。
You can use the model for continuation writing tasks.
```python
from fengshen import RoFormerModel
from transformers import AutoTokenizer
import torch
import numpy as np
sentence = '清华大学位于'
max_length = 32
tokenizer = AutoTokenizer.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-1.3B")
model = RoFormerModel.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-1.3B")
for i in range(max_length):
encode = torch.tensor(
[[tokenizer.cls_token_id]+tokenizer.encode(sentence, add_special_tokens=False)]).long()
logits = model(encode)[0]
logits = torch.nn.functional.linear(
logits, model.embeddings.word_embeddings.weight)
logits = torch.nn.functional.softmax(
logits, dim=-1).cpu().detach().numpy()[0]
sentence = sentence + \
tokenizer.decode(int(np.random.choice(logits.shape[1], p=logits[-1])))
if sentence[-1] == '。':
break
print(sentence)
```
## 引用 Citation
如果您在您的工作中使用了我们的模型，可以引用我们的[论文](https://arxiv.org/abs/2209.02970)：
If you are using the resource for your work, please cite the our [paper](https://arxiv.org/abs/2209.02970):
```text
@article{fengshenbang,
author = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
title = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
journal = {CoRR},
volume = {abs/2209.02970},
year = {2022}
}
```
也可以引用我们的[网站](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
You can also cite our [website](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
```text
@misc{Fengshenbang-LM,
title={Fengshenbang-LM},
author={IDEA-CCNL},
year={2021},
howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}
``` | {"language": ["zh"], "license": "apache-2.0", "widget": [{"text": "\u751f\u6d3b\u7684\u771f\u8c1b\u662f[MASK]\u3002"}]} | IDEA-CCNL/Zhouwenwang-Unified-1.3B | null | [
"transformers",
"pytorch",
"megatron-bert",
"zh",
"arxiv:2209.02970",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | transformers |
# Zhouwenwang-Unified-110M
- Main Page:[Fengshenbang](https://fengshenbang-lm.com/)
- Github: [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)
## 简介 Brief Introduction
与追一科技合作探索的中文统一模型，1.1亿参数的编码器结构模型。
The Chinese unified model explored in cooperation with Zhuiyi Technology, the encoder structure model with 110M parameters.
## 模型分类 Model Taxonomy
| 需求 Demand | 任务 Task | 系列 Series | 模型 Model | 参数 Parameter | 额外 Extra |
| :----: | :----: | :----: | :----: | :----: | :----: |
| 特殊 Special | 探索 Exploration | 周文王 Zhouwenwang | 待定 TBD | 110M | 中文 Chinese |
## 模型信息 Model Information
IDEA研究院认知计算中心联合追一科技有限公司提出的具有新结构的大模型。该模型在预训练阶段时考虑统一LM和MLM的任务，这让其同时具备生成和理解的能力，并且增加了旋转位置编码技术。我们后续会持续在模型规模、知识融入、监督辅助任务等方向不断优化。
A large-scale model (Zhouwenwang-Unified-1.3B) with a new structure proposed by IDEA CCNL and Zhuiyi Technology. The model considers the task of unifying LM (Language Modeling) and MLM (Masked Language Modeling) during the pre-training phase, which gives it both generative and comprehension capabilities, and applys rotational position encoding. In the future, we will continue to optimize it in the direction of model size, knowledge incorporation, and supervisory assistance tasks.
## 使用 Usage
因为[transformers](https://github.com/huggingface/transformers)库中是没有 Zhouwenwang-Unified-110M相关的模型结构的，所以你可以在我们的[Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM)中找到并且运行代码。
Since there is no structure of Zhouwenwang-Unified-110M in [transformers library](https://github.com/huggingface/transformers), you can find the structure of Zhouwenwang-Unified-110M and run the codes in [Fengshenbang-LM](https://github.com/IDEA-CCNL/Fengshenbang-LM).
```shell
git clone https://github.com/IDEA-CCNL/Fengshenbang-LM.git
```
### 加载模型 Loading Models
```python
from fengshen import RoFormerModel
from fengshen import RoFormerConfig
from transformers import BertTokenizer
tokenizer = BertTokenizer.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-110M")
config = RoFormerConfig.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-110M")
model = RoFormerModel.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-110M")
```
### 使用示例 Usage Examples
你可以使用该模型进行续写任务。
You can use the model for continuation writing tasks.
```python
from fengshen import RoFormerModel
from transformers import AutoTokenizer
import torch
import numpy as np
sentence = '清华大学位于'
max_length = 32
tokenizer = AutoTokenizer.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-110M")
model = RoFormerModel.from_pretrained("IDEA-CCNL/Zhouwenwang-Unified-110M")
for i in range(max_length):
encode = torch.tensor(
[[tokenizer.cls_token_id]+tokenizer.encode(sentence, add_special_tokens=False)]).long()
logits = model(encode)[0]
logits = torch.nn.functional.linear(
logits, model.embeddings.word_embeddings.weight)
logits = torch.nn.functional.softmax(
logits, dim=-1).cpu().detach().numpy()[0]
sentence = sentence + \
tokenizer.decode(int(np.random.choice(logits.shape[1], p=logits[-1])))
if sentence[-1] == '。':
break
print(sentence)
```
## 引用 Citation
如果您在您的工作中使用了我们的模型，可以引用我们的[论文](https://arxiv.org/abs/2209.02970)：
If you are using the resource for your work, please cite the our [paper](https://arxiv.org/abs/2209.02970):
```text
@article{fengshenbang,
author = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
title = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
journal = {CoRR},
volume = {abs/2209.02970},
year = {2022}
}
```
也可以引用我们的[网站](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
You can also cite our [website](https://github.com/IDEA-CCNL/Fengshenbang-LM/):
```text
@misc{Fengshenbang-LM,
title={Fengshenbang-LM},
author={IDEA-CCNL},
year={2021},
howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}
```
| {"language": ["zh"], "license": "apache-2.0", "widget": [{"text": "\u751f\u6d3b\u7684\u771f\u8c1b\u662f[MASK]\u3002"}]} | IDEA-CCNL/Zhouwenwang-Unified-110M | null | [
"transformers",
"pytorch",
"megatron-bert",
"zh",
"arxiv:2209.02970",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Rick And Morty DialoGPT Model | {"tags": ["conversational"]} | ILoveThatLady/DialoGPT-small-rickandmorty | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers | #Slovak RoBERTA Masked Language Model
###83Mil Parameters in small model
Medium and Large models coming soon!
RoBERTA pretrained tokenizer vocab and merges included.
---
##Training params:
- **Dataset**:
8GB Slovak Monolingual dataset including ParaCrawl (monolingual), OSCAR, and several gigs of my own findings and cleaning.
- **Preprocessing**:
Tokenized with a pretrained ByteLevelBPETokenizer trained on the same dataset. Uncased, with s, pad, /s, unk, and mask special tokens.
- **Evaluation results**:
- Mnoho ľudí tu MASK
- žije.
- žijú.
- je.
- trpí.
- Ako sa MASK
- máte
- máš
- má
- hovorí
- Plážová sezóna pod Zoborom patrí medzi MASK obdobia.
- ročné
- najkrajšie
- najobľúbenejšie
- najnáročnejšie
- **Limitations**:
The current model is fairly small, although it works very well. This model is meant to be finetuned on downstream tasks e.g. Part-of-Speech tagging, Question Answering, anything in GLUE or SUPERGLUE.
- **Credit**:
If you use this or any of my models in research or professional work, please credit me - Christopher Brousseau in said work. | {} | IMJONEZZ/SlovenBERTcina | null | [
"transformers",
"pytorch",
"safetensors",
"roberta",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# Hate Speech Classifier for Social Media Content in English Language
A monolingual model for hate speech classification of social media content in English language. The model was trained on 103190 YouTube comments and tested on an independent test set of 20554 YouTube comments. It is based on English BERT base pre-trained language model.
## Please cite:
Kralj Novak, P., Scantamburlo, T., Pelicon, A., Cinelli, M., Mozetič, I., & Zollo, F. (2022, July). __Handling disagreement in hate speech modelling__. In International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems (pp. 681-695). Cham: Springer International Publishing.
https://link.springer.com/chapter/10.1007/978-3-031-08974-9_54
## Tokenizer
During training the text was preprocessed using the original English BERT base tokenizer. We suggest the same tokenizer is used for inference.
## Model output
The model classifies each input into one of four distinct classes:
* 0 - acceptable
* 1 - inappropriate
* 2 - offensive
* 3 - violent
Details on data acquisition and labeling including the Annotation guidelines:
http://imsypp.ijs.si/wp-content/uploads/2021/12/IMSyPP_D2.2_multilingual-dataset.pdf
| {"language": ["en"], "license": "mit", "widget": [{"text": "My name is Mark and I live in London. I am a postgraduate student at Queen Mary University."}]} | IMSyPP/hate_speech_en | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"en",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# Hate Speech Classifier for Social Media Content in Italian Language
A monolingual model for hate speech classification of social media content in Italian language. The model was trained on 119,670 YouTube comments and tested on an independent test set of 21,072 YouTube comments. It is based on Italian ALBERTO pre-trained language model.
## Please cite:
Kralj Novak, P., Scantamburlo, T., Pelicon, A., Cinelli, M., Mozetič, I., & Zollo, F. (2022, July). __Handling disagreement in hate speech modelling__. In International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems (pp. 681-695). Cham: Springer International Publishing.
https://link.springer.com/chapter/10.1007/978-3-031-08974-9_54
## Tokenizer
During training the text was preprocessed using the original Italian ALBERTO tokenizer. We suggest the same tokenizer is used for inference.
## Model output
The model classifies each input into one of four distinct classes:
* 0 - acceptable
* 1 - inappropriate
* 2 - offensive
* 3 - violent | {"language": ["it"], "license": "mit", "widget": [{"text": "Ciao, mi chiamo Marcantonio, sono di Roma. Studio informatica all'Universit\u00e0 di Roma."}]} | IMSyPP/hate_speech_it | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"it",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# Hate Speech Classifier for Social Media Content in Dutch
A monolingual model for hate speech classification of social media content in Dutch. The model was trained on 20000 social media posts (youtube, twitter, facebook) and tested on an independent test set of 2000 posts. It is based on thepre-trained language model [BERTje](https://huggingface.co/wietsedv/bert-base-dutch-cased).
## Tokenizer
During training the text was preprocessed using the BERTje tokenizer. We suggest the same tokenizer is used for inference.
## Model output
The model classifies each input into one of four distinct classes:
* 0 - acceptable
* 1 - inappropriate
* 2 - offensive
* 3 - violent | {"language": ["nl"], "license": "mit"} | IMSyPP/hate_speech_nl | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"nl",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# Hate Speech Classifier for Social Media Content in Slovenian Language
A monolingual model for hate speech classification of social media content in Slovenian language. The model was trained on 50,000 Twitter comments and tested on an independent test set of 10,000 Twitter comments. It is based on multilingual CroSloEngual BERT pre-trained language model.
## Please cite:
Kralj Novak, P., Scantamburlo, T., Pelicon, A., Cinelli, M., Mozetič, I., & Zollo, F. (2022, July). __Handling disagreement in hate speech modelling__. In International Conference on Information Processing and Management of Uncertainty in Knowledge-Based Systems (pp. 681-695). Cham: Springer International Publishing.
https://link.springer.com/chapter/10.1007/978-3-031-08974-9_54
## Tokenizer
During training the text was preprocessed using the original CroSloEngual BERT tokenizer. We suggest the same tokenizer is used for inference.
## Model output
The model classifies each input into one of four distinct classes:
* 0 - acceptable
* 1 - inappropriate
* 2 - offensive
* 3 - violent | {"language": ["sl"], "license": "mit", "pipeline_tag": "text-classification", "inference": true, "widget": [{"text": "Sem Mark in \u017eivim v Ljubljani. Sem doktorski \u0161tudent na Mednarodni podiplomski \u0161oli Jo\u017eefa Stefana."}]} | IMSyPP/hate_speech_slo | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"sl",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers | {"language": ["sl"], "license": "mit"} | IMSyPP/hate_speech_targets_slo | null | [
"transformers",
"pytorch",
"camembert",
"text-classification",
"sl",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers |
# Cyber Bones DialoGPT Model | {"tags": ["conversational"]} | ITNODove/DialoGPT-medium-cyberbones | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | ITworkonline/twitter_tac_model | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers |
<!-- 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. -->
# output
This model is a fine-tuned version of [gpt2](https://huggingface.co/gpt2) on a dataset of Shakespeare's plays.
## Model description
The model is the original gpt-2 model fine-tuned on a custom dataset.
## Intended uses & limitations
The model can be used to generate Shakespearean-like text. Consider that because it comes from plays, such a typographical structure might be reproduced.
## Training and evaluation data
Trained with Shakespeare's plays corpus.
## Training procedure
### Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 1
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 2.0
### Training results
### Framework versions
- Transformers 4.16.0.dev0
- Pytorch 1.10.0+cu111
- Datasets 1.18.0
- Tokenizers 0.11.0
| {"license": "mit", "tags": ["generated_from_trainer"], "model-index": [{"name": "output", "results": []}]} | Iacopo/Shakespear-GPT2 | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"generated_from_trainer",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | IanFaiir/Ian | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | IceeSoHighYetSoLow19/Luna | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers |
# Hank Hill DialoGPT Model | {"tags": ["conversational"]} | Icemiser/chat-test | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | Icezorg/test_predict | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Id405/Adam | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Idrisa/Njema | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text2text-generation | transformers | @inproceedings{adebara-abdul-mageed-2021-improving,
title = "Improving Similar Language Translation With Transfer Learning",
author = "Adebara, Ife and
Abdul-Mageed, Muhammad",
booktitle = "Proceedings of the Sixth Conference on Machine Translation",
month = nov,
year = "2021",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2021.wmt-1.27",
pages = "273--278",
abstract = "We investigate transfer learning based on pre-trained neural machine translation models to translate between (low-resource) similar languages. This work is part of our contribution to the WMT 2021 Similar Languages Translation Shared Task where we submitted models for different language pairs, including French-Bambara, Spanish-Catalan, and Spanish-Portuguese in both directions. Our models for Catalan-Spanish (82.79 BLEU)and Portuguese-Spanish (87.11 BLEU) rank top 1 in the official shared task evaluation, and we are the only team to submit models for the French-Bambara pairs.",
} | {"language": ["bm", "fr"]} | Ife/BM-FR | null | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"bm",
"fr",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text2text-generation | transformers | # Similar-Languages-MT | {} | Ife/CA-ES | null | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text2text-generation | transformers | {} | Ife/ES-CA | null | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text2text-generation | transformers | {} | Ife/ES-PT | null | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text2text-generation | transformers | {} | Ife/FR-BM | null | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text2text-generation | transformers | {} | Ife/PT-ES | null | [
"transformers",
"pytorch",
"marian",
"text2text-generation",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
question-answering | transformers | A distilbert model fine-tuned for question answering. | {"language": ["en"], "datasets": ["squad_v2", "wiki_qa"], "metrics": ["accuracy"], "pipeline_tag": "question-answering"} | Ifenna/dbert-3epoch | null | [
"transformers",
"pytorch",
"safetensors",
"distilbert",
"question-answering",
"en",
"dataset:squad_v2",
"dataset:wiki_qa",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
Забавное для дискордика))00)) https://discord.gg/HpeadKH
Offers
[email protected] | {"tags": ["ru", "4ulan"]} | Ifromspace/GRIEFSOFT-walr | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"ru",
"4ulan",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
**Fork of https://huggingface.co/sberbank-ai/rugpt3large_based_on_gpt2**
Забавное для дискордика))00))
ROADMAP:
- Собираю датасетик из книжек про попаданцев. <------------------------- Сейчас тут.
- Дообучаю.
- Выбрасываю в дискордик.
https://discord.gg/HpeadKH | {"language": ["ru"], "tags": ["PyTorch", "Transformers", "4ulan"]} | Ifromspace/GRIEFSOFT | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"PyTorch",
"Transformers",
"4ulan",
"ru",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | Ignacio/distilbert-base-uncased-finetuned-cola | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Iim/Leha | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Ils/Mk | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Ilya-sig-n/UrFU_neural_network | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | transformers | {} | IlyaGusev/gen_title_tg_bottleneck | null | [
"transformers",
"pytorch",
"encoder-decoder",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
feature-extraction | transformers | {} | IlyaGusev/gen_title_tg_bottleneck_encoder | null | [
"transformers",
"pytorch",
"jax",
"bert",
"feature-extraction",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
summarization | transformers |
# MBARTRuSumGazeta
## Model description
This is a ported version of [fairseq model](https://www.dropbox.com/s/fijtntnifbt9h0k/gazeta_mbart_v2_fairseq.tar.gz).
For more details, please see [Dataset for Automatic Summarization of Russian News](https://arxiv.org/abs/2006.11063).
## Intended uses & limitations
#### How to use
Colab: [link](https://colab.research.google.com/drive/1wdo_nPZPk6dWAn1J8nGx4Z5Ef82jCCob)
```python
from transformers import MBartTokenizer, MBartForConditionalGeneration
model_name = "IlyaGusev/mbart_ru_sum_gazeta"
tokenizer = MBartTokenizer.from_pretrained(model_name)
model = MBartForConditionalGeneration.from_pretrained(model_name)
article_text = "..."
input_ids = tokenizer(
[article_text],
max_length=600,
padding="max_length",
truncation=True,
return_tensors="pt",
)["input_ids"]
output_ids = model.generate(
input_ids=input_ids,
no_repeat_ngram_size=4
)[0]
summary = tokenizer.decode(output_ids, skip_special_tokens=True)
print(summary)
```
#### Limitations and bias
- The model should work well with Gazeta.ru articles, but for any other agencies it can suffer from domain shift
## Training data
- Dataset: [Gazeta](https://huggingface.co/datasets/IlyaGusev/gazeta)
## Training procedure
- Fairseq training script: [train.sh](https://github.com/IlyaGusev/summarus/blob/master/external/bart_scripts/train.sh)
- Porting: [Colab link](https://colab.research.google.com/drive/13jXOlCpArV-lm4jZQ0VgOpj6nFBYrLAr)
## Eval results
* Train dataset: **Gazeta v1 train**
* Test dataset: **Gazeta v1 test**
* Source max_length: **600**
* Target max_length: **200**
* no_repeat_ngram_size: **4**
* num_beams: **5**
| Model | R-1-f | R-2-f | R-L-f | chrF | METEOR | BLEU | Avg char length |
|:--------------------------|:------|:------|:------|:-------|:-------|:-----|:-----|
| [mbart_ru_sum_gazeta](https://huggingface.co/IlyaGusev/mbart_ru_sum_gazeta) | **32.4** | 14.3 | 28.0 | 39.7 | **26.4** | 12.1 | 371 |
| [rut5_base_sum_gazeta](https://huggingface.co/IlyaGusev/rut5_base_sum_gazeta) | 32.2 | **14.4** | **28.1** | **39.8** | 25.7 | **12.3** | 330 |
| [rugpt3medium_sum_gazeta](https://huggingface.co/IlyaGusev/rugpt3medium_sum_gazeta) | 26.2 | 7.7 | 21.7 | 33.8 | 18.2 | 4.3 | 244 |
* Train dataset: **Gazeta v1 train**
* Test dataset: **Gazeta v2 test**
* Source max_length: **600**
* Target max_length: **200**
* no_repeat_ngram_size: **4**
* num_beams: **5**
| Model | R-1-f | R-2-f | R-L-f | chrF | METEOR | BLEU | Avg char length |
|:--------------------------|:------|:------|:------|:-------|:-------|:-----|:-----|
| [mbart_ru_sum_gazeta](https://huggingface.co/IlyaGusev/mbart_ru_sum_gazeta) | **28.7** | **11.1** | 24.4 | **37.3** | **22.7** | **9.4** | 373 |
| [rut5_base_sum_gazeta](https://huggingface.co/IlyaGusev/rut5_base_sum_gazeta) | 28.6 | **11.1** | **24.5** | 37.2 | 22.0 | **9.4** | 331 |
| [rugpt3medium_sum_gazeta](https://huggingface.co/IlyaGusev/rugpt3medium_sum_gazeta) | 24.1 | 6.5 | 19.8 | 32.1 | 16.3 | 3.6 | 242 |
Predicting all summaries:
```python
import json
import torch
from transformers import MBartTokenizer, MBartForConditionalGeneration
from datasets import load_dataset
def gen_batch(inputs, batch_size):
batch_start = 0
while batch_start < len(inputs):
yield inputs[batch_start: batch_start + batch_size]
batch_start += batch_size
def predict(
model_name,
input_records,
output_file,
max_source_tokens_count=600,
batch_size=4
):
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = MBartTokenizer.from_pretrained(model_name)
model = MBartForConditionalGeneration.from_pretrained(model_name).to(device)
predictions = []
for batch in gen_batch(inputs, batch_size):
texts = [r["text"] for r in batch]
input_ids = tokenizer(
batch,
return_tensors="pt",
padding="max_length",
truncation=True,
max_length=max_source_tokens_count
)["input_ids"].to(device)
output_ids = model.generate(
input_ids=input_ids,
no_repeat_ngram_size=4
)
summaries = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
for s in summaries:
print(s)
predictions.extend(summaries)
with open(output_file, "w") as w:
for p in predictions:
w.write(p.strip().replace("\n", " ") + "\n")
gazeta_test = load_dataset('IlyaGusev/gazeta', script_version="v1.0")["test"]
predict("IlyaGusev/mbart_ru_sum_gazeta", list(gazeta_test), "mbart_predictions.txt")
```
Evaluation: https://github.com/IlyaGusev/summarus/blob/master/evaluate.py
Flags: --language ru --tokenize-after --lower
### BibTeX entry and citation info
```bibtex
@InProceedings{10.1007/978-3-030-59082-6_9,
author="Gusev, Ilya",
editor="Filchenkov, Andrey and Kauttonen, Janne and Pivovarova, Lidia",
title="Dataset for Automatic Summarization of Russian News",
booktitle="Artificial Intelligence and Natural Language",
year="2020",
publisher="Springer International Publishing",
address="Cham",
pages="122--134",
isbn="978-3-030-59082-6"
}
```
| {"language": ["ru"], "license": "apache-2.0", "tags": ["summarization", "mbart"], "datasets": ["IlyaGusev/gazeta"], "inference": {"parameters": {"no_repeat_ngram_size": 4}}, "widget": [{"text": "\u0412\u044b\u0441\u043e\u0442\u0430 \u0431\u0430\u0448\u043d\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 324 \u043c\u0435\u0442\u0440\u0430 (1063 \u0444\u0443\u0442\u0430), \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e \u0442\u0430\u043a\u0430\u044f \u0436\u0435 \u0432\u044b\u0441\u043e\u0442\u0430, \u043a\u0430\u043a \u0443 81-\u044d\u0442\u0430\u0436\u043d\u043e\u0433\u043e \u0437\u0434\u0430\u043d\u0438\u044f, \u0438 \u0441\u0430\u043c\u043e\u0435 \u0432\u044b\u0441\u043e\u043a\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u0432 \u041f\u0430\u0440\u0438\u0436\u0435. \u0415\u0433\u043e \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u0438\u0435 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u043e, \u0440\u0430\u0437\u043c\u0435\u0440\u043e\u043c 125 \u043c\u0435\u0442\u0440\u043e\u0432 (410 \u0444\u0443\u0442\u043e\u0432) \u0441 \u043b\u044e\u0431\u043e\u0439 \u0441\u0442\u043e\u0440\u043e\u043d\u044b. \u0412\u043e \u0432\u0440\u0435\u043c\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u043f\u0440\u0435\u0432\u0437\u043e\u0448\u043b\u0430 \u043c\u043e\u043d\u0443\u043c\u0435\u043d\u0442 \u0412\u0430\u0448\u0438\u043d\u0433\u0442\u043e\u043d\u0430, \u0441\u0442\u0430\u0432 \u0441\u0430\u043c\u044b\u043c \u0432\u044b\u0441\u043e\u043a\u0438\u043c \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u043c \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435\u043c \u0432 \u043c\u0438\u0440\u0435, \u0438 \u044d\u0442\u043e\u0442 \u0442\u0438\u0442\u0443\u043b \u043e\u043d\u0430 \u0443\u0434\u0435\u0440\u0436\u0438\u0432\u0430\u043b\u0430 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 41 \u0433\u043e\u0434\u0430 \u0434\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u043e \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u0432 \u041d\u044c\u044e-\u0419\u043e\u0440\u043a\u0435 \u0432 1930 \u0433\u043e\u0434\u0443. \u042d\u0442\u043e \u043f\u0435\u0440\u0432\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0434\u043e\u0441\u0442\u0438\u0433\u043b\u043e \u0432\u044b\u0441\u043e\u0442\u044b 300 \u043c\u0435\u0442\u0440\u043e\u0432. \u0418\u0437-\u0437\u0430 \u0434\u043e\u0431\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0432\u0435\u0449\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0430\u043d\u0442\u0435\u043d\u043d\u044b \u043d\u0430 \u0432\u0435\u0440\u0448\u0438\u043d\u0435 \u0431\u0430\u0448\u043d\u0438 \u0432 1957 \u0433\u043e\u0434\u0443 \u043e\u043d\u0430 \u0441\u0435\u0439\u0447\u0430\u0441 \u0432\u044b\u0448\u0435 \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u043d\u0430 5,2 \u043c\u0435\u0442\u0440\u0430 (17 \u0444\u0443\u0442\u043e\u0432). \u0417\u0430 \u0438\u0441\u043a\u043b\u044e\u0447\u0435\u043d\u0438\u0435\u043c \u043f\u0435\u0440\u0435\u0434\u0430\u0442\u0447\u0438\u043a\u043e\u0432, \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0432\u0442\u043e\u0440\u043e\u0439 \u0441\u0430\u043c\u043e\u0439 \u0432\u044b\u0441\u043e\u043a\u043e\u0439 \u043e\u0442\u0434\u0435\u043b\u044c\u043d\u043e \u0441\u0442\u043e\u044f\u0449\u0435\u0439 \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u043e\u0439 \u0432\u043e \u0424\u0440\u0430\u043d\u0446\u0438\u0438 \u043f\u043e\u0441\u043b\u0435 \u0432\u0438\u0430\u0434\u0443\u043a\u0430 \u041c\u0438\u0439\u043e.", "example_title": "\u0412\u0438\u043a\u0438\u043f\u0435\u0434\u0438\u044f"}, {"text": "\u0421 1 \u0441\u0435\u043d\u0442\u044f\u0431\u0440\u044f \u0432 \u0420\u043e\u0441\u0441\u0438\u0438 \u0432\u0441\u0442\u0443\u043f\u0430\u044e\u0442 \u0432 \u0441\u0438\u043b\u0443 \u043f\u043e\u043f\u0440\u0430\u0432\u043a\u0438 \u0432 \u0437\u0430\u043a\u043e\u043d \u00ab\u041e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u0435\u00bb \u2014 \u0442\u0435\u043f\u0435\u0440\u044c \u0434\u043e\u043b\u0436\u043d\u0438\u043a\u0438 \u0441\u043c\u043e\u0433\u0443\u0442 \u043e\u0441\u0432\u043e\u0431\u043e\u0436\u0434\u0430\u0442\u044c\u0441\u044f \u043e\u0442 \u043d\u0435\u043f\u043e\u0441\u0438\u043b\u044c\u043d\u044b\u0445 \u043e\u0431\u044f\u0437\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432 \u0432\u043e \u0432\u043d\u0435\u0441\u0443\u0434\u0435\u0431\u043d\u043e\u043c \u043f\u043e\u0440\u044f\u0434\u043a\u0435, \u0435\u0441\u043b\u0438 \u0441\u0443\u043c\u043c\u0430 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 \u043d\u0435 \u043c\u0435\u043d\u0435\u0435 50 \u0442\u044b\u0441. \u0440\u0443\u0431\u043b\u0435\u0439 \u0438 \u043d\u0435 \u043f\u0440\u0435\u0432\u044b\u0448\u0430\u0435\u0442 500 \u0442\u044b\u0441. \u0440\u0443\u0431\u043b\u0435\u0439 \u0431\u0435\u0437 \u0443\u0447\u0435\u0442\u0430 \u0448\u0442\u0440\u0430\u0444\u043e\u0432, \u043f\u0435\u043d\u0438, \u043f\u0440\u043e\u0446\u0435\u043d\u0442\u043e\u0432 \u0437\u0430 \u043f\u0440\u043e\u0441\u0440\u043e\u0447\u043a\u0443 \u043f\u043b\u0430\u0442\u0435\u0436\u0430 \u0438 \u043f\u0440\u043e\u0447\u0438\u0445 \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0445 \u0438\u043b\u0438 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u044b\u0445 \u0441\u0430\u043d\u043a\u0446\u0438\u0439. \u0423 \u0444\u0438\u0437\u043b\u0438\u0446 \u0438 \u0438\u043d\u0434\u0438\u0432\u0438\u0434\u0443\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0442\u0435\u043b\u0435\u0439 \u043f\u043e\u044f\u0432\u0438\u043b\u0430\u0441\u044c \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u043e\u0439\u0442\u0438 \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u0443 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u0430 \u0431\u0435\u0437 \u0443\u0447\u0430\u0441\u0442\u0438\u044f \u0441\u0443\u0434\u0430 \u0438 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u043e\u0433\u043e \u0443\u043f\u0440\u0430\u0432\u043b\u044f\u044e\u0449\u0435\u0433\u043e \u2014 \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e \u043f\u043e\u0434\u0430\u0442\u044c \u0441\u043e\u043e\u0442\u0432\u0435\u0442\u0441\u0442\u0432\u0443\u044e\u0449\u0435\u0435 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0435 \u0447\u0435\u0440\u0435\u0437 \u041c\u0424\u0426. \u0421\u0443\u043c\u043c\u0443 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u0438 \u0438 \u0441\u043f\u0438\u0441\u043e\u043a \u0432\u0441\u0435\u0445 \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u044b\u0445 \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044e \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432 \u043d\u0443\u0436\u043d\u043e \u043f\u0440\u0435\u0434\u043e\u0441\u0442\u0430\u0432\u0438\u0442\u044c \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e. \u0415\u0441\u043b\u0438 \u0432\u0441\u0435 \u0443\u0441\u043b\u043e\u0432\u0438\u044f \u0441\u043e\u0431\u043b\u044e\u0434\u0435\u043d\u044b, \u0441\u0432\u0435\u0434\u0435\u043d\u0438\u044f \u0432\u043d\u0435\u0441\u0443\u0442 \u0432 \u0415\u0434\u0438\u043d\u044b\u0439 \u0444\u0435\u0434\u0435\u0440\u0430\u043b\u044c\u043d\u044b\u0439 \u0440\u0435\u0435\u0441\u0442\u0440 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 \u0442\u0440\u0435\u0445 \u0440\u0430\u0431\u043e\u0447\u0438\u0445 \u0434\u043d\u0435\u0439. \u041f\u0440\u0438 \u044d\u0442\u043e\u043c \u043d\u0430 \u043c\u043e\u043c\u0435\u043d\u0442 \u043f\u043e\u0434\u0430\u0447\u0438 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u044f \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044f \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u043e\u043a\u043e\u043d\u0447\u0435\u043d\u043e \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u043e \u0441 \u0432\u043e\u0437\u0432\u0440\u0430\u0449\u0435\u043d\u0438\u0435\u043c \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0433\u043e \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u0430 \u0432\u0437\u044b\u0441\u043a\u0430\u0442\u0435\u043b\u044e. \u042d\u0442\u043e \u0437\u043d\u0430\u0447\u0438\u0442, \u0447\u0442\u043e \u0443 \u043f\u043e\u0442\u0435\u043d\u0446\u0438\u0430\u043b\u044c\u043d\u043e\u0433\u043e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0430 \u043d\u0435 \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0430, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u043c\u043e\u0436\u043d\u043e \u0432\u0437\u044b\u0441\u043a\u0430\u0442\u044c. \u041a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0433\u0440\u0430\u0436\u0434\u0430\u043d\u0438\u043d\u0430 \u043d\u0435 \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u0432\u043e\u0437\u0431\u0443\u0436\u0434\u0435\u043d\u043e \u0434\u0440\u0443\u0433\u043e\u0435 \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u043e. \u0412 \u043f\u0435\u0440\u0438\u043e\u0434 \u0432\u0441\u0435\u0439 \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u044b \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044c \u043d\u0435 \u0441\u043c\u043e\u0436\u0435\u0442 \u0431\u0440\u0430\u0442\u044c \u0437\u0430\u0439\u043c\u044b, \u043a\u0440\u0435\u0434\u0438\u0442\u044b, \u0432\u044b\u0434\u0430\u0432\u0430\u0442\u044c \u043f\u043e\u0440\u0443\u0447\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430, \u0441\u043e\u0432\u0435\u0440\u0448\u0430\u0442\u044c \u0438\u043d\u044b\u0435 \u043e\u0431\u0435\u0441\u043f\u0435\u0447\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0435 \u0441\u0434\u0435\u043b\u043a\u0438. \u0412\u043d\u0435\u0441\u0443\u0434\u0435\u0431\u043d\u043e\u0435 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u043e \u0431\u0443\u0434\u0435\u0442 \u0434\u043b\u0438\u0442\u044c\u0441\u044f \u0448\u0435\u0441\u0442\u044c \u043c\u0435\u0441\u044f\u0446\u0435\u0432, \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0442\u0430\u043a\u0436\u0435 \u0431\u0443\u0434\u0435\u0442 \u0434\u0435\u0439\u0441\u0442\u0432\u043e\u0432\u0430\u0442\u044c \u043c\u043e\u0440\u0430\u0442\u043e\u0440\u0438\u0439 \u043d\u0430 \u0443\u0434\u043e\u0432\u043b\u0435\u0442\u0432\u043e\u0440\u0435\u043d\u0438\u0435 \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u0438\u0439 \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432, \u043e\u0442\u043c\u0435\u0447\u0435\u043d\u043d\u044b\u0445 \u0432 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0438 \u0434\u043e\u043b\u0436\u043d\u0438\u043a\u0430, \u0438 \u043c\u043e\u0440\u0430\u0442\u043e\u0440\u0438\u0439 \u043e\u0431 \u0443\u043f\u043b\u0430\u0442\u0435 \u043e\u0431\u044f\u0437\u0430\u0442\u0435\u043b\u044c\u043d\u044b\u0445 \u043f\u043b\u0430\u0442\u0435\u0436\u0435\u0439. \u041a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u043f\u0440\u0435\u043a\u0440\u0430\u0449\u0430\u0435\u0442\u0441\u044f \u043d\u0430\u0447\u0438\u0441\u043b\u0435\u043d\u0438\u0435 \u043d\u0435\u0443\u0441\u0442\u043e\u0435\u043a \u0438 \u0438\u043d\u044b\u0445 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u044b\u0445 \u0441\u0430\u043d\u043a\u0446\u0438\u0439; \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0435 \u0432\u0437\u044b\u0441\u043a\u0430\u043d\u0438\u044f (\u043a\u0440\u043e\u043c\u0435 \u0430\u043b\u0438\u043c\u0435\u043d\u0442\u043e\u0432) \u0442\u0430\u043a\u0436\u0435 \u0431\u0443\u0434\u0443\u0442 \u043f\u0440\u0438\u043e\u0441\u0442\u0430\u043d\u043e\u0432\u043b\u0435\u043d\u044b. \u041f\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044e \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u044b \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044f \u043e\u0441\u0432\u043e\u0431\u043e\u0434\u044f\u0442 \u043e\u0442 \u0434\u0430\u043b\u044c\u043d\u0435\u0439\u0448\u0435\u0433\u043e \u0432\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u044f \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u0438\u0439 \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432, \u0443\u043a\u0430\u0437\u0430\u043d\u043d\u044b\u0445 \u0432 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0438 \u043e \u043f\u0440\u0438\u0437\u043d\u0430\u043d\u0438\u0438 \u0435\u0433\u043e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u043e\u043c, \u0430 \u044d\u0442\u0430 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u0438\u0437\u043d\u0430\u0435\u0442\u0441\u044f \u0431\u0435\u0437\u043d\u0430\u0434\u0435\u0436\u043d\u043e\u0439. \u0412 \u043f\u0440\u043e\u0448\u043b\u043e\u043c \u043c\u0435\u0441\u044f\u0446\u0435 \u0441\u0442\u0430\u043b\u043e \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u043e, \u0447\u0442\u043e \u0437\u0430 \u043f\u0435\u0440\u0432\u043e\u0435 \u043f\u043e\u043b\u0443\u0433\u043e\u0434\u0438\u0435 2020 \u0433\u043e\u0434\u0430 \u0440\u043e\u0441\u0441\u0438\u0439\u0441\u043a\u0438\u0435 \u0441\u0443\u0434\u044b \u043f\u0440\u0438\u0437\u043d\u0430\u043b\u0438 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0430\u043c\u0438 42,7 \u0442\u044b\u0441. \u0433\u0440\u0430\u0436\u0434\u0430\u043d (\u0432 \u0442\u043e\u043c \u0447\u0438\u0441\u043b\u0435 \u0438\u043d\u0434\u0438\u0432\u0438\u0434\u0443\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0442\u0435\u043b\u0435\u0439) \u2014 \u043f\u043e \u0434\u0430\u043d\u043d\u044b\u043c \u0435\u0434\u0438\u043d\u043e\u0433\u043e \u0440\u0435\u0435\u0441\u0442\u0440\u0430 \u00ab\u0424\u0435\u0434\u0440\u0435\u0441\u0443\u0440\u0441\u00bb, \u044d\u0442\u043e \u043d\u0430 47,2% \u0431\u043e\u043b\u044c\u0448\u0435 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044f \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u043e\u0433\u043e \u043f\u0435\u0440\u0438\u043e\u0434\u0430 2019 \u0433\u043e\u0434\u0430. \u0420\u043e\u0441\u0442 \u0447\u0438\u0441\u043b\u0430 \u043e\u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0438\u0432\u0448\u0438\u0445\u0441\u044f \u0433\u0440\u0430\u0436\u0434\u0430\u043d \u0432\u043e \u0432\u0442\u043e\u0440\u043e\u043c \u043a\u0432\u0430\u0440\u0442\u0430\u043b\u0435 \u043f\u043e \u0441\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u044e \u0441 \u043f\u0435\u0440\u0432\u044b\u043c \u0437\u0430\u043c\u0435\u0434\u043b\u0438\u043b\u0441\u044f \u2014 \u0442\u0430\u043a\u0430\u044f \u0434\u0438\u043d\u0430\u043c\u0438\u043a\u0430 \u043e\u0431\u0443\u0441\u043b\u043e\u0432\u043b\u0435\u043d\u0430 \u0442\u0435\u043c, \u0447\u0442\u043e \u0432 \u043f\u0435\u0440\u0438\u043e\u0434 \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u0438\u0439 \u0441 19 \u043c\u0430\u0440\u0442\u0430 \u043f\u043e 11 \u043c\u0430\u044f \u0441\u0443\u0434\u044b \u0440\u0435\u0434\u043a\u043e \u0440\u0430\u0441\u0441\u043c\u0430\u0442\u0440\u0438\u0432\u0430\u043b\u0438 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u043d\u044b\u0435 \u0434\u0435\u043b\u0430 \u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0439 \u0438 \u043c\u0435\u043d\u044c\u0448\u0435, \u0447\u0435\u043c \u043e\u0431\u044b\u0447\u043d\u043e, \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0433\u0440\u0430\u0436\u0434\u0430\u043d, \u043e\u0431\u044a\u044f\u0441\u043d\u044f\u043b \u0440\u0443\u043a\u043e\u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c \u043f\u0440\u043e\u0435\u043a\u0442\u0430 \u00ab\u0424\u0435\u0434\u0440\u0435\u0441\u0443\u0440\u0441\u00bb \u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u042e\u0445\u043d\u0438\u043d. \u041e\u043d \u043f\u0440\u043e\u0433\u043d\u043e\u0437\u0438\u0440\u0443\u0435\u0442, \u0447\u0442\u043e \u0432\u043e \u0432\u0442\u043e\u0440\u043e\u043c \u043f\u043e\u043b\u0443\u0433\u043e\u0434\u0438\u0438 \u043c\u044b \u0443\u0432\u0438\u0434\u0438\u043c \u0440\u043e\u0441\u0442 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044f, \u043a\u043e\u0433\u0434\u0430 \u0441\u0443\u0434\u044b \u0440\u0430\u0441\u0441\u043c\u043e\u0442\u0440\u044f\u0442 \u0432\u0441\u0435 \u0434\u0435\u043b\u0430, \u0447\u0442\u043e \u043d\u0435 \u0441\u043c\u043e\u0433\u043b\u0438 \u0440\u0430\u043d\u0435\u0435 \u0432 \u0440\u0435\u0436\u0438\u043c\u0435 \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u0438\u0439. \u041f\u043e \u0435\u0433\u043e \u0434\u0430\u043d\u043d\u044b\u043c, \u0443\u0436\u0435 \u0432 \u0438\u044e\u043d\u0435 \u0447\u0438\u0441\u043b\u043e \u043b\u0438\u0447\u043d\u044b\u0445 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432 \u0432\u044b\u0440\u043e\u0441\u043b\u043e \u0434\u043e 11,5 \u0442\u044b\u0441., \u0447\u0442\u043e \u0432 \u0434\u0432\u0430 \u0440\u0430\u0437\u0430 \u043f\u0440\u0435\u0432\u044b\u0448\u0430\u0435\u0442 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044c \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u043e\u0433\u043e \u043f\u0435\u0440\u0438\u043e\u0434\u0430 2019 \u0433\u043e\u0434\u0430.", "example_title": "\u041d\u043e\u0432\u043e\u0441\u0442\u0438"}, {"text": "\u0410\u043a\u0442\u0443\u0430\u043b\u044c\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b. \u042d\u043b\u0435\u043a\u0442\u0440\u043e\u043d\u043d\u0430\u044f \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u044f \u0438\u0433\u0440\u0430\u0435\u0442 \u0432\u0441\u0435 \u0431\u043e\u043b\u044c\u0448\u0443\u044e \u0440\u043e\u043b\u044c \u0432\u043e \u0432\u0441\u0435\u0445 \u0441\u0444\u0435\u0440\u0430\u0445 \u0436\u0438\u0437\u043d\u0438 \u0441\u043e\u0432\u0440\u0435\u043c\u0435\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0449\u0435\u0441\u0442\u0432\u0430. \u0412 \u043f\u043e\u0441\u043b\u0435\u0434\u043d\u0438\u0435 \u0433\u043e\u0434\u044b \u043e\u0431\u044a\u0435\u043c \u043d\u0430\u0443\u0447\u043d\u043e-\u0442\u0435\u0445\u043d\u0438\u0447\u0435\u0441\u043a\u043e\u0439 \u0442\u0435\u043a\u0441\u0442\u043e\u0432\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u0432 \u044d\u043b\u0435\u043a\u0442\u0440\u043e\u043d\u043d\u043e\u043c \u0432\u0438\u0434\u0435 \u0432\u043e\u0437\u0440\u043e\u0441 \u043d\u0430\u0441\u0442\u043e\u043b\u044c\u043a\u043e, \u0447\u0442\u043e \u0432\u043e\u0437\u043d\u0438\u043a\u0430\u0435\u0442 \u0443\u0433\u0440\u043e\u0437\u0430 \u043e\u0431\u0435\u0441\u0446\u0435\u043d\u0438\u0432\u0430\u043d\u0438\u044f \u044d\u0442\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u0432 \u0441\u0432\u044f\u0437\u0438 \u0441 \u0442\u0440\u0443\u0434\u043d\u043e\u0441\u0442\u044f\u043c\u0438 \u043f\u043e\u0438\u0441\u043a\u0430 \u043d\u0435\u043e\u0431\u0445\u043e\u0434\u0438\u043c\u044b\u0445 \u0441\u0432\u0435\u0434\u0435\u043d\u0438\u0439 \u0441\u0440\u0435\u0434\u0438 \u043c\u043d\u043e\u0436\u0435\u0441\u0442\u0432\u0430 \u0434\u043e\u0441\u0442\u0443\u043f\u043d\u044b\u0445 \u0442\u0435\u043a\u0441\u0442\u043e\u0432. \u0420\u0430\u0437\u0432\u0438\u0442\u0438\u0435 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u044b\u0445 \u0440\u0435\u0441\u0443\u0440\u0441\u043e\u0432 \u0418\u043d\u0442\u0435\u0440\u043d\u0435\u0442 \u043c\u043d\u043e\u0433\u043e\u043a\u0440\u0430\u0442\u043d\u043e \u0443\u0441\u0443\u0433\u0443\u0431\u0438\u043b\u043e \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u0443 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u043e\u0439 \u043f\u0435\u0440\u0435\u0433\u0440\u0443\u0437\u043a\u0438. \u0412 \u044d\u0442\u043e\u0439 \u0441\u0438\u0442\u0443\u0430\u0446\u0438\u0438 \u043e\u0441\u043e\u0431\u0435\u043d\u043d\u043e \u0430\u043a\u0442\u0443\u0430\u043b\u044c\u043d\u044b\u043c\u0438 \u0441\u0442\u0430\u043d\u043e\u0432\u044f\u0442\u0441\u044f \u043c\u0435\u0442\u043e\u0434\u044b \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0437\u0430\u0446\u0438\u0438 \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f \u0442\u0435\u043a\u0441\u0442\u043e\u0432\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438, \u0442\u043e \u0435\u0441\u0442\u044c \u043c\u0435\u0442\u043e\u0434\u044b \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u0438\u044f \u0441\u0436\u0430\u0442\u043e\u0433\u043e \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0442\u0435\u043a\u0441\u0442\u043e\u0432\u044b\u0445 \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u043e\u0432\u2013\u0440\u0435\u0444\u0435\u0440\u0430\u0442\u043e\u0432 (\u0430\u043d\u043d\u043e\u0442\u0430\u0446\u0438\u0439). \u041f\u043e\u0441\u0442\u0430\u043d\u043e\u0432\u043a\u0430 \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0433\u043e \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f \u0442\u0435\u043a\u0441\u0442\u0430 \u0438 \u0441\u043e\u043e\u0442\u0432\u0435\u0442\u0441\u0442\u0432\u0435\u043d\u043d\u043e \u043f\u043e\u043f\u044b\u0442\u043a\u0438 \u0435\u0435 \u0440\u0435\u0448\u0435\u043d\u0438\u044f \u0441 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u0438\u0435\u043c \u0440\u0430\u0437\u043b\u0438\u0447\u043d\u044b\u0445 \u043f\u043e\u0434\u0445\u043e\u0434\u043e\u0432 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u043b\u0438\u0441\u044c \u043c\u043d\u043e\u0433\u0438\u043c\u0438 \u0438\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044f\u043c\u0438. \u0418\u0441\u0442\u043e\u0440\u0438\u044f \u043f\u0440\u0438\u043c\u0435\u043d\u0435\u043d\u0438\u044f \u0432\u044b\u0447\u0438\u0441\u043b\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0442\u0435\u0445\u043d\u0438\u043a\u0438 \u0434\u043b\u044f \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f \u043d\u0430\u0441\u0447\u0438\u0442\u044b\u0432\u0430\u0435\u0442 \u0443\u0436\u0435 \u0431\u043e\u043b\u0435\u0435 50 \u043b\u0435\u0442 \u0438 \u0441\u0432\u044f\u0437\u0430\u043d\u0430 \u0441 \u0438\u043c\u0435\u043d\u0430\u043c\u0438 \u0442\u0430\u043a\u0438\u0445 \u0438\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0435\u0439, \u043a\u0430\u043a \u0413.\u041f. \u041b\u0443\u043d, \u0412.\u0415. \u0411\u0435\u0440\u0437\u043e\u043d, \u0418.\u041f. C\u0435\u0432\u0431\u043e, \u042d.\u0424. \u0421\u043a\u043e\u0440\u043e\u0445\u043e\u0434\u044c\u043a\u043e, \u0414.\u0413. \u041b\u0430\u0445\u0443\u0442\u0438, \u0420.\u0413. \u041f\u0438\u043e\u0442\u0440\u043e\u0432\u0441\u043a\u0438\u0439 \u0438 \u0434\u0440. \u0417\u0430 \u044d\u0442\u0438 \u0433\u043e\u0434\u044b \u0432\u044b\u0440\u0430\u0431\u043e\u0442\u0430\u043d\u044b \u043c\u043d\u043e\u0433\u043e\u0447\u0438\u0441\u043b\u0435\u043d\u043d\u044b\u0435 \u043f\u043e\u0434\u0445\u043e\u0434\u044b \u043a \u0440\u0435\u0448\u0435\u043d\u0438\u044e \u0434\u0430\u043d\u043d\u043e\u0439 \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e \u0447\u0435\u0442\u043a\u043e \u043f\u043e\u0434\u0440\u0430\u0437\u0434\u0435\u043b\u044f\u044e\u0442\u0441\u044f \u043d\u0430 \u0434\u0432\u0430 \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u044f: \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0435 \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0435, \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u043d\u043e\u0435 \u043d\u0430 \u044d\u043a\u0441\u0442\u0440\u0430\u0433\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0438 \u0438\u0437 \u043f\u0435\u0440\u0432\u0438\u0447\u043d\u044b\u0445 \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u043e\u0432 \u0441 \u043f\u043e\u043c\u043e\u0449\u044c\u044e \u043e\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u043d\u044b\u0445 \u0444\u043e\u0440\u043c\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0438\u0437\u043d\u0430\u043a\u043e\u0432 \u00ab\u043d\u0430\u0438\u0431\u043e\u043b\u0435\u0435 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0442\u0438\u0432\u043d\u044b\u0445\u00bb \u0444\u0440\u0430\u0437 (\u0444\u0440\u0430\u0433\u043c\u0435\u043d\u0442\u043e\u0432), \u0441\u043e\u0432\u043e\u043a\u0443\u043f\u043d\u043e\u0441\u0442\u044c \u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u043e\u0431\u0440\u0430\u0437\u0443\u0435\u0442 \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0439 \u044d\u043a\u0441\u0442\u0440\u0430\u043a\u0442; \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0435 \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0435, \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u043d\u043e\u0435 \u043d\u0430 \u0432\u044b\u0434\u0435\u043b\u0435\u043d\u0438\u0438 \u0438\u0437 \u0442\u0435\u043a\u0441\u0442\u043e\u0432 \u0441 \u043f\u043e\u043c\u043e\u0449\u044c\u044e \u0441\u043f\u0435\u0446\u0438\u0430\u043b\u044c\u043d\u044b\u0445 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u044b\u0445 \u044f\u0437\u044b\u043a\u043e\u0432 \u043d\u0430\u0438\u0431\u043e\u043b\u0435\u0435 \u0441\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u0438 \u043f\u043e\u0440\u043e\u0436\u0434\u0435\u043d\u0438\u0438 \u043d\u043e\u0432\u044b\u0445 \u0442\u0435\u043a\u0441\u0442\u043e\u0432 (\u0440\u0435\u0444\u0435\u0440\u0430\u0442\u043e\u0432), \u0441\u043e\u0434\u0435\u0440\u0436\u0430\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0431\u043e\u0431\u0449\u0430\u044e\u0449\u0438\u0445 \u043f\u0435\u0440\u0432\u0438\u0447\u043d\u044b\u0435 \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u044b.", "example_title": "\u041d\u0430\u0443\u0447\u043d\u0430\u044f \u0441\u0442\u0430\u0442\u044c\u044f"}]} | IlyaGusev/mbart_ru_sum_gazeta | null | [
"transformers",
"pytorch",
"safetensors",
"mbart",
"text2text-generation",
"summarization",
"ru",
"dataset:IlyaGusev/gazeta",
"arxiv:2006.11063",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | transformers |
# NewsTgRuBERT
Training script: https://github.com/dialogue-evaluation/Russian-News-Clustering-and-Headline-Generation/blob/main/train_mlm.py | {"language": ["ru"], "license": "apache-2.0"} | IlyaGusev/news_tg_rubert | null | [
"transformers",
"pytorch",
"ru",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
token-classification | transformers |
# RuBERTExtSumGazeta
## Model description
Model for extractive summarization based on [rubert-base-cased](DeepPavlov/rubert-base-cased)
## Intended uses & limitations
#### How to use
Colab: [link](https://colab.research.google.com/drive/1Q8_v3H-kxdJhZIiyLYat7Kj02qDq7M1L)
```python
import razdel
from transformers import AutoTokenizer, BertForTokenClassification
model_name = "IlyaGusev/rubert_ext_sum_gazeta"
tokenizer = AutoTokenizer.from_pretrained(model_name)
sep_token = tokenizer.sep_token
sep_token_id = tokenizer.sep_token_id
model = BertForTokenClassification.from_pretrained(model_name)
article_text = "..."
sentences = [s.text for s in razdel.sentenize(article_text)]
article_text = sep_token.join(sentences)
inputs = tokenizer(
[article_text],
max_length=500,
padding="max_length",
truncation=True,
return_tensors="pt",
)
sep_mask = inputs["input_ids"][0] == sep_token_id
# Fix token_type_ids
current_token_type_id = 0
for pos, input_id in enumerate(inputs["input_ids"][0]):
inputs["token_type_ids"][0][pos] = current_token_type_id
if input_id == sep_token_id:
current_token_type_id = 1 - current_token_type_id
# Infer model
with torch.no_grad():
outputs = model(**inputs)
logits = outputs.logits[0, :, 1]
# Choose sentences
logits = logits[sep_mask]
logits, indices = logits.sort(descending=True)
logits, indices = logits.cpu().tolist(), indices.cpu().tolist()
pairs = list(zip(logits, indices))
pairs = pairs[:3]
indices = list(sorted([idx for _, idx in pairs]))
summary = " ".join([sentences[idx] for idx in indices])
print(summary)
```
#### Limitations and bias
- The model should work well with Gazeta.ru articles, but for any other agencies it can suffer from domain shift
## Training data
- Dataset: [Gazeta](https://huggingface.co/datasets/IlyaGusev/gazeta)
## Training procedure
TBD
## Eval results
TBD
Evaluation: https://github.com/IlyaGusev/summarus/blob/master/evaluate.py
Flags: --language ru --tokenize-after --lower
| {"language": ["ru"], "license": "apache-2.0", "tags": ["summarization", "token-classification", "t5"], "datasets": ["IlyaGusev/gazeta"], "inference": false, "widget": [{"text": "\u0421 1 \u0441\u0435\u043d\u0442\u044f\u0431\u0440\u044f \u0432 \u0420\u043e\u0441\u0441\u0438\u0438 \u0432\u0441\u0442\u0443\u043f\u0430\u044e\u0442 \u0432 \u0441\u0438\u043b\u0443 \u043f\u043e\u043f\u0440\u0430\u0432\u043a\u0438 \u0432 \u0437\u0430\u043a\u043e\u043d \u00ab\u041e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u0435\u00bb \u2014 \u0442\u0435\u043f\u0435\u0440\u044c \u0434\u043e\u043b\u0436\u043d\u0438\u043a\u0438 \u0441\u043c\u043e\u0433\u0443\u0442 \u043e\u0441\u0432\u043e\u0431\u043e\u0436\u0434\u0430\u0442\u044c\u0441\u044f \u043e\u0442 \u043d\u0435\u043f\u043e\u0441\u0438\u043b\u044c\u043d\u044b\u0445 \u043e\u0431\u044f\u0437\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432 \u0432\u043e \u0432\u043d\u0435\u0441\u0443\u0434\u0435\u0431\u043d\u043e\u043c \u043f\u043e\u0440\u044f\u0434\u043a\u0435, \u0435\u0441\u043b\u0438 \u0441\u0443\u043c\u043c\u0430 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 \u043d\u0435 \u043c\u0435\u043d\u0435\u0435 50 \u0442\u044b\u0441. \u0440\u0443\u0431\u043b\u0435\u0439 \u0438 \u043d\u0435 \u043f\u0440\u0435\u0432\u044b\u0448\u0430\u0435\u0442 500 \u0442\u044b\u0441. \u0440\u0443\u0431\u043b\u0435\u0439 \u0431\u0435\u0437 \u0443\u0447\u0435\u0442\u0430 \u0448\u0442\u0440\u0430\u0444\u043e\u0432, \u043f\u0435\u043d\u0438, \u043f\u0440\u043e\u0446\u0435\u043d\u0442\u043e\u0432 \u0437\u0430 \u043f\u0440\u043e\u0441\u0440\u043e\u0447\u043a\u0443 \u043f\u043b\u0430\u0442\u0435\u0436\u0430 \u0438 \u043f\u0440\u043e\u0447\u0438\u0445 \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0445 \u0438\u043b\u0438 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u044b\u0445 \u0441\u0430\u043d\u043a\u0446\u0438\u0439.[SEP]\u0423 \u0444\u0438\u0437\u043b\u0438\u0446 \u0438 \u0438\u043d\u0434\u0438\u0432\u0438\u0434\u0443\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0442\u0435\u043b\u0435\u0439 \u043f\u043e\u044f\u0432\u0438\u043b\u0430\u0441\u044c \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u043e\u0439\u0442\u0438 \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u0443 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u0430 \u0431\u0435\u0437 \u0443\u0447\u0430\u0441\u0442\u0438\u044f \u0441\u0443\u0434\u0430 \u0438 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u043e\u0433\u043e \u0443\u043f\u0440\u0430\u0432\u043b\u044f\u044e\u0449\u0435\u0433\u043e \u2014 \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e \u043f\u043e\u0434\u0430\u0442\u044c \u0441\u043e\u043e\u0442\u0432\u0435\u0442\u0441\u0442\u0432\u0443\u044e\u0449\u0435\u0435 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0435 \u0447\u0435\u0440\u0435\u0437 \u041c\u0424\u0426.[SEP]\u0421\u0443\u043c\u043c\u0443 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u0438 \u0438 \u0441\u043f\u0438\u0441\u043e\u043a \u0432\u0441\u0435\u0445 \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u044b\u0445 \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044e \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432 \u043d\u0443\u0436\u043d\u043e \u043f\u0440\u0435\u0434\u043e\u0441\u0442\u0430\u0432\u0438\u0442\u044c \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e.[SEP]\u0415\u0441\u043b\u0438 \u0432\u0441\u0435 \u0443\u0441\u043b\u043e\u0432\u0438\u044f \u0441\u043e\u0431\u043b\u044e\u0434\u0435\u043d\u044b, \u0441\u0432\u0435\u0434\u0435\u043d\u0438\u044f \u0432\u043d\u0435\u0441\u0443\u0442 \u0432 \u0415\u0434\u0438\u043d\u044b\u0439 \u0444\u0435\u0434\u0435\u0440\u0430\u043b\u044c\u043d\u044b\u0439 \u0440\u0435\u0435\u0441\u0442\u0440 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 \u0442\u0440\u0435\u0445 \u0440\u0430\u0431\u043e\u0447\u0438\u0445 \u0434\u043d\u0435\u0439.[SEP]\u041f\u0440\u0438 \u044d\u0442\u043e\u043c \u043d\u0430 \u043c\u043e\u043c\u0435\u043d\u0442 \u043f\u043e\u0434\u0430\u0447\u0438 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u044f \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044f \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u043e\u043a\u043e\u043d\u0447\u0435\u043d\u043e \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u043e \u0441 \u0432\u043e\u0437\u0432\u0440\u0430\u0449\u0435\u043d\u0438\u0435\u043c \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0433\u043e \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u0430 \u0432\u0437\u044b\u0441\u043a\u0430\u0442\u0435\u043b\u044e.[SEP]\u042d\u0442\u043e \u0437\u043d\u0430\u0447\u0438\u0442, \u0447\u0442\u043e \u0443 \u043f\u043e\u0442\u0435\u043d\u0446\u0438\u0430\u043b\u044c\u043d\u043e\u0433\u043e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0430 \u043d\u0435 \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0430, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u043c\u043e\u0436\u043d\u043e \u0432\u0437\u044b\u0441\u043a\u0430\u0442\u044c.[SEP]\u041a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0433\u0440\u0430\u0436\u0434\u0430\u043d\u0438\u043d\u0430 \u043d\u0435 \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u0432\u043e\u0437\u0431\u0443\u0436\u0434\u0435\u043d\u043e \u0434\u0440\u0443\u0433\u043e\u0435 \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u043e.[SEP]\u0412 \u043f\u0435\u0440\u0438\u043e\u0434 \u0432\u0441\u0435\u0439 \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u044b \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044c \u043d\u0435 \u0441\u043c\u043e\u0436\u0435\u0442 \u0431\u0440\u0430\u0442\u044c \u0437\u0430\u0439\u043c\u044b, \u043a\u0440\u0435\u0434\u0438\u0442\u044b, \u0432\u044b\u0434\u0430\u0432\u0430\u0442\u044c \u043f\u043e\u0440\u0443\u0447\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430, \u0441\u043e\u0432\u0435\u0440\u0448\u0430\u0442\u044c \u0438\u043d\u044b\u0435 \u043e\u0431\u0435\u0441\u043f\u0435\u0447\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0435 \u0441\u0434\u0435\u043b\u043a\u0438.[SEP]\u0412\u043d\u0435\u0441\u0443\u0434\u0435\u0431\u043d\u043e\u0435 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u043e \u0431\u0443\u0434\u0435\u0442 \u0434\u043b\u0438\u0442\u044c\u0441\u044f \u0448\u0435\u0441\u0442\u044c \u043c\u0435\u0441\u044f\u0446\u0435\u0432, \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0442\u0430\u043a\u0436\u0435 \u0431\u0443\u0434\u0435\u0442 \u0434\u0435\u0439\u0441\u0442\u0432\u043e\u0432\u0430\u0442\u044c \u043c\u043e\u0440\u0430\u0442\u043e\u0440\u0438\u0439 \u043d\u0430 \u0443\u0434\u043e\u0432\u043b\u0435\u0442\u0432\u043e\u0440\u0435\u043d\u0438\u0435 \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u0438\u0439 \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432, \u043e\u0442\u043c\u0435\u0447\u0435\u043d\u043d\u044b\u0445 \u0432 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0438 \u0434\u043e\u043b\u0436\u043d\u0438\u043a\u0430, \u0438 \u043c\u043e\u0440\u0430\u0442\u043e\u0440\u0438\u0439 \u043e\u0431 \u0443\u043f\u043b\u0430\u0442\u0435 \u043e\u0431\u044f\u0437\u0430\u0442\u0435\u043b\u044c\u043d\u044b\u0445 \u043f\u043b\u0430\u0442\u0435\u0436\u0435\u0439.[SEP]\u041a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u043f\u0440\u0435\u043a\u0440\u0430\u0449\u0430\u0435\u0442\u0441\u044f \u043d\u0430\u0447\u0438\u0441\u043b\u0435\u043d\u0438\u0435 \u043d\u0435\u0443\u0441\u0442\u043e\u0435\u043a \u0438 \u0438\u043d\u044b\u0445 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u044b\u0445 \u0441\u0430\u043d\u043a\u0446\u0438\u0439; \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0435 \u0432\u0437\u044b\u0441\u043a\u0430\u043d\u0438\u044f (\u043a\u0440\u043e\u043c\u0435 \u0430\u043b\u0438\u043c\u0435\u043d\u0442\u043e\u0432) \u0442\u0430\u043a\u0436\u0435 \u0431\u0443\u0434\u0443\u0442 \u043f\u0440\u0438\u043e\u0441\u0442\u0430\u043d\u043e\u0432\u043b\u0435\u043d\u044b.[SEP]\u041f\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044e \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u044b \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044f \u043e\u0441\u0432\u043e\u0431\u043e\u0434\u044f\u0442 \u043e\u0442 \u0434\u0430\u043b\u044c\u043d\u0435\u0439\u0448\u0435\u0433\u043e \u0432\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u044f \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u0438\u0439 \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432, \u0443\u043a\u0430\u0437\u0430\u043d\u043d\u044b\u0445 \u0432 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0438 \u043e \u043f\u0440\u0438\u0437\u043d\u0430\u043d\u0438\u0438 \u0435\u0433\u043e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u043e\u043c, \u0430 \u044d\u0442\u0430 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u0438\u0437\u043d\u0430\u0435\u0442\u0441\u044f \u0431\u0435\u0437\u043d\u0430\u0434\u0435\u0436\u043d\u043e\u0439.[SEP]\u0412 \u043f\u0440\u043e\u0448\u043b\u043e\u043c \u043c\u0435\u0441\u044f\u0446\u0435 \u0441\u0442\u0430\u043b\u043e \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u043e, \u0447\u0442\u043e \u0437\u0430 \u043f\u0435\u0440\u0432\u043e\u0435 \u043f\u043e\u043b\u0443\u0433\u043e\u0434\u0438\u0435 2020 \u0433\u043e\u0434\u0430 \u0440\u043e\u0441\u0441\u0438\u0439\u0441\u043a\u0438\u0435 \u0441\u0443\u0434\u044b \u043f\u0440\u0438\u0437\u043d\u0430\u043b\u0438 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0430\u043c\u0438 42,7 \u0442\u044b\u0441. \u0433\u0440\u0430\u0436\u0434\u0430\u043d (\u0432 \u0442\u043e\u043c \u0447\u0438\u0441\u043b\u0435 \u0438\u043d\u0434\u0438\u0432\u0438\u0434\u0443\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0442\u0435\u043b\u0435\u0439) \u2014 \u043f\u043e \u0434\u0430\u043d\u043d\u044b\u043c \u0435\u0434\u0438\u043d\u043e\u0433\u043e \u0440\u0435\u0435\u0441\u0442\u0440\u0430 \u00ab\u0424\u0435\u0434\u0440\u0435\u0441\u0443\u0440\u0441\u00bb, \u044d\u0442\u043e \u043d\u0430 47,2% \u0431\u043e\u043b\u044c\u0448\u0435 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044f \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u043e\u0433\u043e \u043f\u0435\u0440\u0438\u043e\u0434\u0430 2019 \u0433\u043e\u0434\u0430.[SEP]\u0420\u043e\u0441\u0442 \u0447\u0438\u0441\u043b\u0430 \u043e\u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0438\u0432\u0448\u0438\u0445\u0441\u044f \u0433\u0440\u0430\u0436\u0434\u0430\u043d \u0432\u043e \u0432\u0442\u043e\u0440\u043e\u043c \u043a\u0432\u0430\u0440\u0442\u0430\u043b\u0435 \u043f\u043e \u0441\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u044e \u0441 \u043f\u0435\u0440\u0432\u044b\u043c \u0437\u0430\u043c\u0435\u0434\u043b\u0438\u043b\u0441\u044f \u2014 \u0442\u0430\u043a\u0430\u044f \u0434\u0438\u043d\u0430\u043c\u0438\u043a\u0430 \u043e\u0431\u0443\u0441\u043b\u043e\u0432\u043b\u0435\u043d\u0430 \u0442\u0435\u043c, \u0447\u0442\u043e \u0432 \u043f\u0435\u0440\u0438\u043e\u0434 \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u0438\u0439 \u0441 19 \u043c\u0430\u0440\u0442\u0430 \u043f\u043e 11 \u043c\u0430\u044f \u0441\u0443\u0434\u044b \u0440\u0435\u0434\u043a\u043e \u0440\u0430\u0441\u0441\u043c\u0430\u0442\u0440\u0438\u0432\u0430\u043b\u0438 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u043d\u044b\u0435 \u0434\u0435\u043b\u0430 \u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0439 \u0438 \u043c\u0435\u043d\u044c\u0448\u0435, \u0447\u0435\u043c \u043e\u0431\u044b\u0447\u043d\u043e, \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0433\u0440\u0430\u0436\u0434\u0430\u043d, \u043e\u0431\u044a\u044f\u0441\u043d\u044f\u043b \u0440\u0443\u043a\u043e\u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c \u043f\u0440\u043e\u0435\u043a\u0442\u0430 \u00ab\u0424\u0435\u0434\u0440\u0435\u0441\u0443\u0440\u0441\u00bb \u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u042e\u0445\u043d\u0438\u043d.[SEP]", "example_title": "\u041d\u043e\u0432\u043e\u0441\u0442\u0438"}]} | IlyaGusev/rubert_ext_sum_gazeta | null | [
"transformers",
"pytorch",
"bert",
"token-classification",
"summarization",
"t5",
"ru",
"dataset:IlyaGusev/gazeta",
"license:apache-2.0",
"autotrain_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
summarization | transformers |
# RuBertTelegramHeadlines
## Model description
Example model for [Headline generation competition](https://competitions.codalab.org/competitions/29905)
Based on [RuBERT](http://docs.deeppavlov.ai/en/master/features/models/bert.html) model
## Intended uses & limitations
#### How to use
```python
from transformers import AutoTokenizer, EncoderDecoderModel
model_name = "IlyaGusev/rubert_telegram_headlines"
tokenizer = AutoTokenizer.from_pretrained(model_name, do_lower_case=False, do_basic_tokenize=False, strip_accents=False)
model = EncoderDecoderModel.from_pretrained(model_name)
article_text = "..."
input_ids = tokenizer(
[article_text],
add_special_tokens=True,
max_length=256,
padding="max_length",
truncation=True,
return_tensors="pt",
)["input_ids"]
output_ids = model.generate(
input_ids=input_ids,
max_length=64,
no_repeat_ngram_size=3,
num_beams=10,
top_p=0.95
)[0]
headline = tokenizer.decode(output_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True)
print(headline)
```
## Training data
- Dataset: [ru_all_split.tar.gz](https://www.dropbox.com/s/ykqk49a8avlmnaf/ru_all_split.tar.gz)
## Training procedure
```python
import random
import torch
from torch.utils.data import Dataset
from tqdm.notebook import tqdm
from transformers import BertTokenizer, EncoderDecoderModel, Trainer, TrainingArguments, logging
def convert_to_tensors(
tokenizer,
text,
max_text_tokens_count,
max_title_tokens_count = None,
title = None
):
inputs = tokenizer(
text,
add_special_tokens=True,
max_length=max_text_tokens_count,
padding="max_length",
truncation=True
)
result = {
"input_ids": torch.tensor(inputs["input_ids"]),
"attention_mask": torch.tensor(inputs["attention_mask"]),
}
if title is not None:
outputs = tokenizer(
title,
add_special_tokens=True,
max_length=max_title_tokens_count,
padding="max_length",
truncation=True
)
decoder_input_ids = torch.tensor(outputs["input_ids"])
decoder_attention_mask = torch.tensor(outputs["attention_mask"])
labels = decoder_input_ids.clone()
labels[decoder_attention_mask == 0] = -100
result.update({
"labels": labels,
"decoder_input_ids": decoder_input_ids,
"decoder_attention_mask": decoder_attention_mask
})
return result
class GetTitleDataset(Dataset):
def __init__(
self,
original_records,
sample_rate,
tokenizer,
max_text_tokens_count,
max_title_tokens_count
):
self.original_records = original_records
self.sample_rate = sample_rate
self.tokenizer = tokenizer
self.max_text_tokens_count = max_text_tokens_count
self.max_title_tokens_count = max_title_tokens_count
self.records = []
for record in tqdm(original_records):
if random.random() > self.sample_rate:
continue
tensors = convert_to_tensors(
tokenizer=tokenizer,
title=record["title"],
text=record["text"],
max_title_tokens_count=self.max_title_tokens_count,
max_text_tokens_count=self.max_text_tokens_count
)
self.records.append(tensors)
def __len__(self):
return len(self.records)
def __getitem__(self, index):
return self.records[index]
def train(
train_records,
val_records,
pretrained_model_path,
train_sample_rate=1.0,
val_sample_rate=1.0,
output_model_path="models",
checkpoint=None,
max_text_tokens_count=256,
max_title_tokens_count=64,
batch_size=8,
logging_steps=1000,
eval_steps=10000,
save_steps=10000,
learning_rate=0.00003,
warmup_steps=2000,
num_train_epochs=3
):
logging.set_verbosity_info()
tokenizer = BertTokenizer.from_pretrained(
pretrained_model_path,
do_lower_case=False,
do_basic_tokenize=False,
strip_accents=False
)
train_dataset = GetTitleDataset(
train_records,
train_sample_rate,
tokenizer,
max_text_tokens_count=max_text_tokens_count,
max_title_tokens_count=max_title_tokens_count
)
val_dataset = GetTitleDataset(
val_records,
val_sample_rate,
tokenizer,
max_text_tokens_count=max_text_tokens_count,
max_title_tokens_count=max_title_tokens_count
)
model = EncoderDecoderModel.from_encoder_decoder_pretrained(pretrained_model_path, pretrained_model_path)
training_args = TrainingArguments(
output_dir=output_model_path,
per_device_train_batch_size=batch_size,
per_device_eval_batch_size=batch_size,
do_train=True,
do_eval=True,
overwrite_output_dir=False,
logging_steps=logging_steps,
eval_steps=eval_steps,
evaluation_strategy="steps",
save_steps=save_steps,
learning_rate=learning_rate,
warmup_steps=warmup_steps,
num_train_epochs=num_train_epochs,
max_steps=-1,
save_total_limit=1,
)
trainer = Trainer(
model=model,
args=training_args,
train_dataset=train_dataset,
eval_dataset=val_dataset
)
trainer.train(checkpoint)
model.save_pretrained(output_model_path)
``` | {"language": ["ru"], "license": "apache-2.0", "tags": ["summarization"], "inference": {"parameters": {"no_repeat_ngram_size": 4}}} | IlyaGusev/rubert_telegram_headlines | null | [
"transformers",
"pytorch",
"encoder-decoder",
"text2text-generation",
"summarization",
"ru",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# RuBERTConv Toxic Classifier
## Model description
Based on [rubert-base-cased-conversational](https://huggingface.co/DeepPavlov/rubert-base-cased-conversational) model
## Intended uses & limitations
#### How to use
Colab: [link](https://colab.research.google.com/drive/1veKO9hke7myxKigZtZho_F-UM2fD9kp8)
```python
from transformers import pipeline
model_name = "IlyaGusev/rubertconv_toxic_clf"
pipe = pipeline("text-classification", model=model_name, tokenizer=model_name, framework="pt")
text = "Ты придурок из интернета"
pipe([text])
```
## Training data
Datasets:
- [2ch]( https://www.kaggle.com/blackmoon/russian-language-toxic-comments)
- [Odnoklassniki](https://www.kaggle.com/alexandersemiletov/toxic-russian-comments)
- [Toloka Persona Chat Rus](https://toloka.ai/ru/datasets)
- [Koziev's Conversations](https://github.com/Koziev/NLP_Datasets/blob/master/Conversations/Data) with [toxic words vocabulary](https://www.dropbox.com/s/ou6lx03b10yhrfl/bad_vocab.txt.tar.gz)
Augmentations:
- ё -> е
- Remove or add "?" or "!"
- Fix CAPS
- Concatenate toxic and non-toxic texts
- Concatenate two non-toxic texts
- Add toxic words from vocabulary
- Add typos
- Mask toxic words with "*", "@", "$"
## Training procedure
TBA | {"language": ["ru"], "license": "apache-2.0", "tags": ["text-classification"]} | IlyaGusev/rubertconv_toxic_clf | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"ru",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
token-classification | transformers |
# RuBERTConv Toxic Editor
## Model description
Tagging model for detoxification based on [rubert-base-cased-conversational](https://huggingface.co/DeepPavlov/rubert-base-cased-conversational).
4 possible classes:
- Equal = save tokens
- Replace = replace tokens with mask
- Delete = remove tokens
- Insert = insert mask before tokens
Use in pair with [mask filler](https://huggingface.co/IlyaGusev/sber_rut5_filler).
## Intended uses & limitations
#### How to use
Colab: [link](https://colab.research.google.com/drive/1NUSO1QGlDgD-IWXa2SpeND089eVxrCJW)
```python
import torch
from transformers import AutoTokenizer, pipeline
tagger_model_name = "IlyaGusev/rubertconv_toxic_editor"
device = "cuda" if torch.cuda.is_available() else "cpu"
device_num = 0 if device == "cuda" else -1
tagger_pipe = pipeline(
"token-classification",
model=tagger_model_name,
tokenizer=tagger_model_name,
framework="pt",
device=device_num,
aggregation_strategy="max"
)
text = "..."
tagger_predictions = tagger_pipe([text], batch_size=1)
sample_predictions = tagger_predictions[0]
print(sample_predictions)
```
## Training data
- Dataset: [russe_detox_2022](https://github.com/skoltech-nlp/russe_detox_2022/tree/main/data)
## Training procedure
- Parallel corpus convertion: [compute_tags.py](https://github.com/IlyaGusev/rudetox/blob/main/rudetox/marker/compute_tags.py)
- Training script: [train.py](https://github.com/IlyaGusev/rudetox/blob/main/rudetox/marker/train.py)
- Pipeline step: [dvc.yaml, train_marker](https://github.com/IlyaGusev/rudetox/blob/main/dvc.yaml#L367)
## Eval results
TBA | {"language": ["ru"], "license": "apache-2.0", "tags": ["token-classification"], "widget": [{"text": "\u0401\u043f\u0442\u0430, \u043c\u0435\u043d\u044f \u0437\u043e\u0432\u0443\u0442 \u043f\u0440\u0438\u0434\u0443\u0440\u043e\u043a \u0438 \u044f \u0436\u0438\u0432\u0443 \u0432 \u0436\u043e\u043f\u0435"}]} | IlyaGusev/rubertconv_toxic_editor | null | [
"transformers",
"pytorch",
"bert",
"token-classification",
"ru",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
summarization | transformers |
# RuGPT3MediumSumGazeta
## Model description
This is the model for abstractive summarization for Russian based on [rugpt3medium_based_on_gpt2](https://huggingface.co/sberbank-ai/rugpt3medium_based_on_gpt2).
## Intended uses & limitations
#### How to use
Colab: [link](https://colab.research.google.com/drive/1eR-ev0Y5ISWIwGnzYYoHyGMaSIUz8GTN)
```python
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
model_name = "IlyaGusev/rugpt3medium_sum_gazeta"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)
article_text = "..."
text_tokens = tokenizer(
article_text,
max_length=600,
add_special_tokens=False,
padding=False,
truncation=True
)["input_ids"]
input_ids = text_tokens + [tokenizer.sep_token_id]
input_ids = torch.LongTensor([input_ids])
output_ids = model.generate(
input_ids=input_ids,
no_repeat_ngram_size=4
)
summary = tokenizer.decode(output_ids[0], skip_special_tokens=False)
summary = summary.split(tokenizer.sep_token)[1]
summary = summary.split(tokenizer.eos_token)[0]
print(summary)
```
## Training data
- Dataset: [Gazeta](https://huggingface.co/datasets/IlyaGusev/gazeta)
## Training procedure
- Training script: [train.py](https://github.com/IlyaGusev/summarus/blob/master/external/hf_scripts/train.py)
- Config: [gpt_training_config.json](https://github.com/IlyaGusev/summarus/blob/master/external/hf_scripts/configs/gpt_training_config.json)
## Eval results
* Train dataset: **Gazeta v1 train**
* Test dataset: **Gazeta v1 test**
* Source max_length: **600**
* Target max_length: **200**
* no_repeat_ngram_size: **4**
* num_beams: **5**
| Model | R-1-f | R-2-f | R-L-f | chrF | METEOR | BLEU | Avg char length |
|:--------------------------|:------|:------|:------|:-------|:-------|:-----|:-----|
| [mbart_ru_sum_gazeta](https://huggingface.co/IlyaGusev/mbart_ru_sum_gazeta) | **32.4** | 14.3 | 28.0 | 39.7 | **26.4** | 12.1 | 371 |
| [rut5_base_sum_gazeta](https://huggingface.co/IlyaGusev/rut5_base_sum_gazeta) | 32.2 | **14.4** | **28.1** | **39.8** | 25.7 | **12.3** | 330 |
| [rugpt3medium_sum_gazeta](https://huggingface.co/IlyaGusev/rugpt3medium_sum_gazeta) | 26.2 | 7.7 | 21.7 | 33.8 | 18.2 | 4.3 | 244 |
* Train dataset: **Gazeta v1 train**
* Test dataset: **Gazeta v2 test**
* Source max_length: **600**
* Target max_length: **200**
* no_repeat_ngram_size: **4**
* num_beams: **5**
| Model | R-1-f | R-2-f | R-L-f | chrF | METEOR | BLEU | Avg char length |
|:--------------------------|:------|:------|:------|:-------|:-------|:-----|:-----|
| [mbart_ru_sum_gazeta](https://huggingface.co/IlyaGusev/mbart_ru_sum_gazeta) | **28.7** | **11.1** | 24.4 | **37.3** | **22.7** | **9.4** | 373 |
| [rut5_base_sum_gazeta](https://huggingface.co/IlyaGusev/rut5_base_sum_gazeta) | 28.6 | **11.1** | **24.5** | 37.2 | 22.0 | **9.4** | 331 |
| [rugpt3medium_sum_gazeta](https://huggingface.co/IlyaGusev/rugpt3medium_sum_gazeta) | 24.1 | 6.5 | 19.8 | 32.1 | 16.3 | 3.6 | 242 |
Evaluation script: [evaluate.py](https://github.com/IlyaGusev/summarus/blob/master/evaluate.py)
Flags: --language ru --tokenize-after --lower
| {"language": ["ru"], "license": ["apache-2.0"], "tags": ["causal-lm", "summarization"], "datasets": ["IlyaGusev/gazeta"], "inference": false, "widget": [{"text": "\u0412\u044b\u0441\u043e\u0442\u0430 \u0431\u0430\u0448\u043d\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 324 \u043c\u0435\u0442\u0440\u0430 (1063 \u0444\u0443\u0442\u0430), \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e \u0442\u0430\u043a\u0430\u044f \u0436\u0435 \u0432\u044b\u0441\u043e\u0442\u0430, \u043a\u0430\u043a \u0443 81-\u044d\u0442\u0430\u0436\u043d\u043e\u0433\u043e \u0437\u0434\u0430\u043d\u0438\u044f, \u0438 \u0441\u0430\u043c\u043e\u0435 \u0432\u044b\u0441\u043e\u043a\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u0432 \u041f\u0430\u0440\u0438\u0436\u0435. \u0415\u0433\u043e \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u0438\u0435 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u043e, \u0440\u0430\u0437\u043c\u0435\u0440\u043e\u043c 125 \u043c\u0435\u0442\u0440\u043e\u0432 (410 \u0444\u0443\u0442\u043e\u0432) \u0441 \u043b\u044e\u0431\u043e\u0439 \u0441\u0442\u043e\u0440\u043e\u043d\u044b. \u0412\u043e \u0432\u0440\u0435\u043c\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u043f\u0440\u0435\u0432\u0437\u043e\u0448\u043b\u0430 \u043c\u043e\u043d\u0443\u043c\u0435\u043d\u0442 \u0412\u0430\u0448\u0438\u043d\u0433\u0442\u043e\u043d\u0430, \u0441\u0442\u0430\u0432 \u0441\u0430\u043c\u044b\u043c \u0432\u044b\u0441\u043e\u043a\u0438\u043c \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u043c \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435\u043c \u0432 \u043c\u0438\u0440\u0435, \u0438 \u044d\u0442\u043e\u0442 \u0442\u0438\u0442\u0443\u043b \u043e\u043d\u0430 \u0443\u0434\u0435\u0440\u0436\u0438\u0432\u0430\u043b\u0430 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 41 \u0433\u043e\u0434\u0430 \u0434\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u043e \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u0432 \u041d\u044c\u044e-\u0419\u043e\u0440\u043a\u0435 \u0432 1930 \u0433\u043e\u0434\u0443. \u042d\u0442\u043e \u043f\u0435\u0440\u0432\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0434\u043e\u0441\u0442\u0438\u0433\u043b\u043e \u0432\u044b\u0441\u043e\u0442\u044b 300 \u043c\u0435\u0442\u0440\u043e\u0432. \u0418\u0437-\u0437\u0430 \u0434\u043e\u0431\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0432\u0435\u0449\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0430\u043d\u0442\u0435\u043d\u043d\u044b \u043d\u0430 \u0432\u0435\u0440\u0448\u0438\u043d\u0435 \u0431\u0430\u0448\u043d\u0438 \u0432 1957 \u0433\u043e\u0434\u0443 \u043e\u043d\u0430 \u0441\u0435\u0439\u0447\u0430\u0441 \u0432\u044b\u0448\u0435 \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u043d\u0430 5,2 \u043c\u0435\u0442\u0440\u0430 (17 \u0444\u0443\u0442\u043e\u0432). \u0417\u0430 \u0438\u0441\u043a\u043b\u044e\u0447\u0435\u043d\u0438\u0435\u043c \u043f\u0435\u0440\u0435\u0434\u0430\u0442\u0447\u0438\u043a\u043e\u0432, \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0432\u0442\u043e\u0440\u043e\u0439 \u0441\u0430\u043c\u043e\u0439 \u0432\u044b\u0441\u043e\u043a\u043e\u0439 \u043e\u0442\u0434\u0435\u043b\u044c\u043d\u043e \u0441\u0442\u043e\u044f\u0449\u0435\u0439 \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u043e\u0439 \u0432\u043e \u0424\u0440\u0430\u043d\u0446\u0438\u0438 \u043f\u043e\u0441\u043b\u0435 \u0432\u0438\u0430\u0434\u0443\u043a\u0430 \u041c\u0438\u0439\u043e.<s>", "example_title": "\u0412\u0438\u043a\u0438\u043f\u0435\u0434\u0438\u044f"}]} | IlyaGusev/rugpt3medium_sum_gazeta | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"causal-lm",
"summarization",
"ru",
"dataset:IlyaGusev/gazeta",
"license:apache-2.0",
"autotrain_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
summarization | transformers |
# RuT5TelegramHeadlines
## Model description
Based on [rut5-base](https://huggingface.co/cointegrated/rut5-base) model
## Intended uses & limitations
#### How to use
```python
from transformers import AutoTokenizer, T5ForConditionalGeneration
model_name = "IlyaGusev/rut5_base_headline_gen_telegram"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)
article_text = "..."
input_ids = tokenizer(
[article_text],
max_length=600,
add_special_tokens=True,
padding="max_length",
truncation=True,
return_tensors="pt"
)["input_ids"]
output_ids = model.generate(
input_ids=input_ids
)[0]
headline = tokenizer.decode(output_ids, skip_special_tokens=True)
print(headline)
```
## Training data
- Dataset: [ru_all_split.tar.gz](https://www.dropbox.com/s/ykqk49a8avlmnaf/ru_all_split.tar.gz)
## Training procedure
- Training script: [train.py](https://github.com/IlyaGusev/summarus/blob/master/external/hf_scripts/train.py) | {"language": ["ru"], "license": "apache-2.0", "tags": ["summarization"], "widget": [{"text": "\u041a\u043e\u043c\u0438\u0441\u0441\u0438\u044f \u0421\u043e\u0432\u0435\u0442\u0430 \u0424\u0435\u0434\u0435\u0440\u0430\u0446\u0438\u0438 \u043f\u043e \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u043e\u0439 \u043f\u043e\u043b\u0438\u0442\u0438\u043a\u0435 \u0438 \u0432\u0437\u0430\u0438\u043c\u043e\u0434\u0435\u0439\u0441\u0442\u0432\u0438\u044e \u0441\u043e \u0421\u041c\u0418 \u0441\u043e\u0432\u043c\u0435\u0441\u0442\u043d\u043e \u0441 \u0437\u0430\u0438\u043d\u0442\u0435\u0440\u0435\u0441\u043e\u0432\u0430\u043d\u043d\u044b\u043c\u0438 \u0432\u0435\u0434\u043e\u043c\u0441\u0442\u0432\u0430\u043c\u0438 \u0434\u0443\u043c\u0430\u0435\u0442 \u043d\u0430\u0434 \u0440\u0430\u0437\u0440\u0430\u0431\u043e\u0442\u043a\u043e\u0439 \u043d\u0430\u0446\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u043e\u0433\u043e \u0437\u0430\u043a\u043e\u043d\u043e\u0434\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u0432 \u043e\u0431\u043b\u0430\u0441\u0442\u0438 \u043d\u0430\u043b\u043e\u0433\u043e\u043e\u0431\u043b\u043e\u0436\u0435\u043d\u0438\u044f \u0433\u043b\u043e\u0431\u0430\u043b\u044c\u043d\u044b\u0445 \u0438\u043d\u0442\u0435\u0440\u043d\u0435\u0442-\u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0439, \u0442\u0430\u043a\u0438\u0445 \u043a\u0430\u043a Google \u0438 Facebook. \u041e\u0431 \u044d\u0442\u043e\u043c \u0441\u043e\u043e\u0431\u0449\u0438\u043b \u0422\u0410\u0421\u0421 \u043f\u0440\u0435\u0434\u0441\u0435\u0434\u0430\u0442\u0435\u043b\u044c \u043a\u043e\u043c\u0438\u0441\u0441\u0438\u0438 \u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u041f\u0443\u0448\u043a\u043e\u0432. \u00ab\u0412 \u043d\u0430\u0441\u0442\u043e\u044f\u0449\u0435\u0435 \u0432\u0440\u0435\u043c\u044f \u043f\u043e \u043b\u0438\u043d\u0438\u0438 \u041e\u042d\u0421\u0420 [\u041e\u0440\u0433\u0430\u043d\u0438\u0437\u0430\u0446\u0438\u044f \u044d\u043a\u043e\u043d\u043e\u043c\u0438\u0447\u0435\u0441\u043a\u043e\u0433\u043e \u0441\u043e\u0442\u0440\u0443\u0434\u043d\u0438\u0447\u0435\u0441\u0442\u0432\u0430 \u0438 \u0440\u0430\u0437\u0432\u0438\u0442\u0438\u044f] \u0432\u0435\u0434\u0435\u0442\u0441\u044f \u0440\u0430\u0437\u0440\u0430\u0431\u043e\u0442\u043a\u0430 \u043c\u0435\u0436\u0434\u0443\u043d\u0430\u0440\u043e\u0434\u043d\u043e\u0439 \u043a\u043e\u043d\u0432\u0435\u043d\u0446\u0438\u0438, \u043e\u0434\u043d\u0430\u043a\u043e \u0440\u0430\u0431\u043e\u0442\u0430 \u043d\u0430\u0434 \u043d\u0435\u0439 \u0435\u0449\u0435 \u043d\u0435 \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0430. \u0412 \u044d\u0442\u0438\u0445 \u0443\u0441\u043b\u043e\u0432\u0438\u044f\u0445 \u043c\u044b \u0438\u0441\u0445\u043e\u0434\u0438\u043c \u0438\u0437 \u0442\u043e\u0433\u043e, \u0447\u0442\u043e \u0441\u0430\u043c\u0430\u044f \u0440\u0430\u0437\u0443\u043c\u043d\u0430\u044f \u043f\u043e\u0437\u0438\u0446\u0438\u044f - \u043d\u0430\u0447\u0430\u0442\u044c \u0440\u0430\u0437\u0440\u0430\u0431\u043e\u0442\u043a\u0443 \u043d\u0430\u0446\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u043e\u0433\u043e \u0437\u0430\u043a\u043e\u043d\u043e\u0434\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430, \u043d\u0435 \u0434\u043e\u0436\u0438\u0434\u0430\u044f\u0441\u044c \u043a\u043e\u043d\u0432\u0435\u043d\u0446\u0438\u0438\u00bb, \u2014 \u043f\u043e\u044f\u0441\u043d\u0438\u043b \u0441\u0435\u043d\u0430\u0442\u043e\u0440. \u041f\u0443\u0448\u043a\u043e\u0432 \u043e\u0442\u043c\u0435\u0442\u0438\u043b, \u0447\u0442\u043e \u043f\u043e \u0442\u0430\u043a\u043e\u043c\u0443 \u043f\u0443\u0442\u0438 \u043f\u043e\u0448\u043b\u0438 \u0435\u0449\u0435 \u043d\u0435\u0441\u043a\u043e\u043b\u044c\u043a\u043e \u0441\u0442\u0440\u0430\u043d, \u0432 \u0447\u0438\u0441\u043b\u0435 \u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0424\u0440\u0430\u043d\u0446\u0438\u044f, \u0410\u0432\u0441\u0442\u0440\u0430\u043b\u0438\u044f \u0438 \u0422\u0443\u0440\u0446\u0438\u044f. \u041f\u043e \u0435\u0433\u043e \u0441\u043b\u043e\u0432\u0430\u043c, \u0432 \u0420\u043e\u0441\u0441\u0438\u0438 \u0432\u0430\u0436\u043d\u043e \u0437\u0430\u0434\u0435\u0439\u0441\u0442\u0432\u043e\u0432\u0430\u0442\u044c \u0432 \u044d\u0442\u043e\u0439 \u0440\u0430\u0431\u043e\u0442\u0435 \u041c\u0438\u043d\u0444\u0438\u043d, \u0424\u041d\u0421, \u041c\u0418\u0414 \u0420\u0424 \u0438 \u0420\u043e\u0441\u043a\u043e\u043c\u043d\u0430\u0434\u0437\u043e\u0440. \u00ab\u0418\u043d\u0442\u0435\u0440\u043d\u0435\u0442-\u043f\u043b\u0430\u0442\u0444\u043e\u0440\u043c\u044b \u043d\u0435 \u0444\u0438\u0433\u0443\u0440\u0438\u0440\u0443\u044e\u0442 \u0443 \u043d\u0430\u0441 \u0441\u0435\u0439\u0447\u0430\u0441 \u043a\u0430\u043a \u043e\u0442\u0434\u0435\u043b\u044c\u043d\u044b\u0439 \u043e\u0431\u044a\u0435\u043a\u0442 \u043d\u0430\u043b\u043e\u0433\u043e\u043e\u0431\u043b\u043e\u0436\u0435\u043d\u0438\u044f. \u041a\u043e\u0433\u0434\u0430 \u043e\u043d\u0438 \u043e\u0442\u043a\u0440\u043e\u044e\u0442 \u0432 \u0420\u043e\u0441\u0441\u0438\u0438 \u0441\u0432\u043e\u0438 \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u0432 \u0440\u0430\u043c\u043a\u0430\u0445 \u0437\u0430\u043a\u043e\u043d\u0430 \u043e \u00ab\u043f\u0440\u0438\u0437\u0435\u043c\u043b\u0435\u043d\u0438\u0438\u00bb, \u0432\u043e\u0437\u043d\u0438\u043a\u043d\u0435\u0442 \u0432\u043e\u043f\u0440\u043e\u0441: \u043a\u0430\u043a \u0438\u0445 \u043e\u0444\u0438\u0446\u0438\u0430\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u0438\u0441\u0443\u0442\u0441\u0442\u0432\u0438\u0435 \u043d\u0430 \u0442\u0435\u0440\u0440\u0438\u0442\u043e\u0440\u0438\u0438 \u0420\u043e\u0441\u0441\u0438\u0438, \u043a\u043e\u0442\u043e\u0440\u043e\u0433\u043e \u0441\u0435\u0439\u0447\u0430\u0441 \u043d\u0435\u0442, \u0431\u0443\u0434\u0435\u0442 \u0441\u043e\u043e\u0442\u043d\u043e\u0441\u0438\u0442\u044c\u0441\u044f \u0441 \u043d\u0430\u0448\u0438\u043c \u043d\u0430\u043b\u043e\u0433\u043e\u0432\u044b\u043c \u0440\u0435\u0436\u0438\u043c\u043e\u043c. \u041c\u044b \u0441\u0435\u0439\u0447\u0430\u0441 \u043f\u0440\u043e\u0434\u0443\u043c\u044b\u0432\u0430\u0435\u043c, \u043a\u0430\u043a \u0443\u0441\u0442\u0430\u043d\u043e\u0432\u0438\u0442\u044c \u044d\u0442\u0443 \u0432\u0437\u0430\u0438\u043c\u043e\u0441\u0432\u044f\u0437\u044c\u00bb, \u2014 \u0441\u043a\u0430\u0437\u0430\u043b \u041f\u0443\u0448\u043a\u043e\u0432, \u0434\u043e\u0431\u0430\u0432\u043b\u044f\u044f, \u0447\u0442\u043e \u0432\u043e\u043f\u0440\u043e\u0441 \u0432\u043d\u0435\u0441\u0435\u043d\u0438\u044f \u0438\u0437\u043c\u0435\u043d\u0435\u043d\u0438\u0439 \u0432 \u0440\u043e\u0441\u0441\u0438\u0439\u0441\u043a\u043e\u0435 \u0437\u0430\u043a\u043e\u043d\u043e\u0434\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u043e \u0432 \u0447\u0430\u0441\u0442\u0438 \u043d\u0430\u043b\u043e\u0433\u043e\u043e\u0431\u043b\u043e\u0436\u0435\u043d\u0438\u044f \u043a\u0440\u0443\u043f\u043d\u044b\u0445 IT-\u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0439 \u043d\u0430\u0445\u043e\u0434\u0438\u0442\u0441\u044f \u00ab\u043d\u0430 \u043f\u0435\u0440\u0432\u043e\u0439 \u0441\u0442\u0430\u0434\u0438\u0438 \u0438\u0437\u0443\u0447\u0435\u043d\u0438\u044f\u00bb. \u0421\u0430\u043c \u0441\u0435\u043d\u0430\u0442\u043e\u0440 \u0432\u044b\u0441\u0442\u0443\u043f\u0430\u0435\u0442 \u0437\u0430 \u0432\u0432\u0435\u0434\u0435\u043d\u0438\u0435 \u043f\u0440\u043e\u0433\u0440\u0435\u0441\u0441\u0438\u0432\u043d\u043e\u0439 \u0441\u0442\u0430\u0432\u043a\u0438 \u043d\u0430\u043b\u043e\u0433\u0430 \u0432 \u0437\u0430\u0432\u0438\u0441\u0438\u043c\u043e\u0441\u0442\u0438 \u043e\u0442 \u043f\u0440\u0438\u0431\u044b\u043b\u0438 IT-\u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0439 \u043d\u0430 \u0442\u0435\u0440\u0440\u0438\u0442\u043e\u0440\u0438\u0438 \u0441\u0442\u0440\u0430\u043d\u044b. \u041f\u0440\u0438 \u044d\u0442\u043e\u043c, \u043f\u043e\u0434\u0447\u0435\u0440\u043a\u043d\u0443\u043b \u043e\u043d, \u043e\u0434\u043d\u0430 \u0438\u0437 \u0437\u0430\u0434\u0430\u0447 \u043d\u0430\u0446\u0438\u043e\u043d\u0430\u043b\u044c\u043d\u043e\u0439 \u0441\u0438\u0441\u0442\u0435\u043c\u044b \u043d\u0430\u043b\u043e\u0433\u043e\u043e\u0431\u043b\u043e\u0436\u0435\u043d\u0438\u044f \u0431\u0443\u0434\u0435\u0442 \u0437\u0430\u043a\u043b\u044e\u0447\u0430\u0442\u044c\u0441\u044f \u0432 \u043f\u043e\u0434\u0441\u0447\u0435\u0442\u0435 \u043d\u0430\u043b\u043e\u0433\u043e\u043e\u0431\u043b\u0430\u0433\u0430\u0435\u043c\u043e\u0439 \u0431\u0430\u0437\u044b. \u0421\u0435\u0439\u0447\u0430\u0441 \u043a\u0440\u0443\u043f\u043d\u044b\u0435 \u0418\u0422-\u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0438 \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0442\u0447\u0438\u0442\u044b\u0432\u0430\u044e\u0442\u0441\u044f \u043e \u0441\u0432\u043e\u0435\u0439 \u043f\u0440\u0438\u0431\u044b\u043b\u0438. \u041e\u0434\u043d\u0430\u043a\u043e \u0420\u043e\u0441\u0441\u0438\u0438 \u043d\u0443\u0436\u043d\u0430 \u0441\u043e\u0431\u0441\u0442\u0432\u0435\u043d\u043d\u0430\u044f \u0441\u0438\u0441\u0442\u0435\u043c\u0430 \u043f\u043e\u0434\u0441\u0447\u0435\u0442\u0430 \u0438\u0445 \u0434\u043e\u0445\u043e\u0434\u043e\u0432, \u043a\u043e\u0442\u043e\u0440\u0430\u044f \u043f\u043e\u0437\u0432\u043e\u043b\u0438\u0442 \u043e\u043f\u0440\u0435\u0434\u0435\u043b\u0438\u0442\u044c \u0438\u0445 \u00ab\u0440\u0435\u0430\u043b\u044c\u043d\u0443\u044e \u043d\u0430\u043b\u043e\u0433\u043e\u043e\u0431\u043b\u0430\u0433\u0430\u0435\u043c\u0443\u044e \u0431\u0430\u0437\u0443\u00bb, \u0441\u0447\u0438\u0442\u0430\u0435\u0442 \u041f\u0443\u0448\u043a\u043e\u0432. (https://www.gazeta.ru/tech/news/2021/12/17/n_17024239.shtml)", "example_title": "\u041d\u043e\u0432\u043e\u0441\u0442\u044c \u043f\u0440\u043e \u043d\u0430\u043b\u043e\u0433\u0438 \u0432 IT"}, {"text": "\u041f\u0435\u0440\u0432\u0443\u044e \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0443, \u0443 \u043a\u043e\u0442\u043e\u0440\u043e\u0439 \u0431\u043e\u043b\u0435\u0435 \u0442\u044b\u0441\u044f\u0447\u0438 \u043d\u043e\u0433, \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0438\u043b\u0438 \u0432 \u0430\u0432\u0441\u0442\u0440\u0430\u043b\u0438\u0439\u0441\u043a\u0438\u0445 \u043f\u0435\u0449\u0435\u0440\u0430\u0445 \u0431\u0438\u043e\u043b\u043e\u0433\u0438, \u0438\u0437\u0443\u0447\u0430\u0432\u0448\u0438\u0435 \u0442\u0430\u043c \u043f\u043e\u0434\u0437\u0435\u043c\u043d\u044b\u0435 \u0432\u043e\u0434\u044b. \u041f\u0440\u0435\u0434\u044b\u0434\u0443\u0449\u0435\u0439 \u0440\u0435\u043a\u043e\u0440\u0434\u0441\u043c\u0435\u043d\u043a\u043e\u0439 \u043f\u043e \u043a\u043e\u043b\u0438\u0447\u0435\u0441\u0442\u0432\u0443 \u043d\u043e\u0433 \u0431\u044b\u043b\u0430 700-\u043d\u043e\u0433\u0430\u044f \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0430. \u041d\u043e\u0432\u044b\u0439 \u0432\u0438\u0434 \u0438\u043c\u0435\u0435\u0442 \u0434\u043b\u0438\u043d\u043d\u043e\u0435 \u0442\u043e\u043d\u043a\u043e\u0435 \u0442\u0435\u043b\u043e, \u043f\u043e\u0445\u043e\u0436\u0435\u0435 \u043d\u0430 \u043d\u0438\u0442\u044c, \u0438 \u0431\u043e\u043b\u044c\u0448\u043e\u0435 \u043a\u043e\u043b\u0438\u0447\u0435\u0441\u0442\u0432\u043e \u043a\u043e\u043d\u0435\u0447\u043d\u043e\u0441\u0442\u0435\u0439, \u043f\u043e-\u0432\u0438\u0434\u0438\u043c\u043e\u043c\u0443, \u0434\u0430\u0435\u0442 \u043f\u0440\u0435\u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0430 \u0434\u043b\u044f \u0431\u044b\u0441\u0442\u0440\u043e\u0433\u043e \u043f\u0435\u0440\u0435\u043c\u0435\u0449\u0435\u043d\u0438\u044f \u0438 \u043f\u0440\u043e\u043d\u0438\u043a\u043d\u043e\u0432\u0435\u043d\u0438\u044f \u0432 \u0442\u0440\u0443\u0434\u043d\u043e\u0434\u043e\u0441\u0442\u0443\u043f\u043d\u044b\u0435 \u043c\u0435\u0441\u0442\u0430 \u2014 \u0443\u0447\u0435\u043d\u044b\u0435 \u043f\u043e\u043b\u0430\u0433\u0430\u044e\u0442, \u0442\u0430\u043a\u0430\u044f \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0430 \u043c\u043e\u0436\u0435\u0442 \u0441\u043f\u043e\u043a\u043e\u0439\u043d\u043e \u043f\u0435\u0440\u0435\u043c\u0435\u0449\u0430\u0442\u044c\u0441\u044f \u043f\u043e \u0442\u0440\u0435\u0449\u0438\u043d\u0430\u043c \u0432 \u043a\u0430\u043c\u043d\u044f\u0445. \u0410\u0432\u0441\u0442\u0440\u0430\u043b\u0438\u044f \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u0430 \u0441\u0432\u043e\u0438\u043c\u0438 \u043e\u0433\u0440\u043e\u043c\u043d\u044b\u043c\u0438 \u0438 \u0436\u0443\u0442\u043a\u043e\u0432\u0430\u0442\u044b\u043c\u0438 \u0436\u0438\u0432\u043e\u0442\u043d\u044b\u043c\u0438 \u0432\u0440\u043e\u0434\u0435 25-\u0441\u0430\u043d\u0442\u0438\u043c\u0435\u0442\u0440\u043e\u0432\u044b\u0445 \u043f\u0430\u0443\u043a\u043e\u0432. \u0422\u0435\u043f\u0435\u0440\u044c \u0441\u043f\u0438\u0441\u043e\u043a \u043f\u0443\u0433\u0430\u044e\u0449\u0438\u0445 \u0447\u043b\u0435\u043d\u0438\u0441\u0442\u043e\u043d\u043e\u0433\u0438\u0445 \u043f\u043e\u043f\u043e\u043b\u043d\u0438\u043b\u0441\u044f \u0441\u0430\u043c\u043e\u0439 \u00ab\u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0433\u043e\u0439\u00bb \u0432 \u043c\u0438\u0440\u0435 \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u043e\u0439, \u0443 \u043a\u043e\u0442\u043e\u0440\u043e\u0439 \u0431\u043e\u043b\u0435\u0435 \u0442\u044b\u0441\u044f\u0447\u0438 \u043d\u043e\u0433. \u041d\u0435\u043e\u0431\u044b\u0447\u043d\u043e\u0435 \u0436\u0438\u0432\u043e\u0442\u043d\u043e\u0435 \u043e\u0431\u043d\u0430\u0440\u0443\u0436\u0438\u043b\u0430 \u0433\u0440\u0443\u043f\u043f\u0430 \u0438\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0435\u0439 \u0438\u0437 \u0410\u0432\u0441\u0442\u0440\u0430\u043b\u0438\u0438 \u0438 \u0421\u0428\u0410 \u0432 \u043f\u0435\u0449\u0435\u0440\u0430\u0445 \u043d\u0430 \u0437\u0430\u043f\u0430\u0434\u0435 \u0441\u0442\u0440\u0430\u043d\u044b. \u041f\u043e\u0434\u0440\u043e\u0431\u043d\u0435\u0435 \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0443 \u0443\u0447\u0435\u043d\u044b\u0435 \u043e\u043f\u0438\u0441\u0430\u043b\u0438 \u0432 \u0441\u0442\u0430\u0442\u044c\u0435 \u0432 \u0436\u0443\u0440\u043d\u0430\u043b\u0435 Scientific Reports. \u0418\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0438 \u0437\u0430\u043d\u0438\u043c\u0430\u043b\u0438\u0441\u044c \u043e\u0446\u0435\u043d\u043a\u043e\u0439 \u0432\u043e\u0437\u0434\u0435\u0439\u0441\u0442\u0432\u0438\u044f \u043f\u043e\u0434\u0437\u0435\u043c\u043d\u044b\u0445 \u0432\u043e\u0434 \u043d\u0430 \u043e\u043a\u0440\u0443\u0436\u0430\u044e\u0449\u0443\u044e \u0441\u0440\u0435\u0434\u0443 \u0432 \u0437\u043e\u043d\u0435 \u0434\u043e\u0431\u044b\u0447\u0438 \u043f\u043e\u043b\u0435\u0437\u043d\u044b\u0445 \u0438\u0441\u043a\u043e\u043f\u0430\u0435\u043c\u044b\u0445 \u043d\u0430 \u0437\u0430\u043f\u0430\u0434\u0435 \u0441\u0442\u0440\u0430\u043d\u044b, \u043a\u043e\u0433\u0434\u0430 \u043d\u0430\u0442\u043a\u043d\u0443\u043b\u0438\u0441\u044c \u043d\u0430 \u043d\u043e\u0432\u044b\u0439 \u0432\u0438\u0434 \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u0435\u043a. \u0412 \u043e\u0442\u043b\u0438\u0447\u0438\u0435 \u043e\u0442 \u0431\u043e\u043b\u044c\u0448\u0438\u043d\u0441\u0442\u0432\u0430 \u0441\u043e\u0440\u043e\u0434\u0438\u0447\u0435\u0439, \u0436\u0438\u0432\u0443\u0449\u0438\u0445 \u043d\u0430 \u043f\u043e\u0432\u0435\u0440\u0445\u043d\u043e\u0441\u0442\u0438, \u044d\u0442\u0438 \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0438 \u043e\u0431\u0438\u0442\u0430\u043b\u0438 \u0432 \u043f\u0435\u0449\u0435\u0440\u0430\u0445 \u043d\u0430 \u0433\u043b\u0443\u0431\u0438\u043d\u0435 \u0434\u043e 60 \u043c\u0435\u0442\u0440\u043e\u0432. \u041d\u043e\u0432\u044b\u0439 \u0432\u0438\u0434 \u0438\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0438 \u043d\u0430\u0437\u0432\u0430\u043b\u0438 Eumillipes persephone, \u0432 \u0447\u0435\u0441\u0442\u044c \u041f\u0435\u0440\u0441\u0435\u0444\u043e\u043d\u044b \u2014 \u0434\u0440\u0435\u0432\u043d\u0435\u0433\u0440\u0435\u0447\u0435\u0441\u043a\u043e\u0439 \u0431\u043e\u0433\u0438\u043d\u0438 \u043f\u043e\u0434\u0437\u0435\u043c\u043d\u043e\u0433\u043e \u043c\u0438\u0440\u0430. \u0423 \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0438 \u043e\u043a\u0430\u0437\u0430\u043b\u043e\u0441\u044c 1306 \u043d\u043e\u0433 \u2014 \u0431\u043e\u043b\u044c\u0448\u0435, \u0447\u0435\u043c \u0443 \u043b\u044e\u0431\u043e\u0433\u043e \u0434\u0440\u0443\u0433\u043e\u0433\u043e \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u043e\u0433\u043e \u0432\u0438\u0434\u0430. \u041f\u0440\u0435\u0434\u044b\u0434\u0443\u0449\u0435\u0439 \u0440\u0435\u043a\u043e\u0440\u0434\u0441\u043c\u0435\u043d\u043a\u043e\u0439 \u0431\u044b\u043b\u0430 \u043a\u0430\u043b\u0438\u0444\u043e\u0440\u043d\u0438\u0439\u0441\u043a\u0430\u044f Illacme plenipes, \u0443 \u043a\u043e\u0442\u043e\u0440\u043e\u0439 \u043d\u0430\u0441\u0447\u0438\u0442\u044b\u0432\u0430\u043b\u043e\u0441\u044c \u0434\u043e 750 \u043d\u043e\u0433. \u00ab\u042d\u0442\u0438 \u0436\u0438\u0432\u043e\u0442\u043d\u044b\u0435 \u0431\u044b\u043b\u0438 \u043d\u0430\u0441\u0442\u043e\u043b\u044c\u043a\u043e \u0443\u043d\u0438\u043a\u0430\u043b\u044c\u043d\u044b, \u2014 \u0433\u043e\u0432\u043e\u0440\u0438\u0442 \u0431\u0438\u043e\u043b\u043e\u0433 \u0411\u0440\u0443\u043d\u043e \u0411\u0443\u0437\u0430\u0442\u0442\u043e. \u2014 \u041a\u0430\u043a \u0442\u043e\u043b\u044c\u043a\u043e \u044f \u043f\u043e\u043d\u044f\u043b, \u043a\u0430\u043a\u043e\u0439 \u0434\u043b\u0438\u043d\u044b \u043e\u043d\u0438 \u0431\u044b\u043b\u0438... \u0421\u0442\u0430\u043b\u043e \u044f\u0441\u043d\u043e, \u0447\u0442\u043e \u044d\u0442\u043e \u0447\u0442\u043e-\u0442\u043e \u0441\u043e\u0432\u0435\u0440\u0448\u0435\u043d\u043d\u043e \u043d\u043e\u0432\u043e\u0435\u00bb. \u0423 \u0415. persephone \u043d\u0438\u0442\u0435\u0432\u0438\u0434\u043d\u043e\u0435 \u0442\u0435\u043b\u043e \u0434\u043b\u0438\u043d\u043e\u0439 \u043e\u043a\u043e\u043b\u043e 9,5 \u0441\u043c \u0438 \u0448\u0438\u0440\u0438\u043d\u043e\u0439 \u0432\u0441\u0435\u0433\u043e \u043c\u0438\u043b\u043b\u0438\u043c\u0435\u0442\u0440, \u0441\u043e\u0441\u0442\u043e\u044f\u0449\u0435\u0435 \u0438\u0437 330 \u0441\u0435\u0433\u043c\u0435\u043d\u0442\u043e\u0432, \u043a\u043e\u0440\u043e\u0442\u043a\u0438\u0435 \u043d\u043e\u0433\u0438 \u0438 \u043a\u043e\u043d\u0443\u0441\u043e\u043e\u0431\u0440\u0430\u0437\u043d\u0430\u044f \u0433\u043e\u043b\u043e\u0432\u0430. \u041a\u0430\u043a \u0438 \u0434\u0440\u0443\u0433\u0438\u0435 \u0436\u0438\u0432\u043e\u0442\u043d\u044b\u0435, \u0436\u0438\u0432\u0443\u0449\u0438\u0435 \u0432 \u043f\u043e\u0441\u0442\u043e\u044f\u043d\u043d\u043e\u0439 \u0442\u0435\u043c\u043d\u043e\u0442\u0435, \u044d\u0442\u0438 \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0438 \u0431\u043b\u0435\u0434\u043d\u044b \u0438 \u0441\u043b\u0435\u043f\u044b. \u042d\u043d\u0442\u043e\u043c\u043e\u043b\u043e\u0433 \u041f\u043e\u043b \u041c\u0430\u0440\u0435\u043a \u0441\u0440\u0430\u0432\u043d\u0438\u0432\u0430\u0435\u0442 \u0435\u0435 \u0441 \u0431\u0435\u043b\u043e\u0439 \u043d\u0438\u0442\u044c\u044e, \u0432\u044b\u0434\u0435\u0440\u043d\u0443\u0442\u043e\u0439 \u0438\u0437 \u0440\u0443\u0431\u0430\u0448\u043a\u0438. \u0427\u0442\u043e\u0431\u044b \u043f\u043e\u0441\u0447\u0438\u0442\u0430\u0442\u044c \u043a\u043e\u043b\u0438\u0447\u0435\u0441\u0442\u0432\u043e \u043d\u043e\u0433, \u0443\u0447\u0435\u043d\u044b\u043c \u043f\u0440\u0438\u0448\u043b\u043e\u0441\u044c \u0441\u043d\u0430\u0447\u0430\u043b\u0430 \u0441\u043d\u044f\u0442\u044c \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0443 \u0432 \u0432\u044b\u0441\u043e\u043a\u043e\u043c \u0440\u0430\u0437\u0440\u0435\u0448\u0435\u043d\u0438\u0438, \u0430 \u0437\u0430\u0442\u0435\u043c \u0437\u0430\u043a\u0440\u0430\u0448\u0438\u0432\u0430\u0442\u044c \u043d\u0430 \u0444\u043e\u0442\u043e \u043a\u0430\u0436\u0434\u044b\u0439 \u0434\u0435\u0441\u044f\u0442\u043e\u043a \u043d\u043e\u0433 \u0434\u0440\u0443\u0433\u0438\u043c \u0446\u0432\u0435\u0442\u043e\u043c. (https://www.gazeta.ru/science/2021/12/17_a_14325355.shtml)", "example_title": "\u041d\u043e\u0432\u043e\u0441\u0442\u044c \u043f\u0440\u043e \u043c\u043d\u043e\u0433\u043e\u043d\u043e\u0436\u043a\u0443"}, {"text": "\u0412\u044b\u0441\u043e\u0442\u0430 \u0431\u0430\u0448\u043d\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 324 \u043c\u0435\u0442\u0440\u0430 (1063 \u0444\u0443\u0442\u0430), \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e \u0442\u0430\u043a\u0430\u044f \u0436\u0435 \u0432\u044b\u0441\u043e\u0442\u0430, \u043a\u0430\u043a \u0443 81-\u044d\u0442\u0430\u0436\u043d\u043e\u0433\u043e \u0437\u0434\u0430\u043d\u0438\u044f, \u0438 \u0441\u0430\u043c\u043e\u0435 \u0432\u044b\u0441\u043e\u043a\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u0432 \u041f\u0430\u0440\u0438\u0436\u0435. \u0415\u0433\u043e \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u0438\u0435 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u043e, \u0440\u0430\u0437\u043c\u0435\u0440\u043e\u043c 125 \u043c\u0435\u0442\u0440\u043e\u0432 (410 \u0444\u0443\u0442\u043e\u0432) \u0441 \u043b\u044e\u0431\u043e\u0439 \u0441\u0442\u043e\u0440\u043e\u043d\u044b. \u0412\u043e \u0432\u0440\u0435\u043c\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u043f\u0440\u0435\u0432\u0437\u043e\u0448\u043b\u0430 \u043c\u043e\u043d\u0443\u043c\u0435\u043d\u0442 \u0412\u0430\u0448\u0438\u043d\u0433\u0442\u043e\u043d\u0430, \u0441\u0442\u0430\u0432 \u0441\u0430\u043c\u044b\u043c \u0432\u044b\u0441\u043e\u043a\u0438\u043c \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u043c \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435\u043c \u0432 \u043c\u0438\u0440\u0435, \u0438 \u044d\u0442\u043e\u0442 \u0442\u0438\u0442\u0443\u043b \u043e\u043d\u0430 \u0443\u0434\u0435\u0440\u0436\u0438\u0432\u0430\u043b\u0430 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 41 \u0433\u043e\u0434\u0430 \u0434\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u043e \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u0432 \u041d\u044c\u044e-\u0419\u043e\u0440\u043a\u0435 \u0432 1930 \u0433\u043e\u0434\u0443. \u042d\u0442\u043e \u043f\u0435\u0440\u0432\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0434\u043e\u0441\u0442\u0438\u0433\u043b\u043e \u0432\u044b\u0441\u043e\u0442\u044b 300 \u043c\u0435\u0442\u0440\u043e\u0432. \u0418\u0437-\u0437\u0430 \u0434\u043e\u0431\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0432\u0435\u0449\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0430\u043d\u0442\u0435\u043d\u043d\u044b \u043d\u0430 \u0432\u0435\u0440\u0448\u0438\u043d\u0435 \u0431\u0430\u0448\u043d\u0438 \u0432 1957 \u0433\u043e\u0434\u0443 \u043e\u043d\u0430 \u0441\u0435\u0439\u0447\u0430\u0441 \u0432\u044b\u0448\u0435 \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u043d\u0430 5,2 \u043c\u0435\u0442\u0440\u0430 (17 \u0444\u0443\u0442\u043e\u0432). \u0417\u0430 \u0438\u0441\u043a\u043b\u044e\u0447\u0435\u043d\u0438\u0435\u043c \u043f\u0435\u0440\u0435\u0434\u0430\u0442\u0447\u0438\u043a\u043e\u0432, \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0432\u0442\u043e\u0440\u043e\u0439 \u0441\u0430\u043c\u043e\u0439 \u0432\u044b\u0441\u043e\u043a\u043e\u0439 \u043e\u0442\u0434\u0435\u043b\u044c\u043d\u043e \u0441\u0442\u043e\u044f\u0449\u0435\u0439 \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u043e\u0439 \u0432\u043e \u0424\u0440\u0430\u043d\u0446\u0438\u0438 \u043f\u043e\u0441\u043b\u0435 \u0432\u0438\u0430\u0434\u0443\u043a\u0430 \u041c\u0438\u0439\u043e.", "example_title": "\u0412\u0438\u043a\u0438\u043f\u0435\u0434\u0438\u044f"}]} | IlyaGusev/rut5_base_headline_gen_telegram | null | [
"transformers",
"pytorch",
"t5",
"text2text-generation",
"summarization",
"ru",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
summarization | transformers |
# RuT5SumGazeta
## Model description
This is the model for abstractive summarization for Russian based on [rut5-base](https://huggingface.co/cointegrated/rut5-base).
## Intended uses & limitations
#### How to use
Colab: [link](https://colab.research.google.com/drive/1re5E26ZIDUpAx1gOCZkbF3hcwjozmgG0)
```python
from transformers import AutoTokenizer, T5ForConditionalGeneration
model_name = "IlyaGusev/rut5_base_sum_gazeta"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name)
article_text = "..."
input_ids = tokenizer(
[article_text],
max_length=600,
add_special_tokens=True,
padding="max_length",
truncation=True,
return_tensors="pt"
)["input_ids"]
output_ids = model.generate(
input_ids=input_ids,
no_repeat_ngram_size=4
)[0]
summary = tokenizer.decode(output_ids, skip_special_tokens=True)
print(summary)
```
## Training data
- Dataset: [Gazeta](https://huggingface.co/datasets/IlyaGusev/gazeta)
## Training procedure
- Training script: [train.py](https://github.com/IlyaGusev/summarus/blob/master/external/hf_scripts/train.py)
- Config: [t5_training_config.json](https://github.com/IlyaGusev/summarus/blob/master/external/hf_scripts/configs/t5_training_config.json)
## Eval results
* Train dataset: **Gazeta v1 train**
* Test dataset: **Gazeta v1 test**
* Source max_length: **600**
* Target max_length: **200**
* no_repeat_ngram_size: **4**
* num_beams: **5**
| Model | R-1-f | R-2-f | R-L-f | chrF | METEOR | BLEU | Avg char length |
|:--------------------------|:------|:------|:------|:-------|:-------|:-----|:-----|
| [mbart_ru_sum_gazeta](https://huggingface.co/IlyaGusev/mbart_ru_sum_gazeta) | **32.4** | 14.3 | 28.0 | 39.7 | **26.4** | 12.1 | 371 |
| [rut5_base_sum_gazeta](https://huggingface.co/IlyaGusev/rut5_base_sum_gazeta) | 32.2 | **14.4** | **28.1** | **39.8** | 25.7 | **12.3** | 330 |
| [rugpt3medium_sum_gazeta](https://huggingface.co/IlyaGusev/rugpt3medium_sum_gazeta) | 26.2 | 7.7 | 21.7 | 33.8 | 18.2 | 4.3 | 244 |
* Train dataset: **Gazeta v1 train**
* Test dataset: **Gazeta v2 test**
* Source max_length: **600**
* Target max_length: **200**
* no_repeat_ngram_size: **4**
* num_beams: **5**
| Model | R-1-f | R-2-f | R-L-f | chrF | METEOR | BLEU | Avg char length |
|:--------------------------|:------|:------|:------|:-------|:-------|:-----|:-----|
| [mbart_ru_sum_gazeta](https://huggingface.co/IlyaGusev/mbart_ru_sum_gazeta) | **28.7** | **11.1** | 24.4 | **37.3** | **22.7** | **9.4** | 373 |
| [rut5_base_sum_gazeta](https://huggingface.co/IlyaGusev/rut5_base_sum_gazeta) | 28.6 | **11.1** | **24.5** | 37.2 | 22.0 | **9.4** | 331 |
| [rugpt3medium_sum_gazeta](https://huggingface.co/IlyaGusev/rugpt3medium_sum_gazeta) | 24.1 | 6.5 | 19.8 | 32.1 | 16.3 | 3.6 | 242 |
Predicting all summaries:
```python
import json
import torch
from transformers import AutoTokenizer, T5ForConditionalGeneration
from datasets import load_dataset
def gen_batch(inputs, batch_size):
batch_start = 0
while batch_start < len(inputs):
yield inputs[batch_start: batch_start + batch_size]
batch_start += batch_size
def predict(
model_name,
input_records,
output_file,
max_source_tokens_count=600,
batch_size=8
):
device = "cuda" if torch.cuda.is_available() else "cpu"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = T5ForConditionalGeneration.from_pretrained(model_name).to(device)
predictions = []
for batch in gen_batch(input_records, batch_size):
texts = [r["text"] for r in batch]
input_ids = tokenizer(
texts,
add_special_tokens=True,
max_length=max_source_tokens_count,
padding="max_length",
truncation=True,
return_tensors="pt"
)["input_ids"].to(device)
output_ids = model.generate(
input_ids=input_ids,
no_repeat_ngram_size=4
)
summaries = tokenizer.batch_decode(output_ids, skip_special_tokens=True)
for s in summaries:
print(s)
predictions.extend(summaries)
with open(output_file, "w") as w:
for p in predictions:
w.write(p.strip().replace("\n", " ") + "\n")
gazeta_test = load_dataset('IlyaGusev/gazeta', script_version="v1.0")["test"]
predict("IlyaGusev/rut5_base_sum_gazeta", list(gazeta_test), "t5_predictions.txt")
```
Evaluation script: [evaluate.py](https://github.com/IlyaGusev/summarus/blob/master/evaluate.py)
Flags: --language ru --tokenize-after --lower
| {"language": ["ru"], "license": ["apache-2.0"], "tags": ["summarization", "t5"], "datasets": ["IlyaGusev/gazeta"], "inference": {"parameters": {"no_repeat_ngram_size": 4}}, "widget": [{"text": "\u0412\u044b\u0441\u043e\u0442\u0430 \u0431\u0430\u0448\u043d\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 324 \u043c\u0435\u0442\u0440\u0430 (1063 \u0444\u0443\u0442\u0430), \u043f\u0440\u0438\u043c\u0435\u0440\u043d\u043e \u0442\u0430\u043a\u0430\u044f \u0436\u0435 \u0432\u044b\u0441\u043e\u0442\u0430, \u043a\u0430\u043a \u0443 81-\u044d\u0442\u0430\u0436\u043d\u043e\u0433\u043e \u0437\u0434\u0430\u043d\u0438\u044f, \u0438 \u0441\u0430\u043c\u043e\u0435 \u0432\u044b\u0441\u043e\u043a\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u0432 \u041f\u0430\u0440\u0438\u0436\u0435. \u0415\u0433\u043e \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u0438\u0435 \u043a\u0432\u0430\u0434\u0440\u0430\u0442\u043d\u043e, \u0440\u0430\u0437\u043c\u0435\u0440\u043e\u043c 125 \u043c\u0435\u0442\u0440\u043e\u0432 (410 \u0444\u0443\u0442\u043e\u0432) \u0441 \u043b\u044e\u0431\u043e\u0439 \u0441\u0442\u043e\u0440\u043e\u043d\u044b. \u0412\u043e \u0432\u0440\u0435\u043c\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430 \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u043f\u0440\u0435\u0432\u0437\u043e\u0448\u043b\u0430 \u043c\u043e\u043d\u0443\u043c\u0435\u043d\u0442 \u0412\u0430\u0448\u0438\u043d\u0433\u0442\u043e\u043d\u0430, \u0441\u0442\u0430\u0432 \u0441\u0430\u043c\u044b\u043c \u0432\u044b\u0441\u043e\u043a\u0438\u043c \u0438\u0441\u043a\u0443\u0441\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u043c \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435\u043c \u0432 \u043c\u0438\u0440\u0435, \u0438 \u044d\u0442\u043e\u0442 \u0442\u0438\u0442\u0443\u043b \u043e\u043d\u0430 \u0443\u0434\u0435\u0440\u0436\u0438\u0432\u0430\u043b\u0430 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 41 \u0433\u043e\u0434\u0430 \u0434\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044f \u0441\u0442\u0440\u043e\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u043e \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u0432 \u041d\u044c\u044e-\u0419\u043e\u0440\u043a\u0435 \u0432 1930 \u0433\u043e\u0434\u0443. \u042d\u0442\u043e \u043f\u0435\u0440\u0432\u043e\u0435 \u0441\u043e\u043e\u0440\u0443\u0436\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u0434\u043e\u0441\u0442\u0438\u0433\u043b\u043e \u0432\u044b\u0441\u043e\u0442\u044b 300 \u043c\u0435\u0442\u0440\u043e\u0432. \u0418\u0437-\u0437\u0430 \u0434\u043e\u0431\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0432\u0435\u0449\u0430\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0430\u043d\u0442\u0435\u043d\u043d\u044b \u043d\u0430 \u0432\u0435\u0440\u0448\u0438\u043d\u0435 \u0431\u0430\u0448\u043d\u0438 \u0432 1957 \u0433\u043e\u0434\u0443 \u043e\u043d\u0430 \u0441\u0435\u0439\u0447\u0430\u0441 \u0432\u044b\u0448\u0435 \u0437\u0434\u0430\u043d\u0438\u044f \u041a\u0440\u0430\u0439\u0441\u043b\u0435\u0440 \u043d\u0430 5,2 \u043c\u0435\u0442\u0440\u0430 (17 \u0444\u0443\u0442\u043e\u0432). \u0417\u0430 \u0438\u0441\u043a\u043b\u044e\u0447\u0435\u043d\u0438\u0435\u043c \u043f\u0435\u0440\u0435\u0434\u0430\u0442\u0447\u0438\u043a\u043e\u0432, \u042d\u0439\u0444\u0435\u043b\u0435\u0432\u0430 \u0431\u0430\u0448\u043d\u044f \u044f\u0432\u043b\u044f\u0435\u0442\u0441\u044f \u0432\u0442\u043e\u0440\u043e\u0439 \u0441\u0430\u043c\u043e\u0439 \u0432\u044b\u0441\u043e\u043a\u043e\u0439 \u043e\u0442\u0434\u0435\u043b\u044c\u043d\u043e \u0441\u0442\u043e\u044f\u0449\u0435\u0439 \u0441\u0442\u0440\u0443\u043a\u0442\u0443\u0440\u043e\u0439 \u0432\u043e \u0424\u0440\u0430\u043d\u0446\u0438\u0438 \u043f\u043e\u0441\u043b\u0435 \u0432\u0438\u0430\u0434\u0443\u043a\u0430 \u041c\u0438\u0439\u043e.", "example_title": "\u0412\u0438\u043a\u0438\u043f\u0435\u0434\u0438\u044f"}, {"text": "\u0421 1 \u0441\u0435\u043d\u0442\u044f\u0431\u0440\u044f \u0432 \u0420\u043e\u0441\u0441\u0438\u0438 \u0432\u0441\u0442\u0443\u043f\u0430\u044e\u0442 \u0432 \u0441\u0438\u043b\u0443 \u043f\u043e\u043f\u0440\u0430\u0432\u043a\u0438 \u0432 \u0437\u0430\u043a\u043e\u043d \u00ab\u041e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u0435\u00bb \u2014 \u0442\u0435\u043f\u0435\u0440\u044c \u0434\u043e\u043b\u0436\u043d\u0438\u043a\u0438 \u0441\u043c\u043e\u0433\u0443\u0442 \u043e\u0441\u0432\u043e\u0431\u043e\u0436\u0434\u0430\u0442\u044c\u0441\u044f \u043e\u0442 \u043d\u0435\u043f\u043e\u0441\u0438\u043b\u044c\u043d\u044b\u0445 \u043e\u0431\u044f\u0437\u0430\u0442\u0435\u043b\u044c\u0441\u0442\u0432 \u0432\u043e \u0432\u043d\u0435\u0441\u0443\u0434\u0435\u0431\u043d\u043e\u043c \u043f\u043e\u0440\u044f\u0434\u043a\u0435, \u0435\u0441\u043b\u0438 \u0441\u0443\u043c\u043c\u0430 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u0438 \u0441\u043e\u0441\u0442\u0430\u0432\u043b\u044f\u0435\u0442 \u043d\u0435 \u043c\u0435\u043d\u0435\u0435 50 \u0442\u044b\u0441. \u0440\u0443\u0431\u043b\u0435\u0439 \u0438 \u043d\u0435 \u043f\u0440\u0435\u0432\u044b\u0448\u0430\u0435\u0442 500 \u0442\u044b\u0441. \u0440\u0443\u0431\u043b\u0435\u0439 \u0431\u0435\u0437 \u0443\u0447\u0435\u0442\u0430 \u0448\u0442\u0440\u0430\u0444\u043e\u0432, \u043f\u0435\u043d\u0438, \u043f\u0440\u043e\u0446\u0435\u043d\u0442\u043e\u0432 \u0437\u0430 \u043f\u0440\u043e\u0441\u0440\u043e\u0447\u043a\u0443 \u043f\u043b\u0430\u0442\u0435\u0436\u0430 \u0438 \u043f\u0440\u043e\u0447\u0438\u0445 \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0445 \u0438\u043b\u0438 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u044b\u0445 \u0441\u0430\u043d\u043a\u0446\u0438\u0439. \u0423 \u0444\u0438\u0437\u043b\u0438\u0446 \u0438 \u0438\u043d\u0434\u0438\u0432\u0438\u0434\u0443\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0442\u0435\u043b\u0435\u0439 \u043f\u043e\u044f\u0432\u0438\u043b\u0430\u0441\u044c \u0432\u043e\u0437\u043c\u043e\u0436\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u043e\u0439\u0442\u0438 \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u0443 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u0430 \u0431\u0435\u0437 \u0443\u0447\u0430\u0441\u0442\u0438\u044f \u0441\u0443\u0434\u0430 \u0438 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u043e\u0433\u043e \u0443\u043f\u0440\u0430\u0432\u043b\u044f\u044e\u0449\u0435\u0433\u043e \u2014 \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e \u043f\u043e\u0434\u0430\u0442\u044c \u0441\u043e\u043e\u0442\u0432\u0435\u0442\u0441\u0442\u0432\u0443\u044e\u0449\u0435\u0435 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0435 \u0447\u0435\u0440\u0435\u0437 \u041c\u0424\u0426. \u0421\u0443\u043c\u043c\u0443 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u0438 \u0438 \u0441\u043f\u0438\u0441\u043e\u043a \u0432\u0441\u0435\u0445 \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u044b\u0445 \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044e \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432 \u043d\u0443\u0436\u043d\u043e \u043f\u0440\u0435\u0434\u043e\u0441\u0442\u0430\u0432\u0438\u0442\u044c \u0441\u0430\u043c\u043e\u0441\u0442\u043e\u044f\u0442\u0435\u043b\u044c\u043d\u043e. \u0415\u0441\u043b\u0438 \u0432\u0441\u0435 \u0443\u0441\u043b\u043e\u0432\u0438\u044f \u0441\u043e\u0431\u043b\u044e\u0434\u0435\u043d\u044b, \u0441\u0432\u0435\u0434\u0435\u043d\u0438\u044f \u0432\u043d\u0435\u0441\u0443\u0442 \u0432 \u0415\u0434\u0438\u043d\u044b\u0439 \u0444\u0435\u0434\u0435\u0440\u0430\u043b\u044c\u043d\u044b\u0439 \u0440\u0435\u0435\u0441\u0442\u0440 \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 \u0442\u0440\u0435\u0445 \u0440\u0430\u0431\u043e\u0447\u0438\u0445 \u0434\u043d\u0435\u0439. \u041f\u0440\u0438 \u044d\u0442\u043e\u043c \u043d\u0430 \u043c\u043e\u043c\u0435\u043d\u0442 \u043f\u043e\u0434\u0430\u0447\u0438 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u044f \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044f \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u043e\u043a\u043e\u043d\u0447\u0435\u043d\u043e \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u043e \u0441 \u0432\u043e\u0437\u0432\u0440\u0430\u0449\u0435\u043d\u0438\u0435\u043c \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0433\u043e \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u0430 \u0432\u0437\u044b\u0441\u043a\u0430\u0442\u0435\u043b\u044e. \u042d\u0442\u043e \u0437\u043d\u0430\u0447\u0438\u0442, \u0447\u0442\u043e \u0443 \u043f\u043e\u0442\u0435\u043d\u0446\u0438\u0430\u043b\u044c\u043d\u043e\u0433\u043e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0430 \u043d\u0435 \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0430, \u043a\u043e\u0442\u043e\u0440\u043e\u0435 \u043c\u043e\u0436\u043d\u043e \u0432\u0437\u044b\u0441\u043a\u0430\u0442\u044c. \u041a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0433\u0440\u0430\u0436\u0434\u0430\u043d\u0438\u043d\u0430 \u043d\u0435 \u0434\u043e\u043b\u0436\u043d\u043e \u0431\u044b\u0442\u044c \u0432\u043e\u0437\u0431\u0443\u0436\u0434\u0435\u043d\u043e \u0434\u0440\u0443\u0433\u043e\u0435 \u0438\u0441\u043f\u043e\u043b\u043d\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0435 \u043f\u0440\u043e\u0438\u0437\u0432\u043e\u0434\u0441\u0442\u0432\u043e. \u0412 \u043f\u0435\u0440\u0438\u043e\u0434 \u0432\u0441\u0435\u0439 \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u044b \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044c \u043d\u0435 \u0441\u043c\u043e\u0436\u0435\u0442 \u0431\u0440\u0430\u0442\u044c \u0437\u0430\u0439\u043c\u044b, \u043a\u0440\u0435\u0434\u0438\u0442\u044b, \u0432\u044b\u0434\u0430\u0432\u0430\u0442\u044c \u043f\u043e\u0440\u0443\u0447\u0438\u0442\u0435\u043b\u044c\u0441\u0442\u0432\u0430, \u0441\u043e\u0432\u0435\u0440\u0448\u0430\u0442\u044c \u0438\u043d\u044b\u0435 \u043e\u0431\u0435\u0441\u043f\u0435\u0447\u0438\u0442\u0435\u043b\u044c\u043d\u044b\u0435 \u0441\u0434\u0435\u043b\u043a\u0438. \u0412\u043d\u0435\u0441\u0443\u0434\u0435\u0431\u043d\u043e\u0435 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432\u043e \u0431\u0443\u0434\u0435\u0442 \u0434\u043b\u0438\u0442\u044c\u0441\u044f \u0448\u0435\u0441\u0442\u044c \u043c\u0435\u0441\u044f\u0446\u0435\u0432, \u0432 \u0442\u0435\u0447\u0435\u043d\u0438\u0435 \u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u0442\u0430\u043a\u0436\u0435 \u0431\u0443\u0434\u0435\u0442 \u0434\u0435\u0439\u0441\u0442\u0432\u043e\u0432\u0430\u0442\u044c \u043c\u043e\u0440\u0430\u0442\u043e\u0440\u0438\u0439 \u043d\u0430 \u0443\u0434\u043e\u0432\u043b\u0435\u0442\u0432\u043e\u0440\u0435\u043d\u0438\u0435 \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u0438\u0439 \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432, \u043e\u0442\u043c\u0435\u0447\u0435\u043d\u043d\u044b\u0445 \u0432 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0438 \u0434\u043e\u043b\u0436\u043d\u0438\u043a\u0430, \u0438 \u043c\u043e\u0440\u0430\u0442\u043e\u0440\u0438\u0439 \u043e\u0431 \u0443\u043f\u043b\u0430\u0442\u0435 \u043e\u0431\u044f\u0437\u0430\u0442\u0435\u043b\u044c\u043d\u044b\u0445 \u043f\u043b\u0430\u0442\u0435\u0436\u0435\u0439. \u041a\u0440\u043e\u043c\u0435 \u0442\u043e\u0433\u043e, \u043f\u0440\u0435\u043a\u0440\u0430\u0449\u0430\u0435\u0442\u0441\u044f \u043d\u0430\u0447\u0438\u0441\u043b\u0435\u043d\u0438\u0435 \u043d\u0435\u0443\u0441\u0442\u043e\u0435\u043a \u0438 \u0438\u043d\u044b\u0445 \u0444\u0438\u043d\u0430\u043d\u0441\u043e\u0432\u044b\u0445 \u0441\u0430\u043d\u043a\u0446\u0438\u0439; \u0438\u043c\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u044b\u0435 \u0432\u0437\u044b\u0441\u043a\u0430\u043d\u0438\u044f (\u043a\u0440\u043e\u043c\u0435 \u0430\u043b\u0438\u043c\u0435\u043d\u0442\u043e\u0432) \u0442\u0430\u043a\u0436\u0435 \u0431\u0443\u0434\u0443\u0442 \u043f\u0440\u0438\u043e\u0441\u0442\u0430\u043d\u043e\u0432\u043b\u0435\u043d\u044b. \u041f\u043e \u0437\u0430\u0432\u0435\u0440\u0448\u0435\u043d\u0438\u044e \u043f\u0440\u043e\u0446\u0435\u0434\u0443\u0440\u044b \u0437\u0430\u044f\u0432\u0438\u0442\u0435\u043b\u044f \u043e\u0441\u0432\u043e\u0431\u043e\u0434\u044f\u0442 \u043e\u0442 \u0434\u0430\u043b\u044c\u043d\u0435\u0439\u0448\u0435\u0433\u043e \u0432\u044b\u043f\u043e\u043b\u043d\u0435\u043d\u0438\u044f \u0442\u0440\u0435\u0431\u043e\u0432\u0430\u043d\u0438\u0439 \u043a\u0440\u0435\u0434\u0438\u0442\u043e\u0440\u043e\u0432, \u0443\u043a\u0430\u0437\u0430\u043d\u043d\u044b\u0445 \u0432 \u0437\u0430\u044f\u0432\u043b\u0435\u043d\u0438\u0438 \u043e \u043f\u0440\u0438\u0437\u043d\u0430\u043d\u0438\u0438 \u0435\u0433\u043e \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u043e\u043c, \u0430 \u044d\u0442\u0430 \u0437\u0430\u0434\u043e\u043b\u0436\u0435\u043d\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u0438\u0437\u043d\u0430\u0435\u0442\u0441\u044f \u0431\u0435\u0437\u043d\u0430\u0434\u0435\u0436\u043d\u043e\u0439. \u0412 \u043f\u0440\u043e\u0448\u043b\u043e\u043c \u043c\u0435\u0441\u044f\u0446\u0435 \u0441\u0442\u0430\u043b\u043e \u0438\u0437\u0432\u0435\u0441\u0442\u043d\u043e, \u0447\u0442\u043e \u0437\u0430 \u043f\u0435\u0440\u0432\u043e\u0435 \u043f\u043e\u043b\u0443\u0433\u043e\u0434\u0438\u0435 2020 \u0433\u043e\u0434\u0430 \u0440\u043e\u0441\u0441\u0438\u0439\u0441\u043a\u0438\u0435 \u0441\u0443\u0434\u044b \u043f\u0440\u0438\u0437\u043d\u0430\u043b\u0438 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0430\u043c\u0438 42,7 \u0442\u044b\u0441. \u0433\u0440\u0430\u0436\u0434\u0430\u043d (\u0432 \u0442\u043e\u043c \u0447\u0438\u0441\u043b\u0435 \u0438\u043d\u0434\u0438\u0432\u0438\u0434\u0443\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u0442\u0435\u043b\u0435\u0439) \u2014 \u043f\u043e \u0434\u0430\u043d\u043d\u044b\u043c \u0435\u0434\u0438\u043d\u043e\u0433\u043e \u0440\u0435\u0435\u0441\u0442\u0440\u0430 \u00ab\u0424\u0435\u0434\u0440\u0435\u0441\u0443\u0440\u0441\u00bb, \u044d\u0442\u043e \u043d\u0430 47,2% \u0431\u043e\u043b\u044c\u0448\u0435 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044f \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u043e\u0433\u043e \u043f\u0435\u0440\u0438\u043e\u0434\u0430 2019 \u0433\u043e\u0434\u0430. \u0420\u043e\u0441\u0442 \u0447\u0438\u0441\u043b\u0430 \u043e\u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0438\u0432\u0448\u0438\u0445\u0441\u044f \u0433\u0440\u0430\u0436\u0434\u0430\u043d \u0432\u043e \u0432\u0442\u043e\u0440\u043e\u043c \u043a\u0432\u0430\u0440\u0442\u0430\u043b\u0435 \u043f\u043e \u0441\u0440\u0430\u0432\u043d\u0435\u043d\u0438\u044e \u0441 \u043f\u0435\u0440\u0432\u044b\u043c \u0437\u0430\u043c\u0435\u0434\u043b\u0438\u043b\u0441\u044f \u2014 \u0442\u0430\u043a\u0430\u044f \u0434\u0438\u043d\u0430\u043c\u0438\u043a\u0430 \u043e\u0431\u0443\u0441\u043b\u043e\u0432\u043b\u0435\u043d\u0430 \u0442\u0435\u043c, \u0447\u0442\u043e \u0432 \u043f\u0435\u0440\u0438\u043e\u0434 \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u0438\u0439 \u0441 19 \u043c\u0430\u0440\u0442\u0430 \u043f\u043e 11 \u043c\u0430\u044f \u0441\u0443\u0434\u044b \u0440\u0435\u0434\u043a\u043e \u0440\u0430\u0441\u0441\u043c\u0430\u0442\u0440\u0438\u0432\u0430\u043b\u0438 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u043d\u044b\u0435 \u0434\u0435\u043b\u0430 \u043a\u043e\u043c\u043f\u0430\u043d\u0438\u0439 \u0438 \u043c\u0435\u043d\u044c\u0448\u0435, \u0447\u0435\u043c \u043e\u0431\u044b\u0447\u043d\u043e, \u0432 \u043e\u0442\u043d\u043e\u0448\u0435\u043d\u0438\u0438 \u0433\u0440\u0430\u0436\u0434\u0430\u043d, \u043e\u0431\u044a\u044f\u0441\u043d\u044f\u043b \u0440\u0443\u043a\u043e\u0432\u043e\u0434\u0438\u0442\u0435\u043b\u044c \u043f\u0440\u043e\u0435\u043a\u0442\u0430 \u00ab\u0424\u0435\u0434\u0440\u0435\u0441\u0443\u0440\u0441\u00bb \u0410\u043b\u0435\u043a\u0441\u0435\u0439 \u042e\u0445\u043d\u0438\u043d. \u041e\u043d \u043f\u0440\u043e\u0433\u043d\u043e\u0437\u0438\u0440\u0443\u0435\u0442, \u0447\u0442\u043e \u0432\u043e \u0432\u0442\u043e\u0440\u043e\u043c \u043f\u043e\u043b\u0443\u0433\u043e\u0434\u0438\u0438 \u043c\u044b \u0443\u0432\u0438\u0434\u0438\u043c \u0440\u043e\u0441\u0442 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044f, \u043a\u043e\u0433\u0434\u0430 \u0441\u0443\u0434\u044b \u0440\u0430\u0441\u0441\u043c\u043e\u0442\u0440\u044f\u0442 \u0432\u0441\u0435 \u0434\u0435\u043b\u0430, \u0447\u0442\u043e \u043d\u0435 \u0441\u043c\u043e\u0433\u043b\u0438 \u0440\u0430\u043d\u0435\u0435 \u0432 \u0440\u0435\u0436\u0438\u043c\u0435 \u043e\u0433\u0440\u0430\u043d\u0438\u0447\u0435\u043d\u0438\u0439. \u041f\u043e \u0435\u0433\u043e \u0434\u0430\u043d\u043d\u044b\u043c, \u0443\u0436\u0435 \u0432 \u0438\u044e\u043d\u0435 \u0447\u0438\u0441\u043b\u043e \u043b\u0438\u0447\u043d\u044b\u0445 \u0431\u0430\u043d\u043a\u0440\u043e\u0442\u0441\u0442\u0432 \u0432\u044b\u0440\u043e\u0441\u043b\u043e \u0434\u043e 11,5 \u0442\u044b\u0441., \u0447\u0442\u043e \u0432 \u0434\u0432\u0430 \u0440\u0430\u0437\u0430 \u043f\u0440\u0435\u0432\u044b\u0448\u0430\u0435\u0442 \u043f\u043e\u043a\u0430\u0437\u0430\u0442\u0435\u043b\u044c \u0430\u043d\u0430\u043b\u043e\u0433\u0438\u0447\u043d\u043e\u0433\u043e \u043f\u0435\u0440\u0438\u043e\u0434\u0430 2019 \u0433\u043e\u0434\u0430.", "example_title": "\u041d\u043e\u0432\u043e\u0441\u0442\u0438"}, {"text": "\u0410\u043a\u0442\u0443\u0430\u043b\u044c\u043d\u043e\u0441\u0442\u044c \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b. \u042d\u043b\u0435\u043a\u0442\u0440\u043e\u043d\u043d\u0430\u044f \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u044f \u0438\u0433\u0440\u0430\u0435\u0442 \u0432\u0441\u0435 \u0431\u043e\u043b\u044c\u0448\u0443\u044e \u0440\u043e\u043b\u044c \u0432\u043e \u0432\u0441\u0435\u0445 \u0441\u0444\u0435\u0440\u0430\u0445 \u0436\u0438\u0437\u043d\u0438 \u0441\u043e\u0432\u0440\u0435\u043c\u0435\u043d\u043d\u043e\u0433\u043e \u043e\u0431\u0449\u0435\u0441\u0442\u0432\u0430. \u0412 \u043f\u043e\u0441\u043b\u0435\u0434\u043d\u0438\u0435 \u0433\u043e\u0434\u044b \u043e\u0431\u044a\u0435\u043c \u043d\u0430\u0443\u0447\u043d\u043e-\u0442\u0435\u0445\u043d\u0438\u0447\u0435\u0441\u043a\u043e\u0439 \u0442\u0435\u043a\u0441\u0442\u043e\u0432\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u0432 \u044d\u043b\u0435\u043a\u0442\u0440\u043e\u043d\u043d\u043e\u043c \u0432\u0438\u0434\u0435 \u0432\u043e\u0437\u0440\u043e\u0441 \u043d\u0430\u0441\u0442\u043e\u043b\u044c\u043a\u043e, \u0447\u0442\u043e \u0432\u043e\u0437\u043d\u0438\u043a\u0430\u0435\u0442 \u0443\u0433\u0440\u043e\u0437\u0430 \u043e\u0431\u0435\u0441\u0446\u0435\u043d\u0438\u0432\u0430\u043d\u0438\u044f \u044d\u0442\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u0432 \u0441\u0432\u044f\u0437\u0438 \u0441 \u0442\u0440\u0443\u0434\u043d\u043e\u0441\u0442\u044f\u043c\u0438 \u043f\u043e\u0438\u0441\u043a\u0430 \u043d\u0435\u043e\u0431\u0445\u043e\u0434\u0438\u043c\u044b\u0445 \u0441\u0432\u0435\u0434\u0435\u043d\u0438\u0439 \u0441\u0440\u0435\u0434\u0438 \u043c\u043d\u043e\u0436\u0435\u0441\u0442\u0432\u0430 \u0434\u043e\u0441\u0442\u0443\u043f\u043d\u044b\u0445 \u0442\u0435\u043a\u0441\u0442\u043e\u0432. \u0420\u0430\u0437\u0432\u0438\u0442\u0438\u0435 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u044b\u0445 \u0440\u0435\u0441\u0443\u0440\u0441\u043e\u0432 \u0418\u043d\u0442\u0435\u0440\u043d\u0435\u0442 \u043c\u043d\u043e\u0433\u043e\u043a\u0440\u0430\u0442\u043d\u043e \u0443\u0441\u0443\u0433\u0443\u0431\u0438\u043b\u043e \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u0443 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u043e\u0439 \u043f\u0435\u0440\u0435\u0433\u0440\u0443\u0437\u043a\u0438. \u0412 \u044d\u0442\u043e\u0439 \u0441\u0438\u0442\u0443\u0430\u0446\u0438\u0438 \u043e\u0441\u043e\u0431\u0435\u043d\u043d\u043e \u0430\u043a\u0442\u0443\u0430\u043b\u044c\u043d\u044b\u043c\u0438 \u0441\u0442\u0430\u043d\u043e\u0432\u044f\u0442\u0441\u044f \u043c\u0435\u0442\u043e\u0434\u044b \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0437\u0430\u0446\u0438\u0438 \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f \u0442\u0435\u043a\u0441\u0442\u043e\u0432\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438, \u0442\u043e \u0435\u0441\u0442\u044c \u043c\u0435\u0442\u043e\u0434\u044b \u043f\u043e\u043b\u0443\u0447\u0435\u043d\u0438\u044f \u0441\u0436\u0430\u0442\u043e\u0433\u043e \u043f\u0440\u0435\u0434\u0441\u0442\u0430\u0432\u043b\u0435\u043d\u0438\u044f \u0442\u0435\u043a\u0441\u0442\u043e\u0432\u044b\u0445 \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u043e\u0432\u2013\u0440\u0435\u0444\u0435\u0440\u0430\u0442\u043e\u0432 (\u0430\u043d\u043d\u043e\u0442\u0430\u0446\u0438\u0439). \u041f\u043e\u0441\u0442\u0430\u043d\u043e\u0432\u043a\u0430 \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0433\u043e \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f \u0442\u0435\u043a\u0441\u0442\u0430 \u0438 \u0441\u043e\u043e\u0442\u0432\u0435\u0442\u0441\u0442\u0432\u0435\u043d\u043d\u043e \u043f\u043e\u043f\u044b\u0442\u043a\u0438 \u0435\u0435 \u0440\u0435\u0448\u0435\u043d\u0438\u044f \u0441 \u0438\u0441\u043f\u043e\u043b\u044c\u0437\u043e\u0432\u0430\u043d\u0438\u0435\u043c \u0440\u0430\u0437\u043b\u0438\u0447\u043d\u044b\u0445 \u043f\u043e\u0434\u0445\u043e\u0434\u043e\u0432 \u043f\u0440\u0435\u0434\u043f\u0440\u0438\u043d\u0438\u043c\u0430\u043b\u0438\u0441\u044c \u043c\u043d\u043e\u0433\u0438\u043c\u0438 \u0438\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u044f\u043c\u0438. \u0418\u0441\u0442\u043e\u0440\u0438\u044f \u043f\u0440\u0438\u043c\u0435\u043d\u0435\u043d\u0438\u044f \u0432\u044b\u0447\u0438\u0441\u043b\u0438\u0442\u0435\u043b\u044c\u043d\u043e\u0439 \u0442\u0435\u0445\u043d\u0438\u043a\u0438 \u0434\u043b\u044f \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u044f \u043d\u0430\u0441\u0447\u0438\u0442\u044b\u0432\u0430\u0435\u0442 \u0443\u0436\u0435 \u0431\u043e\u043b\u0435\u0435 50 \u043b\u0435\u0442 \u0438 \u0441\u0432\u044f\u0437\u0430\u043d\u0430 \u0441 \u0438\u043c\u0435\u043d\u0430\u043c\u0438 \u0442\u0430\u043a\u0438\u0445 \u0438\u0441\u0441\u043b\u0435\u0434\u043e\u0432\u0430\u0442\u0435\u043b\u0435\u0439, \u043a\u0430\u043a \u0413.\u041f. \u041b\u0443\u043d, \u0412.\u0415. \u0411\u0435\u0440\u0437\u043e\u043d, \u0418.\u041f. C\u0435\u0432\u0431\u043e, \u042d.\u0424. \u0421\u043a\u043e\u0440\u043e\u0445\u043e\u0434\u044c\u043a\u043e, \u0414.\u0413. \u041b\u0430\u0445\u0443\u0442\u0438, \u0420.\u0413. \u041f\u0438\u043e\u0442\u0440\u043e\u0432\u0441\u043a\u0438\u0439 \u0438 \u0434\u0440. \u0417\u0430 \u044d\u0442\u0438 \u0433\u043e\u0434\u044b \u0432\u044b\u0440\u0430\u0431\u043e\u0442\u0430\u043d\u044b \u043c\u043d\u043e\u0433\u043e\u0447\u0438\u0441\u043b\u0435\u043d\u043d\u044b\u0435 \u043f\u043e\u0434\u0445\u043e\u0434\u044b \u043a \u0440\u0435\u0448\u0435\u043d\u0438\u044e \u0434\u0430\u043d\u043d\u043e\u0439 \u043f\u0440\u043e\u0431\u043b\u0435\u043c\u044b, \u043a\u043e\u0442\u043e\u0440\u044b\u0435 \u0434\u043e\u0441\u0442\u0430\u0442\u043e\u0447\u043d\u043e \u0447\u0435\u0442\u043a\u043e \u043f\u043e\u0434\u0440\u0430\u0437\u0434\u0435\u043b\u044f\u044e\u0442\u0441\u044f \u043d\u0430 \u0434\u0432\u0430 \u043d\u0430\u043f\u0440\u0430\u0432\u043b\u0435\u043d\u0438\u044f: \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0435 \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0435, \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u043d\u043e\u0435 \u043d\u0430 \u044d\u043a\u0441\u0442\u0440\u0430\u0433\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0438 \u0438\u0437 \u043f\u0435\u0440\u0432\u0438\u0447\u043d\u044b\u0445 \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u043e\u0432 \u0441 \u043f\u043e\u043c\u043e\u0449\u044c\u044e \u043e\u043f\u0440\u0435\u0434\u0435\u043b\u0435\u043d\u043d\u044b\u0445 \u0444\u043e\u0440\u043c\u0430\u043b\u044c\u043d\u044b\u0445 \u043f\u0440\u0438\u0437\u043d\u0430\u043a\u043e\u0432 \u00ab\u043d\u0430\u0438\u0431\u043e\u043b\u0435\u0435 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0442\u0438\u0432\u043d\u044b\u0445\u00bb \u0444\u0440\u0430\u0437 (\u0444\u0440\u0430\u0433\u043c\u0435\u043d\u0442\u043e\u0432), \u0441\u043e\u0432\u043e\u043a\u0443\u043f\u043d\u043e\u0441\u0442\u044c \u043a\u043e\u0442\u043e\u0440\u044b\u0445 \u043e\u0431\u0440\u0430\u0437\u0443\u0435\u0442 \u043d\u0435\u043a\u043e\u0442\u043e\u0440\u044b\u0439 \u044d\u043a\u0441\u0442\u0440\u0430\u043a\u0442; \u0430\u0432\u0442\u043e\u043c\u0430\u0442\u0438\u0447\u0435\u0441\u043a\u043e\u0435 \u0440\u0435\u0444\u0435\u0440\u0438\u0440\u043e\u0432\u0430\u043d\u0438\u0435, \u043e\u0441\u043d\u043e\u0432\u0430\u043d\u043d\u043e\u0435 \u043d\u0430 \u0432\u044b\u0434\u0435\u043b\u0435\u043d\u0438\u0438 \u0438\u0437 \u0442\u0435\u043a\u0441\u0442\u043e\u0432 \u0441 \u043f\u043e\u043c\u043e\u0449\u044c\u044e \u0441\u043f\u0435\u0446\u0438\u0430\u043b\u044c\u043d\u044b\u0445 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u043e\u043d\u043d\u044b\u0445 \u044f\u0437\u044b\u043a\u043e\u0432 \u043d\u0430\u0438\u0431\u043e\u043b\u0435\u0435 \u0441\u0443\u0449\u0435\u0441\u0442\u0432\u0435\u043d\u043d\u043e\u0439 \u0438\u043d\u0444\u043e\u0440\u043c\u0430\u0446\u0438\u0438 \u0438 \u043f\u043e\u0440\u043e\u0436\u0434\u0435\u043d\u0438\u0438 \u043d\u043e\u0432\u044b\u0445 \u0442\u0435\u043a\u0441\u0442\u043e\u0432 (\u0440\u0435\u0444\u0435\u0440\u0430\u0442\u043e\u0432), \u0441\u043e\u0434\u0435\u0440\u0436\u0430\u0442\u0435\u043b\u044c\u043d\u043e \u043e\u0431\u043e\u0431\u0449\u0430\u044e\u0449\u0438\u0445 \u043f\u0435\u0440\u0432\u0438\u0447\u043d\u044b\u0435 \u0434\u043e\u043a\u0443\u043c\u0435\u043d\u0442\u044b.", "example_title": "\u041d\u0430\u0443\u0447\u043d\u0430\u044f \u0441\u0442\u0430\u0442\u044c\u044f"}]} | IlyaGusev/rut5_base_sum_gazeta | null | [
"transformers",
"pytorch",
"t5",
"text2text-generation",
"summarization",
"ru",
"dataset:IlyaGusev/gazeta",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text2text-generation | transformers | {"language": ["ru"], "license": "apache-2.0", "widget": [{"text": "\u042d\u0442\u0430 \u0431\u043b\u044f\u0434\u044c \u043c\u0435\u043d\u044f \u0437\u0430\u0435\u0431\u0430\u043b\u0430</s> \u042d\u0442\u0430 <extra_id_0> \u043c\u0435\u043d\u044f <extra_id_1>"}]} | IlyaGusev/sber_rut5_filler | null | [
"transformers",
"pytorch",
"t5",
"text2text-generation",
"ru",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-classification | transformers |
# XLM-RoBERTa HeadlineCause Full
## Model description
This model was trained to predict the presence of causal relations between two headlines. This model is for the Full task with 7 possible labels: titles are almost the same, A causes B, B causes A, A refutes B, B refutes A, A linked with B in another way, A is not linked to B. English and Russian languages are supported.
You can use hosted inference API to infer a label for a headline pair. To do this, you shoud seperate headlines with ```</s>``` token.
For example:
```
Песков опроверг свой перевод на удаленку</s>Дмитрий Песков перешел на удаленку
```
## Intended uses & limitations
#### How to use
```python
from tqdm.notebook import tqdm
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
def get_batch(data, batch_size):
start_index = 0
while start_index < len(data):
end_index = start_index + batch_size
batch = data[start_index:end_index]
yield batch
start_index = end_index
def pipe_predict(data, pipe, batch_size=64):
raw_preds = []
for batch in tqdm(get_batch(data, batch_size)):
raw_preds += pipe(batch)
return raw_preds
MODEL_NAME = TOKENIZER_NAME = "IlyaGusev/xlm_roberta_large_headline_cause_full"
tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME, do_lower_case=False)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
model.eval()
pipe = pipeline("text-classification", model=model, tokenizer=tokenizer, framework="pt", return_all_scores=True)
texts = [
(
"Judge issues order to allow indoor worship in NC churches",
"Some local churches resume indoor services after judge lifted NC governor’s restriction"
),
(
"Gov. Kevin Stitt defends $2 million purchase of malaria drug touted by Trump",
"Oklahoma spent $2 million on malaria drug touted by Trump"
),
(
"Песков опроверг свой перевод на удаленку",
"Дмитрий Песков перешел на удаленку"
)
]
pipe_predict(texts, pipe)
```
#### Limitations and bias
The models are intended to be used on news headlines. No other limitations are known.
## Training data
* HuggingFace dataset: [IlyaGusev/headline_cause](https://huggingface.co/datasets/IlyaGusev/headline_cause)
* GitHub: [IlyaGusev/HeadlineCause](https://github.com/IlyaGusev/HeadlineCause)
## Training procedure
* Notebook: [HeadlineCause](https://colab.research.google.com/drive/1NAnD0OJ0TnYCJRsHpYUyYkjr_yi8ObcA)
* Stand-alone script: [train.py](https://github.com/IlyaGusev/HeadlineCause/blob/main/headline_cause/train.py)
## Eval results
Evaluation results can be found in the [arxiv paper](https://arxiv.org/pdf/2108.12626.pdf).
### BibTeX entry and citation info
```bibtex
@misc{gusev2021headlinecause,
title={HeadlineCause: A Dataset of News Headlines for Detecting Causalities},
author={Ilya Gusev and Alexey Tikhonov},
year={2021},
eprint={2108.12626},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
| {"language": ["ru", "en"], "license": "apache-2.0", "tags": ["xlm-roberta-large"], "datasets": ["IlyaGusev/headline_cause"], "widget": [{"text": "\u041f\u0435\u0441\u043a\u043e\u0432 \u043e\u043f\u0440\u043e\u0432\u0435\u0440\u0433 \u0441\u0432\u043e\u0439 \u043f\u0435\u0440\u0435\u0432\u043e\u0434 \u043d\u0430 \u0443\u0434\u0430\u043b\u0435\u043d\u043a\u0443</s>\u0414\u043c\u0438\u0442\u0440\u0438\u0439 \u041f\u0435\u0441\u043a\u043e\u0432 \u043f\u0435\u0440\u0435\u0448\u0435\u043b \u043d\u0430 \u0443\u0434\u0430\u043b\u0435\u043d\u043a\u0443"}]} | IlyaGusev/xlm_roberta_large_headline_cause_full | null | [
"transformers",
"pytorch",
"xlm-roberta",
"text-classification",
"xlm-roberta-large",
"ru",
"en",
"dataset:IlyaGusev/headline_cause",
"arxiv:2108.12626",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# XLM-RoBERTa HeadlineCause Simple
## Model description
This model was trained to predict the presence of causal relations between two headlines. This model is for the Simple task with 3 possible labels: A causes B, B causes A, no causal relation. English and Russian languages are supported.
You can use hosted inference API to infer a label for a headline pair. To do this, you shoud seperate headlines with ```</s>``` token.
For example:
```
Песков опроверг свой перевод на удаленку</s>Дмитрий Песков перешел на удаленку
```
## Intended uses & limitations
#### How to use
```python
from tqdm.notebook import tqdm
from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
def get_batch(data, batch_size):
start_index = 0
while start_index < len(data):
end_index = start_index + batch_size
batch = data[start_index:end_index]
yield batch
start_index = end_index
def pipe_predict(data, pipe, batch_size=64):
raw_preds = []
for batch in tqdm(get_batch(data, batch_size)):
raw_preds += pipe(batch)
return raw_preds
MODEL_NAME = TOKENIZER_NAME = "IlyaGusev/xlm_roberta_large_headline_cause_simple"
tokenizer = AutoTokenizer.from_pretrained(TOKENIZER_NAME, do_lower_case=False)
model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME)
model.eval()
pipe = pipeline("text-classification", model=model, tokenizer=tokenizer, framework="pt", return_all_scores=True)
texts = [
(
"Judge issues order to allow indoor worship in NC churches",
"Some local churches resume indoor services after judge lifted NC governor’s restriction"
),
(
"Gov. Kevin Stitt defends $2 million purchase of malaria drug touted by Trump",
"Oklahoma spent $2 million on malaria drug touted by Trump"
),
(
"Песков опроверг свой перевод на удаленку",
"Дмитрий Песков перешел на удаленку"
)
]
pipe_predict(texts, pipe)
```
#### Limitations and bias
The models are intended to be used on news headlines. No other limitations are known.
## Training data
* HuggingFace dataset: [IlyaGusev/headline_cause](https://huggingface.co/datasets/IlyaGusev/headline_cause)
* GitHub: [IlyaGusev/HeadlineCause](https://github.com/IlyaGusev/HeadlineCause)
## Training procedure
* Notebook: [HeadlineCause](https://colab.research.google.com/drive/1NAnD0OJ0TnYCJRsHpYUyYkjr_yi8ObcA)
* Stand-alone script: [train.py](https://github.com/IlyaGusev/HeadlineCause/blob/main/headline_cause/train.py)
## Eval results
Evaluation results can be found in the [arxiv paper](https://arxiv.org/pdf/2108.12626.pdf).
### BibTeX entry and citation info
```bibtex
@misc{gusev2021headlinecause,
title={HeadlineCause: A Dataset of News Headlines for Detecting Causalities},
author={Ilya Gusev and Alexey Tikhonov},
year={2021},
eprint={2108.12626},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
``` | {"language": ["ru", "en"], "license": "apache-2.0", "tags": ["xlm-roberta-large"], "datasets": ["IlyaGusev/headline_cause"], "widget": [{"text": "\u041f\u0435\u0441\u043a\u043e\u0432 \u043e\u043f\u0440\u043e\u0432\u0435\u0440\u0433 \u0441\u0432\u043e\u0439 \u043f\u0435\u0440\u0435\u0432\u043e\u0434 \u043d\u0430 \u0443\u0434\u0430\u043b\u0435\u043d\u043a\u0443</s>\u0414\u043c\u0438\u0442\u0440\u0438\u0439 \u041f\u0435\u0441\u043a\u043e\u0432 \u043f\u0435\u0440\u0435\u0448\u0435\u043b \u043d\u0430 \u0443\u0434\u0430\u043b\u0435\u043d\u043a\u0443"}]} | IlyaGusev/xlm_roberta_large_headline_cause_simple | null | [
"transformers",
"pytorch",
"xlm-roberta",
"text-classification",
"xlm-roberta-large",
"ru",
"en",
"dataset:IlyaGusev/headline_cause",
"arxiv:2108.12626",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Harry Botter Model | {"tags": ["conversational"]} | Ilyabarigou/Genesis-harrybotter | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
automatic-speech-recognition | transformers | ## Evaluation on Common Voice FR Test
The script used for training and evaluation can be found here: https://github.com/irebai/wav2vec2
```python
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import (
Wav2Vec2ForCTC,
Wav2Vec2Processor,
)
import re
model_name = "Ilyes/wav2vec2-large-xlsr-53-french"
device = "cpu" # "cuda"
model = Wav2Vec2ForCTC.from_pretrained(model_name).to(device)
processor = Wav2Vec2Processor.from_pretrained(model_name)
ds = load_dataset("common_voice", "fr", split="test", cache_dir="./data/fr")
chars_to_ignore_regex = '[\,\?\.\!\;\:\"\“\%\‘\”\�\‘\’\’\’\‘\…\·\!\ǃ\?\«\‹\»\›“\”\\ʿ\ʾ\„\∞\\|\.\,\;\:\*\—\–\─\―\_\/\:\ː\;\,\=\«\»\→]'
def map_to_array(batch):
speech, _ = torchaudio.load(batch["path"])
batch["speech"] = resampler.forward(speech.squeeze(0)).numpy()
batch["sampling_rate"] = resampler.new_freq
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower().replace("’", "'")
return batch
resampler = torchaudio.transforms.Resample(48_000, 16_000)
ds = ds.map(map_to_array)
def map_to_pred(batch):
features = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0], padding=True, return_tensors="pt")
input_values = features.input_values.to(device)
attention_mask = features.attention_mask.to(device)
with torch.no_grad():
logits = model(input_values, attention_mask=attention_mask).logits
pred_ids = torch.argmax(logits, dim=-1)
batch["predicted"] = processor.batch_decode(pred_ids)
batch["target"] = batch["sentence"]
return batch
result = ds.map(map_to_pred, batched=True, batch_size=16, remove_columns=list(ds.features.keys()))
wer = load_metric("wer")
print(wer.compute(predictions=result["predicted"], references=result["target"]))
```
## Results
WER=12.82%
CER=4.40%
| {"language": "fr", "license": "apache-2.0", "tags": ["audio", "automatic-speech-recognition", "speech", "xlsr-fine-tuning-week"], "datasets": ["common_voice"], "model-index": [{"name": "wav2vec2-large-xlsr-53-French by Ilyes Rebai", "results": [{"task": {"type": "automatic-speech-recognition", "name": "Speech Recognition"}, "dataset": {"name": "Common Voice fr", "type": "common_voice", "args": "fr"}, "metrics": [{"type": "wer", "value": 12.82, "name": "Test WER"}]}]}]} | Ilyes/wav2vec2-large-xlsr-53-french | null | [
"transformers",
"pytorch",
"wav2vec2",
"automatic-speech-recognition",
"audio",
"speech",
"xlsr-fine-tuning-week",
"fr",
"dataset:common_voice",
"license:apache-2.0",
"model-index",
"endpoints_compatible",
"has_space",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
automatic-speech-recognition | transformers | ## Evaluation on Common Voice FR Test
```python
import re
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import (
Wav2Vec2ForCTC,
Wav2Vec2Processor,
)
model_name = "Ilyes/wav2vec2-large-xlsr-53-french_punctuation"
model = Wav2Vec2ForCTC.from_pretrained(model_name).to('cuda')
processor = Wav2Vec2Processor.from_pretrained(model_name)
ds = load_dataset("common_voice", "fr", split="test")
chars_to_ignore_regex = '[\;\:\"\“\%\‘\”\�\‘\’\’\’\‘\…\·\ǃ\«\‹\»\›“\”\\ʿ\ʾ\„\∞\\|\;\:\*\—\–\─\―\_\/\:\ː\;\=\«\»\→]'
def normalize_text(text):
text = text.lower().strip()
text = re.sub('œ', 'oe', text)
text = re.sub('æ', 'ae', text)
text = re.sub("’|´|′|ʼ|‘|ʻ|`", "'", text)
text = re.sub("'+ ", " ", text)
text = re.sub(" '+", " ", text)
text = re.sub("'$", " ", text)
text = re.sub("' ", " ", text)
text = re.sub("−|‐", "-", text)
text = re.sub(" -", "", text)
text = re.sub("- ", "", text)
text = re.sub(chars_to_ignore_regex, '', text)
return text
def map_to_array(batch):
speech, _ = torchaudio.load(batch["path"])
batch["speech"] = resampler.forward(speech.squeeze(0)).numpy()
batch["sampling_rate"] = resampler.new_freq
batch["sentence"] = normalize_text(batch["sentence"])
return batch
ds = ds.map(map_to_array)
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def map_to_pred(batch):
features = processor(batch["speech"], sampling_rate=batch["sampling_rate"][0], padding=True, return_tensors="pt")
input_values = features.input_values.to(device)
attention_mask = features.attention_mask.to(device)
with torch.no_grad():
logits = model(input_values, attention_mask=attention_mask).logits
pred_ids = torch.argmax(logits, dim=-1)
batch["predicted"] = processor.batch_decode(pred_ids)
batch["target"] = batch["sentence"]
# remove duplicates
batch["target"] = re.sub('\.+', '.', batch["target"])
batch["target"] = re.sub('\?+', '?', batch["target"])
batch["target"] = re.sub('!+', '!', batch["target"])
batch["target"] = re.sub(',+', ',', batch["target"])
return batch
result = ds.map(map_to_pred, batched=True, batch_size=16, remove_columns=list(ds.features.keys()))
wer = load_metric("wer")
print(wer.compute(predictions=result["predicted"], references=result["target"]))
```
## Some results
| Reference | Prediction |
| ------------- | ------------- |
| il vécut à new york et y enseigna une grande partie de sa vie. | il a vécu à new york et y enseigna une grande partie de sa vie. |
| au classement par nations, l'allemagne est la tenante du titre. | au classement der nation l'allemagne est la tenante du titre. |
| voici un petit calcul pour fixer les idées. | voici un petit calcul pour fixer les idées. |
| oh! tu dois être beau avec | oh! tu dois être beau avec. |
| babochet vous le voulez? | baboche, vous le voulez? |
| la commission est, par conséquent, défavorable à cet amendement. | la commission est, par conséquent, défavorable à cet amendement. |
All the references and predictions of the test corpus are already available in this repository.
## Results
text + punctuation
WER=21.47% CER=7.21%
text (without punctuation)
WER=19.71% CER=6.91%
| {"language": "fr", "license": "apache-2.0", "tags": ["audio", "automatic-speech-recognition", "speech", "xlsr-fine-tuning"], "datasets": ["common_voice"]} | Ilyes/wav2vec2-large-xlsr-53-french_punctuation | null | [
"transformers",
"pytorch",
"jax",
"wav2vec2",
"automatic-speech-recognition",
"audio",
"speech",
"xlsr-fine-tuning",
"fr",
"dataset:common_voice",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Albert DialoGPT Model | {"tags": ["conversational"]} | ImAPizza/DialoGPT-medium-albert | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
# Alberttwo DialoGPT Model | {"tags": ["conversational"]} | ImAPizza/DialoGPT-medium-alberttwo | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | ImSagnik/model_name | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Imelbay/Brother | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Imperialhost/Pepe | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Incwu/mymodels | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
fill-mask | transformers | {} | InfoCoV/Cro-CoV-BERTic | null | [
"transformers",
"pytorch",
"tensorboard",
"electra",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
fill-mask | transformers | ## Usage:
```
from sentence_transformers import models
from sentence_transformers import SentenceTransformer
word_embedding_model = models.Transformer('Cro-CoV-cseBERT')
pooling_model = models.Pooling(word_embedding_model.get_word_embedding_dimension(),
pooling_mode_mean_tokens=True,
pooling_mode_cls_token=False,
pooling_mode_max_tokens=False)
model = SentenceTransformer(modules=[word_embedding_model, pooling_model], device='') ## device = 'gpu' or 'cpu'
texts_emb = model.encode(texts)
```
## Datasets:
https://github.com/InfoCoV/InfoCoV
## Paper:
Please cite https://www.mdpi.com/2076-3417/11/21/10442 | {} | InfoCoV/Cro-CoV-cseBERT | null | [
"transformers",
"pytorch",
"bert",
"fill-mask",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers | {} | InfoCoV/Senti-Cro-CoV-cseBERT | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers |
# Inkdrop/gpt2-property-classifier
| {"language": ["de"], "license": "mit", "tags": ["text-generation"], "widget": [{"text": "\"Ideal als kleine Aufmerksamkeit mit emotionalem Wert Neue Tuchmasken-Referenz \"Verw\u00f6hnmoment\u00bb exklusiv im Set Langanhaltende Feuchtigkeit und Erholung Mit strahlendem Teint Sofort-Effekt Naturnahe Kosmetik Inhalt: Badekristalle Kleiner Gruss von Herzen 60 g, Tuchmaske Verw\u00f6hnmoment 1x\" is a", "example_title": "Bullet point classification"}]} | Inkdrop/gpt2-property-classifier | null | [
"transformers",
"pytorch",
"tensorboard",
"gpt2",
"text-generation",
"de",
"license:mit",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | Inos/Martin | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | transformers | {} | Insun/wav2vec2_large_xlsr_53_VTCK_16K | null | [
"transformers",
"pytorch",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | # Welcome to my model | {"tags": ["chemistry", "climate"]} | Intae/mymodel | null | [
"chemistry",
"climate",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers |
# Sparse BERT base model fine tuned to MNLI without classifier layer (uncased)
Fine tuned sparse BERT base to MNLI (GLUE Benchmark) task from [bert-base-uncased-sparse-70-unstructured](https://huggingface.co/Intel/bert-base-uncased-sparse-70-unstructured).
<br>
This model doesn't have a classifier layer to enable easier loading of the model for training to other downstream tasks.
In all the other layers this model is similar to [bert-base-uncased-mnli-sparse-70-unstructured](https://huggingface.co/Intel/bert-base-uncased-mnli-sparse-70-unstructured).
<br><br>
Note: This model requires `transformers==2.10.0`
## Evaluation Results
Matched: 82.5%
Mismatched: 83.3%
This model can be further fine-tuned to other tasks and achieve the following evaluation results:
| Task | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) | STS-B (Pears/Spear) | SQuADv1.1 (Acc/F1) |
|------|--------------|------------|-------------|---------------------|--------------------|
| | 90.2/86.7 | 90.3 | 91.5 | 88.9/88.6 | 80.5/88.2 |
| {"language": "en"} | Intel/bert-base-uncased-mnli-sparse-70-unstructured-no-classifier | null | [
"transformers",
"pytorch",
"bert",
"fill-mask",
"en",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-classification | transformers |
# Sparse BERT base model fine tuned to MNLI (uncased)
Fine tuned sparse BERT base to MNLI (GLUE Benchmark) task from [bert-base-uncased-sparse-70-unstructured](https://huggingface.co/Intel/bert-base-uncased-sparse-70-unstructured).
<br><br>
Note: This model requires `transformers==2.10.0`
## Evaluation Results
Matched: 82.5%
Mismatched: 83.3%
This model can be further fine-tuned to other tasks and achieve the following evaluation results:
| Task | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) | STS-B (Pears/Spear) | SQuADv1.1 (Acc/F1) |
|------|--------------|------------|-------------|---------------------|--------------------|
| | 90.2/86.7 | 90.3 | 91.5 | 88.9/88.6 | 80.5/88.2 |
| {"language": "en"} | Intel/bert-base-uncased-mnli-sparse-70-unstructured | null | [
"transformers",
"pytorch",
"bert",
"text-classification",
"en",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | transformers |
# Sparse BERT base model (uncased)
Pretrained model pruned to 1:2 structured sparsity.
The model is a pruned version of the [BERT base model](https://huggingface.co/bert-base-uncased).
## Intended Use
The model can be used for fine-tuning to downstream tasks with sparsity already embeded to the model.
To keep the sparsity a mask should be added to each sparse weight blocking the optimizer from updating the zeros.
## Evaluation Results
We get the following results on the tasks development set, all results are mean of 5 different seeded models:
| Task | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) | STS-B (Pears/Spear) | SQuADv1.1 (Acc/F1) |
|------|--------------|---------------|--------------|------------|-------------|---------------------|--------------------|
| | 83.3 | 83.9 | 90.8/87.6 | 90.4 | 91.3 | 88.8/88.3 | 80.5/88.2 | | {"language": "en"} | Intel/bert-base-uncased-sparse-1_2 | null | [
"transformers",
"pytorch",
"bert",
"pretraining",
"en",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers |
# Sparse BERT base model (uncased)
Pretrained model pruned to 70% sparsity.
The model is a pruned version of the [BERT base model](https://huggingface.co/bert-base-uncased).
## Intended Use
The model can be used for fine-tuning to downstream tasks with sparsity already embeded to the model.
To keep the sparsity a mask should be added to each sparse weight blocking the optimizer from updating the zeros. | {"language": "en"} | Intel/bert-base-uncased-sparse-70-unstructured | null | [
"transformers",
"pytorch",
"bert",
"fill-mask",
"en",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers | ## Model Details: 85% Sparse BERT-Base (uncased) Prune Once for All
This model is a sparse pre-trained model that can be fine-tuned for a wide range of language tasks. The process of weight pruning is forcing some of the weights of the neural network to zero. Setting some of the weights to zero results in sparser matrices. Updating neural network weights does involve matrix multiplication, and if we can keep the matrices sparse while retaining enough important information, we can reduce the overall computational overhead. The term "sparse" in the title of the model indicates a ratio of sparsity in the weights; for more details, you can read [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754).
Visualization of Prunce Once for All method from [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754):
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Date | September 30, 2021 |
| Version | 1 |
| Type | NLP - General sparse language model |
| Architecture | "The method consists of two steps, teacher preparation and student pruning. The sparse pre-trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine-tune the sparse pre-trained models for several language tasks while we prune the pre-trained model only once." [(Zafrir et al., 2021)](https://arxiv.org/abs/2111.05754) |
| Paper or Other Resources | [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | This is a general sparse language model; in its current form, it is not ready for downstream prediction tasks, but it can be fine-tuned for several language tasks including (but not limited to) question-answering, genre natural language inference, and sentiment classification. |
| Primary intended users | Anyone who needs an efficient general language model for other downstream tasks. |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is an example of how to import this model in Python:
```python
import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/bert-base-uncased-sparse-85-unstructured-pruneofa')
```
For more code examples, refer to the [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
### Metrics (Model Performance):
| Model | Model Size | SQuADv1.1 (EM/F1) | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) |
|-------------------------------|:----------:|:-----------------:|:------------:|:-------------:|:------------:|:----------:|:-----------:|
| [80% Sparse BERT-Base uncased fine-tuned on SQuAD1.1](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa) | - | 81.29/88.47 | - | - | - | - | - |
| [**85% Sparse BERT-Base uncased**](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa) | Medium | 81.10/88.42 | 82.71 | 83.67 | 91.15/88.00 | 90.34 | 91.46 |
| [90% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa) | Medium | 79.83/87.25 | 81.45 | 82.43 | 90.93/87.72 | 89.07 | 90.88 |
| [90% Sparse BERT-Large uncased](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa) | Large | 83.35/90.20 | 83.74 | 84.20 | 91.48/88.43 | 91.39 | 92.95 |
| [85% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa) | Small | 78.10/85.82 | 81.35 | 82.03 | 90.29/86.97 | 88.31 | 90.60 |
| [90% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa) | Small | 76.91/84.82 | 80.68 | 81.47 | 90.05/86.67 | 87.66 | 90.02 |
All the results are the mean of two seperate experiments with the same hyper-parameters and different seeds.
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | [English Wikipedia Dataset](https://huggingface.co/datasets/wikipedia) (2500M words). |
| Motivation | To build an efficient and accurate base model for several downstream language tasks. |
| Preprocessing | "We use the English Wikipedia dataset (2500M words) for training the models on the pre-training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers ([Devlin et al., 2019](https://arxiv.org/abs/1810.04805), [Sanh et al., 2019](https://arxiv.org/abs/1910.01108)). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1." |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@article{zafrir2021prune,
title={Prune Once for All: Sparse Pre-Trained Language Models},
author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
journal={arXiv preprint arXiv:2111.05754},
year={2021}
}
``` | {"language": "en", "license": "apache-2.0", "tags": ["fill-mask"], "datasets": ["wikipedia", "bookcorpus"]} | Intel/bert-base-uncased-sparse-85-unstructured-pruneofa | null | [
"transformers",
"pytorch",
"tf",
"bert",
"pretraining",
"fill-mask",
"en",
"dataset:wikipedia",
"dataset:bookcorpus",
"arxiv:2111.05754",
"arxiv:1810.04805",
"arxiv:1910.01108",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers | ## Model Details: 90% Sparse BERT-Base (uncased) Prune Once for All
This model is a sparse pre-trained model that can be fine-tuned for a wide range of language tasks. The process of weight pruning is forcing some of the weights of the neural network to zero. Setting some of the weights to zero results in sparser matrices. Updating neural network weights does involve matrix multiplication, and if we can keep the matrices sparse while retaining enough important information, we can reduce the overall computational overhead. The term "sparse" in the title of the model indicates a ratio of sparsity in the weights; for more details, you can read [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754).
Visualization of Prunce Once for All method from [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754):
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Date | September 30, 2021 |
| Version | 1 |
| Type | NLP - General sparse language model |
| Architecture | "The method consists of two steps, teacher preparation and student pruning. The sparse pre-trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine-tune the sparse pre-trained models for several language tasks while we prune the pre-trained model only once." [(Zafrir et al., 2021)](https://arxiv.org/abs/2111.05754) |
| Paper or Other Resources | [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | This is a general sparse language model; in its current form, it is not ready for downstream prediction tasks, but it can be fine-tuned for several language tasks including (but not limited to) question-answering, genre natural language inference, and sentiment classification. |
| Primary intended users | Anyone who needs an efficient general language model for other downstream tasks. |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is an example of how to import this model in Python:
```python
import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/bert-base-uncased-sparse-90-unstructured-pruneofa')
```
For more code examples, refer to the [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
### Metrics (Model Performance):
| Model | Model Size | SQuADv1.1 (EM/F1) | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) |
|-------------------------------|:----------:|:-----------------:|:------------:|:-------------:|:------------:|:----------:|:-----------:|
| [80% Sparse BERT-Base uncased fine-tuned on SQuAD1.1](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa) | - | 81.29/88.47 | - | - | - | - | - |
| [85% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa) | Medium | 81.10/88.42 | 82.71 | 83.67 | 91.15/88.00 | 90.34 | 91.46 |
| [**90% Sparse BERT-Base uncased**](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa) | Medium | 79.83/87.25 | 81.45 | 82.43 | 90.93/87.72 | 89.07 | 90.88 |
| [90% Sparse BERT-Large uncased](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa) | Large | 83.35/90.20 | 83.74 | 84.20 | 91.48/88.43 | 91.39 | 92.95 |
| [85% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa) | Small | 78.10/85.82 | 81.35 | 82.03 | 90.29/86.97 | 88.31 | 90.60 |
| [90% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa) | Small | 76.91/84.82 | 80.68 | 81.47 | 90.05/86.67 | 87.66 | 90.02 |
All the results are the mean of two seperate experiments with the same hyper-parameters and different seeds.
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | [English Wikipedia Dataset](https://huggingface.co/datasets/wikipedia) (2500M words). |
| Motivation | To build an efficient and accurate base model for several downstream language tasks. |
| Preprocessing | "We use the English Wikipedia dataset (2500M words) for training the models on the pre-training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers ([Devlin et al., 2019](https://arxiv.org/abs/1810.04805), [Sanh et al., 2019](https://arxiv.org/abs/1910.01108)). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1." |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@article{zafrir2021prune,
title={Prune Once for All: Sparse Pre-Trained Language Models},
author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
journal={arXiv preprint arXiv:2111.05754},
year={2021}
}
```
| {"language": "en", "license": "apache-2.0", "tags": ["fill-mask", "bert"], "datasets": ["wikipedia", "bookcorpus"]} | Intel/bert-base-uncased-sparse-90-unstructured-pruneofa | null | [
"transformers",
"pytorch",
"tf",
"bert",
"pretraining",
"fill-mask",
"en",
"dataset:wikipedia",
"dataset:bookcorpus",
"arxiv:2111.05754",
"arxiv:1810.04805",
"arxiv:1910.01108",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
question-answering | transformers | ## Model Details: 80% 1x4 Block Sparse BERT-Base (uncased) Fine Tuned on SQuADv1.1
This model has been fine-tuned for the NLP task of question answering, trained on the SQuAD 1.1 dataset. It is a result of fine-tuning a Prune Once For All 80% 1x4 block sparse pre-trained BERT-Base model, combined with knowledge distillation.
> We present a new method for training sparse pre-trained Transformer language models by integrating weight pruning and model distillation. These sparse pre-trained models can be used to transfer learning for a wide range of tasks while maintaining their sparsity pattern. We show how the compressed sparse pre-trained models we trained transfer their knowledge to five different downstream natural language tasks with minimal accuracy loss.
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Model Card Authors | Intel |
| Date | February 27, 2022 |
| Version | 1 |
| Type | NLP - Question Answering |
| Architecture | "The method consists of two steps, teacher preparation and student pruning. The sparse pre-trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine-tune the sparse pre-trained models for several language tasks while we prune the pre-trained model only once." [(Zafrir et al., 2021)](https://arxiv.org/abs/2111.05754) |
| Paper or Other Resources | [Paper: Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
Visualization of Prunce Once for All method from [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754). More details can be found in their paper.
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | You can use the model for the NLP task of question answering: given a corpus of text, you can ask it a question about that text, and it will find the answer in the text. |
| Primary intended users | Anyone doing question answering |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is how to import this model in Python:
```python
import transformers
import model_compression_research as model_comp
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa')
scheduler = mcr.pruning_scheduler_factory(model, '../../examples/transformers/question-answering/config/lock_config.json')
# Train your model...
scheduler.remove_pruning()
```
For more code examples, refer to the [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
### Metrics (Model Performance):
| Model | Model Size | SQuADv1.1 (EM/F1) | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) |
|-------------------------------|:----------:|:-----------------:|:------------:|:-------------:|:------------:|:----------:|:-----------:|
| [**80% 1x4 Block Sparse BERT-Base uncased**](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa) | - | 81.29/88.47 | - | - | - | - | - |
| [85% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa) | Medium | 81.10/88.42 | 82.71 | 83.67 | 91.15/88.00 | 90.34 | 91.46 |
| [90% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa) | Medium | 79.83/87.25 | 81.45 | 82.43 | 90.93/87.72 | 89.07 | 90.88 |
| [90% Sparse BERT-Large uncased](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa) | Large | 83.35/90.20 | 83.74 | 84.20 | 91.48/88.43 | 91.39 | 92.95 |
| [85% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa) | Small | 78.10/85.82 | 81.35 | 82.03 | 90.29/86.97 | 88.31 | 90.60 |
| [90% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa) | Small | 76.91/84.82 | 80.68 | 81.47 | 90.05/86.67 | 87.66 | 90.02 |
All the results are the mean of two seperate experiments with the same hyper-parameters and different seeds.
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | SQuAD1.1: "Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable." (https://huggingface.co/datasets/squad)|
| Motivation | To build an efficient and accurate model for the question answering task. |
| Preprocessing | "We use the English Wikipedia dataset (2500M words) for training the models on the pre-training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers ([Devlin et al., 2019](https://arxiv.org/abs/1810.04805), [Sanh et al., 2019](https://arxiv.org/abs/1910.01108)). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1." Following the pre-training on Wikipedia, fine-tuning is completed on the SQuAD1.1 dataset. |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@article{zafrir2021prune,
title={Prune Once for All: Sparse Pre-Trained Language Models},
author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
journal={arXiv preprint arXiv:2111.05754},
year={2021}
}
``` | {"language": "en", "license": "apache-2.0", "tags": ["question-answering", "bert"], "datasets": ["squad"]} | Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa | null | [
"transformers",
"pytorch",
"bert",
"question-answering",
"en",
"dataset:squad",
"arxiv:2111.05754",
"arxiv:1810.04805",
"arxiv:1910.01108",
"license:apache-2.0",
"model-index",
"endpoints_compatible",
"has_space",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers | ## Model Details: 90% Sparse BERT-Large (uncased) Prune Once for All
This model is a sparse pre-trained model that can be fine-tuned for a wide range of language tasks. The process of weight pruning is forcing some of the weights of the neural network to zero. Setting some of the weights to zero results in sparser matrices. Updating neural network weights does involve matrix multiplication, and if we can keep the matrices sparse while retaining enough important information, we can reduce the overall computational overhead. The term "sparse" in the title of the model indicates a ratio of sparsity in the weights; for more details, you can read [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754).
Visualization of Prunce Once for All method from [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754):
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Date | September 30, 2021 |
| Version | 1 |
| Type | NLP - General sparse language model |
| Architecture | "The method consists of two steps, teacher preparation and student pruning. The sparse pre-trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine-tune the sparse pre-trained models for several language tasks while we prune the pre-trained model only once." [(Zafrir et al., 2021)](https://arxiv.org/abs/2111.05754) |
| Paper or Other Resources | [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | This is a general sparse language model; in its current form, it is not ready for downstream prediction tasks, but it can be fine-tuned for several language tasks including (but not limited to) question-answering, genre natural language inference, and sentiment classification. |
| Primary intended users | Anyone who needs an efficient general language model for other downstream tasks. |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is an example of how to import this model in Python:
```python
import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/bert-large-uncased-sparse-90-unstructured-pruneofa')
```
For more code examples, refer to the [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
### Metrics (Model Performance):
| Model | Model Size | SQuADv1.1 (EM/F1) | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) |
|-------------------------------|:----------:|:-----------------:|:------------:|:-------------:|:------------:|:----------:|:-----------:|
| [80% Sparse BERT-Base uncased fine-tuned on SQuAD1.1](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa) | - | 81.29/88.47 | - | - | - | - | - |
| [85% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa) | Medium | 81.10/88.42 | 82.71 | 83.67 | 91.15/88.00 | 90.34 | 91.46 |
| [90% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa) | Medium | 79.83/87.25 | 81.45 | 82.43 | 90.93/87.72 | 89.07 | 90.88 |
| [**90% Sparse BERT-Large uncased**](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa) | Large | 83.35/90.20 | 83.74 | 84.20 | 91.48/88.43 | 91.39 | 92.95 |
| [85% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa) | Small | 78.10/85.82 | 81.35 | 82.03 | 90.29/86.97 | 88.31 | 90.60 |
| [90% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa) | Small | 76.91/84.82 | 80.68 | 81.47 | 90.05/86.67 | 87.66 | 90.02 |
All the results are the mean of two seperate experiments with the same hyper-parameters and different seeds.
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | [English Wikipedia Dataset](https://huggingface.co/datasets/wikipedia) (2500M words). |
| Motivation | To build an efficient and accurate base model for several downstream language tasks. |
| Preprocessing | "We use the English Wikipedia dataset (2500M words) for training the models on the pre-training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers ([Devlin et al., 2019](https://arxiv.org/abs/1810.04805), [Sanh et al., 2019](https://arxiv.org/abs/1910.01108)). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1." |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@article{zafrir2021prune,
title={Prune Once for All: Sparse Pre-Trained Language Models},
author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
journal={arXiv preprint arXiv:2111.05754},
year={2021}
}
```
| {"language": "en", "license": "apache-2.0", "tags": ["fill-mask"], "datasets": ["wikipedia", "bookcorpus"]} | Intel/bert-large-uncased-sparse-90-unstructured-pruneofa | null | [
"transformers",
"pytorch",
"tf",
"bert",
"pretraining",
"fill-mask",
"en",
"dataset:wikipedia",
"dataset:bookcorpus",
"arxiv:2111.05754",
"arxiv:1810.04805",
"arxiv:1910.01108",
"license:apache-2.0",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
question-answering | transformers | # 90% Sparse BERT-Large (uncased) Fine Tuned on SQuADv1.1
This model is a result of fine-tuning a Prune OFA 90% sparse pre-trained BERT-Large combined with knowledge distillation.
This model yields the following results on SQuADv1.1 development set:<br>
`{"exact_match": 83.56669820245979, "f1": 90.20829352733487}`
For further details see our paper, [Prune Once for All: Sparse Pre-Trained Language Models](https://arxiv.org/abs/2111.05754), and our open source implementation available [here](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
| {"language": "en"} | Intel/bert-large-uncased-squadv1.1-sparse-90-unstructured | null | [
"transformers",
"pytorch",
"tf",
"bert",
"question-answering",
"en",
"arxiv:2111.05754",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers | ## Model Details: 85% Sparse DistilBERT-Base (uncased) Prune Once for All
This model is a sparse pre-trained model that can be fine-tuned for a wide range of language tasks. The process of weight pruning is forcing some of the weights of the neural network to zero. Setting some of the weights to zero results in sparser matrices. Updating neural network weights does involve matrix multiplication, and if we can keep the matrices sparse while retaining enough important information, we can reduce the overall computational overhead. The term "sparse" in the title of the model indicates a ratio of sparsity in the weights; for more details, you can read [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754).
Visualization of Prunce Once for All method from [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754):
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Date | September 30, 2021 |
| Version | 1 |
| Type | NLP - General sparse language model |
| Architecture | "The method consists of two steps, teacher preparation and student pruning. The sparse pre-trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine-tune the sparse pre-trained models for several language tasks while we prune the pre-trained model only once." [(Zafrir et al., 2021)](https://arxiv.org/abs/2111.05754) |
| Paper or Other Resources | [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | This is a general sparse language model; in its current form, it is not ready for downstream prediction tasks, but it can be fine-tuned for several language tasks including (but not limited to) question-answering, genre natural language inference, and sentiment classification. |
| Primary intended users | Anyone who needs an efficient general language model for other downstream tasks. |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is an example of how to import this model in Python:
```python
import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa')
```
For more code examples, refer to the [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
### Metrics (Model Performance):
| Model | Model Size | SQuADv1.1 (EM/F1) | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) |
|-------------------------------|:----------:|:-----------------:|:------------:|:-------------:|:------------:|:----------:|:-----------:|
| [80% Sparse BERT-Base uncased fine-tuned on SQuAD1.1](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa) | - | 81.29/88.47 | - | - | - | - | - |
| [85% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa) | Medium | 81.10/88.42 | 82.71 | 83.67 | 91.15/88.00 | 90.34 | 91.46 |
| [90% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa) | Medium | 79.83/87.25 | 81.45 | 82.43 | 90.93/87.72 | 89.07 | 90.88 |
| [90% Sparse BERT-Large uncased](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa) | Large | 83.35/90.20 | 83.74 | 84.20 | 91.48/88.43 | 91.39 | 92.95 |
| [**85% Sparse DistilBERT uncased**](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa) | Small | 78.10/85.82 | 81.35 | 82.03 | 90.29/86.97 | 88.31 | 90.60 |
| [90% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa) | Small | 76.91/84.82 | 80.68 | 81.47 | 90.05/86.67 | 87.66 | 90.02 |
All the results are the mean of two seperate experiments with the same hyper-parameters and different seeds.
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | [English Wikipedia Dataset](https://huggingface.co/datasets/wikipedia) (2500M words). |
| Motivation | To build an efficient and accurate base model for several downstream language tasks. |
| Preprocessing | "We use the English Wikipedia dataset (2500M words) for training the models on the pre-training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers ([Devlin et al., 2019](https://arxiv.org/abs/1810.04805), [Sanh et al., 2019](https://arxiv.org/abs/1910.01108)). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1." |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@article{zafrir2021prune,
title={Prune Once for All: Sparse Pre-Trained Language Models},
author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
journal={arXiv preprint arXiv:2111.05754},
year={2021}
}
```
| {"language": "en", "license": "apache-2.0", "datasets": ["wikipedia"]} | Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa | null | [
"transformers",
"pytorch",
"tf",
"distilbert",
"fill-mask",
"en",
"dataset:wikipedia",
"arxiv:2111.05754",
"arxiv:1810.04805",
"arxiv:1910.01108",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
fill-mask | transformers | ### Model Details: 90% Sparse DistilBERT-Base (uncased) Prune Once for All
This model is a sparse pre-trained model that can be fine-tuned for a wide range of language tasks. The process of weight pruning is forcing some of the weights of the neural network to zero. Setting some of the weights to zero results in sparser matrices. Updating neural network weights does involve matrix multiplication, and if we can keep the matrices sparse while retaining enough important information, we can reduce the overall computational overhead. The term "sparse" in the title of the model indicates a ratio of sparsity in the weights; for more details, you can read [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754).
Visualization of Prunce Once for All method from [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754):
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Date | September 30, 2021 |
| Version | 1 |
| Type | NLP - General sparse language model |
| Architecture | "The method consists of two steps, teacher preparation and student pruning. The sparse pre-trained model we trained is the model we use for transfer learning while maintaining its sparsity pattern. We call the method Prune Once for All since we show how to fine-tune the sparse pre-trained models for several language tasks while we prune the pre-trained model only once." [(Zafrir et al., 2021)](https://arxiv.org/abs/2111.05754) |
| Paper or Other Resources | [Zafrir et al. (2021)](https://arxiv.org/abs/2111.05754); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | This is a general sparse language model; in its current form, it is not ready for downstream prediction tasks, but it can be fine-tuned for several language tasks including (but not limited to) question-answering, genre natural language inference, and sentiment classification. |
| Primary intended users | Anyone who needs an efficient general language model for other downstream tasks. |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is an example of how to import this model in Python:
```python
import transformers
model = transformers.AutoModelForQuestionAnswering.from_pretrained('Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa')
```
For more code examples, refer to the [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package/tree/main/research/prune-once-for-all).
### Metrics (Model Performance):
| Model | Model Size | SQuADv1.1 (EM/F1) | MNLI-m (Acc) | MNLI-mm (Acc) | QQP (Acc/F1) | QNLI (Acc) | SST-2 (Acc) |
|-------------------------------|:----------:|:-----------------:|:------------:|:-------------:|:------------:|:----------:|:-----------:|
| [80% Sparse BERT-Base uncased fine-tuned on SQuAD1.1](https://huggingface.co/Intel/bert-base-uncased-squadv1.1-sparse-80-1x4-block-pruneofa) | - | 81.29/88.47 | - | - | - | - | - |
| [85% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-85-unstructured-pruneofa) | Medium | 81.10/88.42 | 82.71 | 83.67 | 91.15/88.00 | 90.34 | 91.46 |
| [90% Sparse BERT-Base uncased](https://huggingface.co/Intel/bert-base-uncased-sparse-90-unstructured-pruneofa) | Medium | 79.83/87.25 | 81.45 | 82.43 | 90.93/87.72 | 89.07 | 90.88 |
| [90% Sparse BERT-Large uncased](https://huggingface.co/Intel/bert-large-uncased-sparse-90-unstructured-pruneofa) | Large | 83.35/90.20 | 83.74 | 84.20 | 91.48/88.43 | 91.39 | 92.95 |
| [85% Sparse DistilBERT uncased](https://huggingface.co/Intel/distilbert-base-uncased-sparse-85-unstructured-pruneofa) | Small | 78.10/85.82 | 81.35 | 82.03 | 90.29/86.97 | 88.31 | 90.60 |
| [**90% Sparse DistilBERT uncased**](https://huggingface.co/Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa) | Small | 76.91/84.82 | 80.68 | 81.47 | 90.05/86.67 | 87.66 | 90.02 |
All the results are the mean of two seperate experiments with the same hyper-parameters and different seeds.
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | [English Wikipedia Dataset](https://huggingface.co/datasets/wikipedia) (2500M words). |
| Motivation | To build an efficient and accurate base model for several downstream language tasks. |
| Preprocessing | "We use the English Wikipedia dataset (2500M words) for training the models on the pre-training task. We split the data into train (95%) and validation (5%) sets. Both sets are preprocessed as described in the models’ original papers ([Devlin et al., 2019](https://arxiv.org/abs/1810.04805), [Sanh et al., 2019](https://arxiv.org/abs/1910.01108)). We process the data to use the maximum sequence length allowed by the models, however, we allow shorter sequences at a probability of 0:1." |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@article{zafrir2021prune,
title={Prune Once for All: Sparse Pre-Trained Language Models},
author={Zafrir, Ofir and Larey, Ariel and Boudoukh, Guy and Shen, Haihao and Wasserblat, Moshe},
journal={arXiv preprint arXiv:2111.05754},
year={2021}
}
``` | {"language": "en", "license": "apache-2.0", "datasets": ["wikipedia"]} | Intel/distilbert-base-uncased-sparse-90-unstructured-pruneofa | null | [
"transformers",
"pytorch",
"tf",
"distilbert",
"fill-mask",
"en",
"dataset:wikipedia",
"arxiv:2111.05754",
"arxiv:1810.04805",
"arxiv:1910.01108",
"license:apache-2.0",
"autotrain_compatible",
"endpoints_compatible",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
question-answering | transformers |
## Model Details: Dynamic-TinyBERT: Boost TinyBERT's Inference Efficiency by Dynamic Sequence Length
Dynamic-TinyBERT has been fine-tuned for the NLP task of question answering, trained on the SQuAD 1.1 dataset. [Guskin et al. (2021)](https://neurips2021-nlp.github.io/papers/16/CameraReady/Dynamic_TinyBERT_NLSP2021_camera_ready.pdf) note:
> Dynamic-TinyBERT is a TinyBERT model that utilizes sequence-length reduction and Hyperparameter Optimization for enhanced inference efficiency per any computational budget. Dynamic-TinyBERT is trained only once, performing on-par with BERT and achieving an accuracy-speedup trade-off superior to any other efficient approaches (up to 3.3x with <1% loss-drop).
| Model Detail | Description |
| ----------- | ----------- |
| Model Authors - Company | Intel |
| Model Card Authors | Intel in collaboration with Hugging Face |
| Date | November 22, 2021 |
| Version | 1 |
| Type | NLP - Question Answering |
| Architecture | "For our Dynamic-TinyBERT model we use the architecture of TinyBERT6L: a small BERT model with 6 layers, a hidden size of 768, a feed forward size of 3072 and 12 heads." [Guskin et al. (2021)](https://gyuwankim.github.io/publication/dynamic-tinybert/poster.pdf) |
| Paper or Other Resources | [Paper](https://neurips2021-nlp.github.io/papers/16/CameraReady/Dynamic_TinyBERT_NLSP2021_camera_ready.pdf); [Poster](https://gyuwankim.github.io/publication/dynamic-tinybert/poster.pdf); [GitHub Repo](https://github.com/IntelLabs/Model-Compression-Research-Package) |
| License | Apache 2.0 |
| Questions or Comments | [Community Tab](https://huggingface.co/Intel/dynamic_tinybert/discussions) and [Intel Developers Discord](https://discord.gg/rv2Gp55UJQ)|
| Intended Use | Description |
| ----------- | ----------- |
| Primary intended uses | You can use the model for the NLP task of question answering: given a corpus of text, you can ask it a question about that text, and it will find the answer in the text. |
| Primary intended users | Anyone doing question answering |
| Out-of-scope uses | The model should not be used to intentionally create hostile or alienating environments for people.|
### How to use
Here is how to import this model in Python:
<details>
<summary> Click to expand </summary>
```python
from transformers import AutoTokenizer, AutoModelForQuestionAnswering
tokenizer = AutoTokenizer.from_pretrained("Intel/dynamic_tinybert")
model = AutoModelForQuestionAnswering.from_pretrained("Intel/dynamic_tinybert")
```
</details>
| Factors | Description |
| ----------- | ----------- |
| Groups | Many Wikipedia articles with question and answer labels are contained in the training data |
| Instrumentation | - |
| Environment | Training was completed on a Titan GPU. |
| Card Prompts | Model deployment on alternate hardware and software will change model performance |
| Metrics | Description |
| ----------- | ----------- |
| Model performance measures | F1 |
| Decision thresholds | - |
| Approaches to uncertainty and variability | - |
| Training and Evaluation Data | Description |
| ----------- | ----------- |
| Datasets | SQuAD1.1: "Stanford Question Answering Dataset (SQuAD) is a reading comprehension dataset, consisting of questions posed by crowdworkers on a set of Wikipedia articles, where the answer to every question is a segment of text, or span, from the corresponding reading passage, or the question might be unanswerable." (https://huggingface.co/datasets/squad)|
| Motivation | To build an efficient and accurate model for the question answering task. |
| Preprocessing | "We start with a pre-trained general-TinyBERT student, which was trained to learn the general knowledge of BERT using the general-distillation method presented by TinyBERT. We perform transformer distillation from a fine- tuned BERT teacher to the student, following the same training steps used in the original TinyBERT: (1) intermediate-layer distillation (ID) — learning the knowledge residing in the hidden states and attentions matrices, and (2) prediction-layer distillation (PD) — fitting the predictions of the teacher." ([Guskin et al., 2021](https://neurips2021-nlp.github.io/papers/16/CameraReady/Dynamic_TinyBERT_NLSP2021_camera_ready.pdf))|
Model Performance Analysis:
| Model | Max F1 (full model) | Best Speedup within BERT-1% |
|------------------|---------------------|-----------------------------|
| Dynamic-TinyBERT | 88.71 | 3.3x |
| Ethical Considerations | Description |
| ----------- | ----------- |
| Data | The training data come from Wikipedia articles |
| Human life | The model is not intended to inform decisions central to human life or flourishing. It is an aggregated set of labelled Wikipedia articles. |
| Mitigations | No additional risk mitigation strategies were considered during model development. |
| Risks and harms | Significant research has explored bias and fairness issues with language models (see, e.g., [Sheng et al., 2021](https://aclanthology.org/2021.acl-long.330.pdf), and [Bender et al., 2021](https://dl.acm.org/doi/pdf/10.1145/3442188.3445922)). Predictions generated by the model may include disturbing and harmful stereotypes across protected classes; identity characteristics; and sensitive, social, and occupational groups. Beyond this, the extent of the risks involved by using the model remain unknown.|
| Use cases | - |
| Caveats and Recommendations |
| ----------- |
| Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. There are no additional caveats or recommendations for this model. |
### BibTeX entry and citation info
```bibtex
@misc{https://doi.org/10.48550/arxiv.2111.09645,
doi = {10.48550/ARXIV.2111.09645},
url = {https://arxiv.org/abs/2111.09645},
author = {Guskin, Shira and Wasserblat, Moshe and Ding, Ke and Kim, Gyuwan},
keywords = {Computation and Language (cs.CL), Machine Learning (cs.LG), FOS: Computer and information sciences, FOS: Computer and information sciences},
title = {Dynamic-TinyBERT: Boost TinyBERT's Inference Efficiency by Dynamic Sequence Length},
publisher = {arXiv},
year = {2021},
``` | {"language": ["en"], "license": "apache-2.0", "tags": ["question-answering", "bert"], "datasets": ["squad"]} | Intel/dynamic_tinybert | null | [
"transformers",
"pytorch",
"bert",
"question-answering",
"en",
"dataset:squad",
"arxiv:2111.09645",
"license:apache-2.0",
"model-index",
"endpoints_compatible",
"has_space",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers | #harry potter | {"tags": ["conversational"]} | Invincible/Chat_bot-Harrypotter-medium | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | transformers |
#harry potter Model | {"tags": ["conversational"]} | Invincible/Chat_bot-Harrypotter-small | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"conversational",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
text-generation | null | #Harry Potter DialoDPT Model | {"tags": ["conversational"]} | Invincible/DialoGPT-medium-harryPotter | null | [
"conversational",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
null | null | {} | Invincible/Harry-potter-chatbot-medium | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Iremtop/dash | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | Irene/blenderbot | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers | {} | Irina/Fairytale | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers | {} | Irina/cyoa_GPT3Medium | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"endpoints_compatible",
"has_space",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers | {} | Irina/fantasy_GPT3Medium | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers | {} | Irina/trans_GPT3Medium | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers | {} | Irina/trans_cyoa_GPT3Medium | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
text-generation | transformers | {} | Irina/trans_cyoa_rollouted | null | [
"transformers",
"pytorch",
"gpt2",
"text-generation",
"autotrain_compatible",
"endpoints_compatible",
"text-generation-inference",
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | IrinaLee/roberta-base-finetuned-ynat | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
|
null | null | {} | IronManProject/DialoGPT-medium-IronMan | null | [
"region:us"
] | null | 2022-03-02T23:29:04+00:00 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.