Files changed (1) hide show
  1. README.md +131 -11
README.md CHANGED
@@ -13,9 +13,70 @@ tags:
13
  license: apache-2.0
14
  ---
15
 
16
- [Google's T5](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html)
17
 
18
- ## PreTraining
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
 
20
  The model was pre-trained on a on a **multi-task mixture of unsupervised (1.) and supervised tasks (2.)**.
21
  Thereby, the following datasets were being used for (1.) and (2.):
@@ -50,22 +111,81 @@ Thereby, the following datasets were being used for (1.) and (2.):
50
  - ReCoRD [Zhang et al., 2018](https://arxiv.org/abs/1810.12885)
51
  - BoolQ [Clark et al., 2019](https://arxiv.org/abs/1905.10044)
52
 
53
- ## All T5 checkpoints
54
 
55
- Other Community Checkpoints: [here](https://huggingface.co/models?search=t5)
56
 
57
- ## Paper
58
 
59
- For more information, please take a look at the original paper.
60
 
61
- Paper: [Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer](https://arxiv.org/pdf/1910.10683.pdf)
62
 
63
- Authors: *Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. Liu*
64
 
 
65
 
66
- **Abstract**
67
 
68
- Transfer learning, where a model is first pre-trained on a data-rich task before being fine-tuned on a downstream task, has emerged as a powerful technique in natural language processing (NLP). The effectiveness of transfer learning has given rise to a diversity of approaches, methodology, and practice. In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts every language problem into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled datasets, transfer approaches, and other factors on dozens of language understanding tasks. By combining the insights from our exploration with scale and our new “Colossal Clean Crawled Corpus”, we achieve state-of-the-art results on many benchmarks covering summarization, question answering, text classification, and more. To facilitate future work on transfer learning for NLP, we release our dataset, pre-trained models, and code.
69
 
70
- ![model image](https://camo.githubusercontent.com/623b4dea0b653f2ad3f36c71ebfe749a677ac0a1/68747470733a2f2f6d69726f2e6d656469756d2e636f6d2f6d61782f343030362f312a44304a31674e51663876727255704b657944387750412e706e67)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
71
 
 
13
  license: apache-2.0
14
  ---
15
 
16
+ # Model Card for T5 Small
17
 
18
+ ![model image](https://camo.githubusercontent.com/623b4dea0b653f2ad3f36c71ebfe749a677ac0a1/68747470733a2f2f6d69726f2e6d656469756d2e636f6d2f6d61782f343030362f312a44304a31674e51663876727255704b657944387750412e706e67)
19
+
20
+ # Table of Contents
21
+
22
+ 1. [Model Details](#model-details)
23
+ 2. [Uses](#uses)
24
+ 3. [Bias, Risks, and Limitations](#bias-risks-and-limitations)
25
+ 4. [Training Details](#training-details)
26
+ 5. [Evaluation](#evaluation)
27
+ 6. [Environmental Impact](#environmental-impact)
28
+ 7. [Citation](#citation)
29
+ 8. [Model Card Authors](#model-card-authors)
30
+ 9. [How To Get Started With the Model](#how-to-get-started-with-the-model)
31
+
32
+ # Model Details
33
+
34
+ ## Model Description
35
+
36
+ The developers of the Text-To-Text Transfer Transformer (T5) [write](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html):
37
+
38
+ > With T5, we propose reframing all NLP tasks into a unified text-to-text-format where the input and output are always text strings, in contrast to BERT-style models that can only output either a class label or a span of the input. Our text-to-text framework allows us to use the same model, loss function, and hyperparameters on any NLP task.
39
+
40
+ T5-Small is the checkpoint with 60 million parameters.
41
+
42
+ - **Developed by:** Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, Peter J. Liu. See [associated paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf) and [GitHub repo](https://github.com/google-research/text-to-text-transfer-transformer#released-model-checkpoints)
43
+ - **Model type:** Language model
44
+ - **Language(s) (NLP):** English, French, Romanian, German
45
+ - **License:** Apache 2.0
46
+ - **Related Models:** [All T5 Checkpoints](https://huggingface.co/models?search=t5)
47
+ - **Resources for more information:**
48
+ - [Research paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf)
49
+ - [Google's T5 Blog Post](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html)
50
+ - [GitHub Repo](https://github.com/google-research/text-to-text-transfer-transformer)
51
+ - [Hugging Face T5 Docs](https://huggingface.co/docs/transformers/model_doc/t5)
52
+
53
+ # Uses
54
+
55
+ ## Direct Use and Downstream Use
56
+
57
+ The developers write in a [blog post](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) that the model:
58
+
59
+ > Our text-to-text framework allows us to use the same model, loss function, and hyperparameters on any NLP task, including machine translation, document summarization, question answering, and classification tasks (e.g., sentiment analysis). We can even apply T5 to regression tasks by training it to predict the string representation of a number instead of the number itself.
60
+
61
+ See the [blog post](https://ai.googleblog.com/2020/02/exploring-transfer-learning-with-t5.html) and [research paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf) for further details.
62
+
63
+ ## Out-of-Scope Use
64
+
65
+ More information needed.
66
+
67
+ # Bias, Risks, and Limitations
68
+
69
+ More information needed.
70
+
71
+ ## Recommendations
72
+
73
+ More information needed.
74
+
75
+ # Training Details
76
+
77
+ ## Training Data
78
+
79
+ The model is pre-trained on the [Colossal Clean Crawled Corpus (C4)](https://www.tensorflow.org/datasets/catalog/c4), which was developed and released in the context of the same [research paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf) as T5.
80
 
81
  The model was pre-trained on a on a **multi-task mixture of unsupervised (1.) and supervised tasks (2.)**.
82
  Thereby, the following datasets were being used for (1.) and (2.):
 
111
  - ReCoRD [Zhang et al., 2018](https://arxiv.org/abs/1810.12885)
112
  - BoolQ [Clark et al., 2019](https://arxiv.org/abs/1905.10044)
113
 
114
+ ## Training Procedure
115
 
116
+ In their [abstract](https://jmlr.org/papers/volume21/20-074/20-074.pdf), the model developers write:
117
 
118
+ > In this paper, we explore the landscape of transfer learning techniques for NLP by introducing a unified framework that converts every language problem into a text-to-text format. Our systematic study compares pre-training objectives, architectures, unlabeled datasets, transfer approaches, and other factors on dozens of language understanding tasks.
119
 
120
+ The framework introduced, the T5 framework, involves a training procedure that brings together the approaches studied in the paper. See the [research paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf) for further details.
121
 
122
+ # Evaluation
123
 
124
+ ## Testing Data, Factors & Metrics
125
 
126
+ The developers evaluated the model on 24 tasks, see the [research paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf) for full details.
127
 
128
+ ## Results
129
 
130
+ For full results for T5-small, see the [research paper](https://jmlr.org/papers/volume21/20-074/20-074.pdf), Table 14.
131
 
132
+ # Environmental Impact
133
+
134
+ Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
135
+
136
+ - **Hardware Type:** Google Cloud TPU Pods
137
+ - **Hours used:** More information needed
138
+ - **Cloud Provider:** GCP
139
+ - **Compute Region:** More information needed
140
+ - **Carbon Emitted:** More information needed
141
+
142
+ # Citation
143
+
144
+ **BibTeX:**
145
+
146
+ ```bibtex
147
+ @article{2020t5,
148
+ author = {Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu},
149
+ title = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer},
150
+ journal = {Journal of Machine Learning Research},
151
+ year = {2020},
152
+ volume = {21},
153
+ number = {140},
154
+ pages = {1-67},
155
+ url = {http://jmlr.org/papers/v21/20-074.html}
156
+ }
157
+ ```
158
+
159
+ **APA:**
160
+ - Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., ... & Liu, P. J. (2020). Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21(140), 1-67.
161
+
162
+ # Model Card Authors
163
+
164
+ This model card was written by the team at Hugging Face.
165
+
166
+ # How to Get Started with the Model
167
+
168
+ Use the code below to get started with the model.
169
+
170
+ <details>
171
+ <summary> Click to expand </summary>
172
+
173
+ ```python
174
+ from transformers import T5Tokenizer, T5Model
175
+
176
+ tokenizer = T5Tokenizer.from_pretrained("t5-small")
177
+ model = T5Model.from_pretrained("t5-small")
178
+
179
+ input_ids = tokenizer(
180
+ "Studies have been shown that owning a dog is good for you", return_tensors="pt"
181
+ ).input_ids # Batch size 1
182
+ decoder_input_ids = tokenizer("Studies show that", return_tensors="pt").input_ids # Batch size 1
183
+
184
+ # forward pass
185
+ outputs = model(input_ids=input_ids, decoder_input_ids=decoder_input_ids)
186
+ last_hidden_states = outputs.last_hidden_state
187
+ ```
188
+
189
+ See the [Hugging Face T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Model) docs and a [Colab Notebook](https://colab.research.google.com/github/google-research/text-to-text-transfer-transformer/blob/main/notebooks/t5-trivia.ipynb) created by the model developers for more examples.
190
+ </details>
191