model documentation

#2
by nazneen - opened
Files changed (1) hide show
  1. README.md +209 -0
README.md ADDED
@@ -0,0 +1,209 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - text-2-text-generation
4
+ - bart
5
+ ---
6
+
7
+ # Model Card for brio-cnndm-uncased
8
+
9
+ # Model Details
10
+
11
+
12
+ ## Model Description
13
+
14
+ Abstractive summarization models are commonly trained using maximum likelihood estimation, which assumes a deterministic (one-point) target distribution in which an ideal model will assign all the probability mass to the reference summary. This assumption may lead to performance degradation during inference, where the model needs to compare several system-generated (candidate) summaries that have deviated from the reference summary. To address this problem, we propose a novel training paradigm which assumes a non-deterministic distribution so that different candidate summaries are assigned probability mass according to their quality.
15
+
16
+ - **Developed by:** Yale LILY Lab
17
+ - **Shared by [Optional]:** Yale LILY Lab
18
+
19
+ - **Model type:** Text2Text Generation
20
+ - **Language(s) (NLP):** More information needed
21
+ - **License:** More information needed
22
+ - **Parent Model:** BART
23
+ - **Resources for more information:**
24
+ - [Github Repo](https://github.com/Yale-LILY/BRIO)
25
+ - [Associated Paper](https://arxiv.org/abs/2203.16804)
26
+
27
+
28
+ # Uses
29
+
30
+
31
+ ## Direct Use
32
+ This model can be used for the task of Text2Text Generation
33
+
34
+ ## Downstream Use [Optional]
35
+
36
+ Further analysis also shows that our model can estimate probabilities of candidate summaries that are more correlated with their level of quality.
37
+
38
+
39
+
40
+
41
+ ## Out-of-Scope Use
42
+
43
+ The model should not be used to intentionally create hostile or alienating environments for people.
44
+
45
+ # Bias, Risks, and Limitations
46
+
47
+
48
+ 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.
49
+
50
+
51
+
52
+ ## Recommendations
53
+
54
+
55
+ Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
56
+
57
+ # Training Details
58
+
59
+ ## Training Data
60
+
61
+ The model creators note in the [associated paper](https://arxiv.org/abs/2203.16804):
62
+ > CNNDM4: is a large scale news dataset.
63
+ Nallapati et al: we treat the news articles as the source documents and the associated highlights as the summaries.
64
+ XSum5: is a highly abstractive dataset of articles from the British Broadcasting Corporation (BBC). NYT6: contains articles from the New York Times and the associated summaries
65
+
66
+
67
+
68
+ ## Training Procedure
69
+
70
+
71
+ ### Preprocessing
72
+
73
+ The model creators note in the [associated paper](https://arxiv.org/abs/2203.16804):
74
+ > We follow Kedzie et al. (2018) for data preprocessing and splitting, and use the associated archival abstracts as the summaries
75
+
76
+
77
+
78
+
79
+
80
+ ### Speeds, Sizes, Times
81
+
82
+ More information needed
83
+
84
+
85
+
86
+ # Evaluation
87
+
88
+
89
+ ## Testing Data, Factors & Metrics
90
+
91
+ ### Testing Data
92
+
93
+ More information needed
94
+
95
+ ### Factors
96
+ More information needed
97
+
98
+ ### Metrics
99
+
100
+ More information needed
101
+
102
+
103
+ ## Results
104
+
105
+
106
+ ### CNNDM
107
+ | | ROUGE-1 | ROUGE-2 | ROUGE-L |
108
+ |----------|---------|---------|---------|
109
+ | BART | 44.16 | 21.28 | 40.90 |
110
+ | Ours | 47.78 | 23.55 | 44.57 |
111
+
112
+
113
+
114
+
115
+ # Model Examination
116
+
117
+ The model creators note in the [associated paper](https://arxiv.org/abs/2203.16804):
118
+ > We attribute BRIO-Ctr’s superior performance to its use of the same model architecture (BART) for both candidate generation and scoring, while SimCLS uses RoBERTa as the evaluation model. As a result, BRIO-Ctr maximizes the parameter sharing between the two stages, and preserves the power of the Seq2Seq model pre-trained on the same dataset.
119
+
120
+
121
+
122
+ # Environmental Impact
123
+
124
+ 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).
125
+
126
+ - **Hardware Type:** More information needed
127
+ - **Hours used:** More information needed
128
+ - **Cloud Provider:** More information needed
129
+ - **Compute Region:** More information needed
130
+ - **Carbon Emitted:** More information needed
131
+
132
+ # Technical Specifications [optional]
133
+
134
+ ## Model Architecture and Objective
135
+
136
+ The model creators note in the [associated paper](https://arxiv.org/abs/2203.16804):
137
+
138
+ > Formulate summarization as a sequence-to-sequence (Seq2Seq) problem
139
+
140
+
141
+ ## Compute Infrastructure
142
+
143
+ More information needed
144
+
145
+ ### Hardware
146
+
147
+
148
+ More information needed
149
+
150
+ ### Software
151
+
152
+ More information needed.
153
+
154
+ # Citation
155
+
156
+
157
+ **BibTeX:**
158
+
159
+
160
+ ```bibtex
161
+ @misc{mesh-transformer-jax,
162
+ @misc{https://doi.org/10.48550/arxiv.2203.16804,
163
+ doi = {10.48550/ARXIV.2203.16804},
164
+
165
+ url = {https://arxiv.org/abs/2203.16804},
166
+
167
+ author = {Liu, Yixin and Liu, Pengfei and Radev, Dragomir and Neubig, Graham},
168
+
169
+ keywords = {Computation and Language (cs.CL), FOS: Computer and information sciences, FOS: Computer and information sciences},
170
+
171
+ title = {BRIO: Bringing Order to Abstractive Summarization},
172
+ ```
173
+
174
+
175
+
176
+
177
+ # Glossary [optional]
178
+
179
+ More information needed
180
+
181
+ # More Information [optional]
182
+ More information needed
183
+
184
+
185
+ # Model Card Authors [optional]
186
+
187
+ Yale LILY Lab in collaboration with Ezi Ozoani and the Hugging Face team
188
+
189
+ # Model Card Contact
190
+
191
+ More information needed
192
+
193
+ # How to Get Started with the Model
194
+
195
+ Use the code below to get started with the model.
196
+
197
+ <details>
198
+ <summary> Click to expand </summary>
199
+
200
+ ```python
201
+ from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
202
+
203
+ tokenizer = AutoTokenizer.from_pretrained("Yale-LILY/brio-cnndm-uncased")
204
+
205
+ model = AutoModelForSeq2SeqLM.from_pretrained("Yale-LILY/brio-cnndm-uncased")
206
+ ```
207
+ </details>
208
+
209
+