File size: 12,440 Bytes
80ba612
 
bbfc484
 
 
 
 
 
 
80ba612
bbfc484
 
 
 
 
 
e9e32bd
bbfc484
 
 
 
 
170e974
bbfc484
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
---
license: apache-2.0
tags:
- generated_from_trainer
datasets:
- NewSHead
model-index:
- name: Centrum
  results: []
---

<!-- 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. -->

# Centrum

Centrum is a pretrained model for multi-document summarization, trained with centroid-based pretraining objective on the NewSHead dataset. It is initialized from [allenai/led-base-16384](https://huggingface.co/allenai/led-base-16384). The details of the approach are mentioned in the preprint [Multi-Document Summarization with Centroid-Based Pretraining](https://arxiv.org/abs/2208.01006) (Ratish Puduppully and Mark Steedman).
It achieves the following results on the evaluation set:
- Loss: 3.5568

## Model description

The script for training and inference of Centrum is available on https://github.com/ratishsp/centrum

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 3e-05
- train_batch_size: 1
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 4
- total_train_batch_size: 16
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 10000
- training_steps: 100000
- mixed_precision_training: Native AMP
- label_smoothing_factor: 0.1

### Training results

| Training Loss | Epoch | Step   | Validation Loss |
|:-------------:|:-----:|:------:|:---------------:|
| 4.1628        | 0.05  | 500    | 4.0732          |
| 4.0278        | 0.09  | 1000   | 3.9800          |
| 4.0008        | 0.14  | 1500   | 3.9283          |
| 3.9564        | 0.19  | 2000   | 3.8941          |
| 3.9193        | 0.23  | 2500   | 3.8780          |
| 3.9185        | 0.28  | 3000   | 3.8501          |
| 3.8881        | 0.32  | 3500   | 3.8334          |
| 3.8869        | 0.37  | 4000   | 3.8211          |
| 3.876         | 0.42  | 4500   | 3.8057          |
| 3.8552        | 0.46  | 5000   | 3.7954          |
| 3.8198        | 0.51  | 5500   | 3.7861          |
| 3.8016        | 0.56  | 6000   | 3.7750          |
| 3.8033        | 0.6   | 6500   | 3.7651          |
| 3.7927        | 0.65  | 7000   | 3.7528          |
| 3.7978        | 0.7   | 7500   | 3.7429          |
| 3.7727        | 0.74  | 8000   | 3.7367          |
| 3.7634        | 0.79  | 8500   | 3.7275          |
| 3.7395        | 0.83  | 9000   | 3.7158          |
| 3.7432        | 0.88  | 9500   | 3.7066          |
| 3.7623        | 0.93  | 10000  | 3.7039          |
| 3.7182        | 0.97  | 10500  | 3.6904          |
| 3.7146        | 1.02  | 11000  | 3.6881          |
| 3.681         | 1.07  | 11500  | 3.6797          |
| 3.6745        | 1.11  | 12000  | 3.6750          |
| 3.6794        | 1.16  | 12500  | 3.6748          |
| 3.6802        | 1.21  | 13000  | 3.6696          |
| 3.665         | 1.25  | 13500  | 3.6609          |
| 3.6516        | 1.3   | 14000  | 3.6633          |
| 3.6577        | 1.34  | 14500  | 3.6573          |
| 3.6409        | 1.39  | 15000  | 3.6519          |
| 3.6691        | 1.44  | 15500  | 3.6490          |
| 3.6521        | 1.48  | 16000  | 3.6475          |
| 3.6435        | 1.53  | 16500  | 3.6465          |
| 3.6466        | 1.58  | 17000  | 3.6392          |
| 3.644         | 1.62  | 17500  | 3.6419          |
| 3.6347        | 1.67  | 18000  | 3.6347          |
| 3.6205        | 1.71  | 18500  | 3.6328          |
| 3.6451        | 1.76  | 19000  | 3.6310          |
| 3.6327        | 1.81  | 19500  | 3.6284          |
| 3.6166        | 1.85  | 20000  | 3.6267          |
| 3.622         | 1.9   | 20500  | 3.6212          |
| 3.6164        | 1.95  | 21000  | 3.6199          |
| 3.6178        | 1.99  | 21500  | 3.6201          |
| 3.5892        | 2.04  | 22000  | 3.6201          |
| 3.5855        | 2.09  | 22500  | 3.6221          |
| 3.5658        | 2.13  | 23000  | 3.6193          |
| 3.5916        | 2.18  | 23500  | 3.6144          |
| 3.5767        | 2.22  | 24000  | 3.6101          |
| 3.5809        | 2.27  | 24500  | 3.6115          |
| 3.5561        | 2.32  | 25000  | 3.6110          |
| 3.5831        | 2.36  | 25500  | 3.6080          |
| 3.5551        | 2.41  | 26000  | 3.6121          |
| 3.5588        | 2.46  | 26500  | 3.6072          |
| 3.5645        | 2.5   | 27000  | 3.6056          |
| 3.5804        | 2.55  | 27500  | 3.6038          |
| 3.5712        | 2.6   | 28000  | 3.6052          |
| 3.5494        | 2.64  | 28500  | 3.6014          |
| 3.582         | 2.69  | 29000  | 3.5995          |
| 3.5487        | 2.73  | 29500  | 3.6051          |
| 3.5709        | 2.78  | 30000  | 3.5954          |
| 3.5546        | 2.83  | 30500  | 3.5941          |
| 3.5525        | 2.87  | 31000  | 3.5952          |
| 3.5603        | 2.92  | 31500  | 3.5972          |
| 3.5572        | 2.97  | 32000  | 3.5947          |
| 3.5106        | 3.01  | 32500  | 3.5952          |
| 3.5142        | 3.06  | 33000  | 3.5937          |
| 3.506         | 3.11  | 33500  | 3.5965          |
| 3.515         | 3.15  | 34000  | 3.5932          |
| 3.5247        | 3.2   | 34500  | 3.5951          |
| 3.5384        | 3.24  | 35000  | 3.5917          |
| 3.5165        | 3.29  | 35500  | 3.5887          |
| 3.5187        | 3.34  | 36000  | 3.5866          |
| 3.5097        | 3.38  | 36500  | 3.5895          |
| 3.5136        | 3.43  | 37000  | 3.5878          |
| 3.5095        | 3.48  | 37500  | 3.5839          |
| 3.5226        | 3.52  | 38000  | 3.5859          |
| 3.5277        | 3.57  | 38500  | 3.5827          |
| 3.4959        | 3.62  | 39000  | 3.5846          |
| 3.5003        | 3.66  | 39500  | 3.5823          |
| 3.5095        | 3.71  | 40000  | 3.5820          |
| 3.4814        | 3.75  | 40500  | 3.5854          |
| 3.5173        | 3.8   | 41000  | 3.5796          |
| 3.4968        | 3.85  | 41500  | 3.5810          |
| 3.5183        | 3.89  | 42000  | 3.5783          |
| 3.512         | 3.94  | 42500  | 3.5784          |
| 3.5069        | 3.99  | 43000  | 3.5775          |
| 3.5014        | 4.03  | 43500  | 3.5819          |
| 3.4787        | 4.08  | 44000  | 3.5836          |
| 3.4625        | 4.12  | 44500  | 3.5788          |
| 3.4902        | 4.17  | 45000  | 3.5784          |
| 3.4927        | 4.22  | 45500  | 3.5773          |
| 3.4813        | 4.26  | 46000  | 3.5769          |
| 3.4637        | 4.31  | 46500  | 3.5761          |
| 3.4731        | 4.36  | 47000  | 3.5771          |
| 3.4856        | 4.4   | 47500  | 3.5786          |
| 3.4579        | 4.45  | 48000  | 3.5790          |
| 3.5032        | 4.5   | 48500  | 3.5738          |
| 3.4826        | 4.54  | 49000  | 3.5749          |
| 3.4709        | 4.59  | 49500  | 3.5746          |
| 3.4916        | 4.63  | 50000  | 3.5745          |
| 3.4715        | 4.68  | 50500  | 3.5706          |
| 3.4926        | 4.73  | 51000  | 3.5729          |
| 3.4974        | 4.77  | 51500  | 3.5725          |
| 3.4796        | 4.82  | 52000  | 3.5683          |
| 3.4817        | 4.87  | 52500  | 3.5707          |
| 3.4683        | 4.91  | 53000  | 3.5721          |
| 3.4986        | 4.96  | 53500  | 3.5689          |
| 3.4763        | 5.01  | 54000  | 3.5716          |
| 3.4668        | 5.05  | 54500  | 3.5700          |
| 3.4274        | 5.1   | 55000  | 3.5724          |
| 3.4499        | 5.14  | 55500  | 3.5717          |
| 3.4507        | 5.19  | 56000  | 3.5706          |
| 3.4343        | 5.24  | 56500  | 3.5697          |
| 3.4151        | 5.28  | 57000  | 3.5710          |
| 3.4469        | 5.33  | 57500  | 3.5712          |
| 3.458         | 5.38  | 58000  | 3.5692          |
| 3.4559        | 5.42  | 58500  | 3.5680          |
| 3.4354        | 5.47  | 59000  | 3.5683          |
| 3.4479        | 5.52  | 59500  | 3.5703          |
| 3.4627        | 5.56  | 60000  | 3.5678          |
| 3.4478        | 5.61  | 60500  | 3.5659          |
| 3.4645        | 5.65  | 61000  | 3.5675          |
| 3.4658        | 5.7   | 61500  | 3.5666          |
| 3.4657        | 5.75  | 62000  | 3.5658          |
| 3.4618        | 5.79  | 62500  | 3.5653          |
| 3.4541        | 5.84  | 63000  | 3.5653          |
| 3.4552        | 5.89  | 63500  | 3.5648          |
| 3.4679        | 5.93  | 64000  | 3.5648          |
| 3.4423        | 5.98  | 64500  | 3.5652          |
| 3.3893        | 6.03  | 65000  | 3.5646          |
| 3.4239        | 6.07  | 65500  | 3.5668          |
| 3.4329        | 6.12  | 66000  | 3.5639          |
| 3.4151        | 6.16  | 66500  | 3.5649          |
| 3.4181        | 6.21  | 67000  | 3.5682          |
| 3.4314        | 6.26  | 67500  | 3.5669          |
| 3.4245        | 6.3   | 68000  | 3.5629          |
| 3.421         | 6.35  | 68500  | 3.5663          |
| 3.4329        | 6.4   | 69000  | 3.5660          |
| 3.4122        | 6.44  | 69500  | 3.5651          |
| 3.4362        | 6.49  | 70000  | 3.5628          |
| 3.4497        | 6.54  | 70500  | 3.5648          |
| 3.431         | 6.58  | 71000  | 3.5626          |
| 3.432         | 6.63  | 71500  | 3.5648          |
| 3.4208        | 6.67  | 72000  | 3.5635          |
| 3.4526        | 6.72  | 72500  | 3.5645          |
| 3.4139        | 6.77  | 73000  | 3.5621          |
| 3.4212        | 6.81  | 73500  | 3.5629          |
| 3.4352        | 6.86  | 74000  | 3.5597          |
| 3.4242        | 6.91  | 74500  | 3.5597          |
| 3.429         | 6.95  | 75000  | 3.5619          |
| 3.4133        | 7.0   | 75500  | 3.5592          |
| 3.4086        | 7.04  | 76000  | 3.5621          |
| 3.4056        | 7.09  | 76500  | 3.5604          |
| 3.4158        | 7.14  | 77000  | 3.5629          |
| 3.4153        | 7.18  | 77500  | 3.5609          |
| 3.4155        | 7.23  | 78000  | 3.5621          |
| 3.4117        | 7.28  | 78500  | 3.5626          |
| 3.407         | 7.32  | 79000  | 3.5638          |
| 3.3977        | 7.37  | 79500  | 3.5604          |
| 3.4134        | 7.42  | 80000  | 3.5611          |
| 3.4403        | 7.46  | 80500  | 3.5630          |
| 3.4002        | 7.51  | 81000  | 3.5601          |
| 3.4147        | 7.55  | 81500  | 3.5577          |
| 3.4068        | 7.6   | 82000  | 3.5588          |
| 3.4165        | 7.65  | 82500  | 3.5613          |
| 3.409         | 7.69  | 83000  | 3.5596          |
| 3.4213        | 7.74  | 83500  | 3.5583          |
| 3.403         | 7.79  | 84000  | 3.5601          |
| 3.3819        | 7.83  | 84500  | 3.5580          |
| 3.4182        | 7.88  | 85000  | 3.5570          |
| 3.4099        | 7.93  | 85500  | 3.5570          |
| 3.3845        | 7.97  | 86000  | 3.5582          |
| 3.411         | 8.02  | 86500  | 3.5610          |
| 3.3952        | 8.06  | 87000  | 3.5588          |
| 3.4211        | 8.11  | 87500  | 3.5588          |
| 3.4171        | 8.16  | 88000  | 3.5570          |
| 3.3825        | 8.2   | 88500  | 3.5607          |
| 3.3807        | 8.25  | 89000  | 3.5579          |
| 3.3842        | 8.3   | 89500  | 3.5583          |
| 3.3809        | 8.34  | 90000  | 3.5596          |
| 3.4033        | 8.39  | 90500  | 3.5590          |
| 3.4156        | 8.44  | 91000  | 3.5577          |
| 3.3927        | 8.48  | 91500  | 3.5585          |
| 3.4041        | 8.53  | 92000  | 3.5596          |
| 3.4006        | 8.57  | 92500  | 3.5600          |
| 3.4007        | 8.62  | 93000  | 3.5578          |
| 3.4047        | 8.67  | 93500  | 3.5572          |
| 3.3904        | 8.71  | 94000  | 3.5571          |
| 3.3888        | 8.76  | 94500  | 3.5581          |
| 3.3876        | 8.81  | 95000  | 3.5572          |
| 3.3872        | 8.85  | 95500  | 3.5575          |
| 3.3753        | 8.9   | 96000  | 3.5577          |
| 3.3961        | 8.95  | 96500  | 3.5568          |
| 3.4131        | 8.99  | 97000  | 3.5579          |
| 3.3647        | 9.04  | 97500  | 3.5573          |
| 3.3792        | 9.08  | 98000  | 3.5576          |
| 3.3755        | 9.13  | 98500  | 3.5575          |
| 3.3981        | 9.18  | 99000  | 3.5573          |
| 3.3914        | 9.22  | 99500  | 3.5573          |
| 3.4136        | 9.27  | 100000 | 3.5575          |


### Framework versions

- Transformers 4.20.0.dev0
- Pytorch 1.11.0
- Datasets 2.2.2
- Tokenizers 0.12.1