File size: 2,887 Bytes
402ff14
 
 
 
 
 
 
 
26c662c
 
 
 
 
402ff14
 
 
 
c3115b6
 
402ff14
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8b09b9c
 
 
 
 
 
 
402ff14
 
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
---
language: en
thumbnail: https://huggingface.co/front/thumbnails/microsoft.png
tags:
- text-classification
license: mit
---

# XtremeDistilTransformers for Distilling Massive Neural Networks

XtremeDistilTransformers is a distilled task-agnostic transformer model that leverages task transfer for learning a small universal model that can be applied to arbitrary tasks and languages as outlined in the paper [XtremeDistilTransformers: Task Transfer for Task-agnostic Distillation](https://arxiv.org/abs/2106.04563).

We leverage task transfer combined with multi-task distillation techniques from the papers [XtremeDistil: Multi-stage Distillation for Massive Multilingual Models](https://www.aclweb.org/anthology/2020.acl-main.202.pdf) and [MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers](https://arxiv.org/abs/2002.10957) with the following [Github code](https://github.com/microsoft/xtreme-distil-transformers).


This l6-h384 checkpoint with **6** layers, **384** hidden size, **12** attention heads corresponds to **22 million** parameters with **5.3x** speedup over BERT-base.

Other available checkpoints: [xtremedistil-l6-h384-uncased](https://huggingface.co/microsoft/xtremedistil-l6-h384-uncased) and [xtremedistil-l12-h384-uncased](https://huggingface.co/microsoft/xtremedistil-l12-h384-uncased) 

The following table shows the results on GLUE dev set and SQuAD-v2.

| Models         | #Params | Speedup | MNLI | QNLI | QQP  | RTE  | SST  | MRPC | SQUAD2 | Avg   |
|----------------|--------|---------|------|------|------|------|------|------|--------|-------|
| BERT        | 109    | 1x       | 84.5 | 91.7 | 91.3 | 68.6 | 93.2 | 87.3 | 76.8   | 84.8 |
| DistilBERT  | 66     | 2x       | 82.2 | 89.2 | 88.5 | 59.9 | 91.3 | 87.5 | 70.7   | 81.3 |
| TinyBERT    | 66     | 2x       | 83.5 | 90.5 | 90.6 | 72.2 | 91.6 | 88.4 | 73.1   | 84.3 |
| MiniLM      | 66     | 2x       | 84.0   | 91.0   | 91.0   | 71.5 | 92.0   | 88.4 | 76.4   | 84.9  |
| MiniLM      | 22     | 5.3x     | 82.8 | 90.3 | 90.6 | 68.9 | 91.3 | 86.6 | 72.9   | 83.3 |
| XtremeDistil-l6-h256   | 13     | 8.7x     | 83.9 | 89.5 | 90.6   | 80.1 | 91.2 | 90.0   | 74.1   | 85.6 |
| XtremeDistil-l6-h384   | 22     | 5.3x     | 85.4 | 90.3 | 91.0   | 80.9 | 92.3 | 90.0   | 76.6   | 86.6 |
| XtremeDistil-l12-h384   | 33     | 2.7x     | 87.2 | 91.9 | 91.3   | 85.6 | 93.1 | 90.4   | 80.2   | 88.5 |

Tested with `tensorflow 2.3.1, transformers 4.1.1, torch 1.6.0`

If you use this checkpoint in your work, please cite:

``` latex
@misc{mukherjee2021xtremedistiltransformers,
      title={XtremeDistilTransformers: Task Transfer for Task-agnostic Distillation}, 
      author={Subhabrata Mukherjee and Ahmed Hassan Awadallah and Jianfeng Gao},
      year={2021},
      eprint={2106.04563},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```