File size: 1,826 Bytes
1c325c7
 
 
 
 
 
 
 
 
0527fb8
 
 
1c325c7
 
 
 
 
 
 
 
 
0527fb8
 
 
 
 
 
 
1c325c7
 
0527fb8
 
 
 
 
 
 
 
1c325c7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
0527fb8
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
---
license: mit
tags:
- generated_from_trainer
datasets:
- snli
model-index:
- name: DeBERTa-finetuned-SNLI2
  results: []
metrics:
- accuracy
library_name: 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. -->

# DeBERTa-finetuned-SNLI2

This model is a fine-tuned version of [gyeoldere/test_trainer](https://huggingface.co/gyeoldere/test_trainer) on the snli dataset.

Test_trainer model is a fine-tuned version of [microsoft/deberta-base](https://huggingface.co/microsoft/deberta-base) on the snli dataset.

This model achieves the following results on the evaluation set:
- NLI accuracy : 0.86
- MLM accuracy : 0.68


## Model description

This model fine-tuned to perform 2 tasks simultaneously; NLI task and MLM task.

Output vector of DeBERTa processed through two different fc layer to predict.
I used layer structure introduced in BERT paper, which is implemented on huggingface transformers; DebertaForTokenClassification and DebertaForMaskedLM.
[https://huggingface.co/docs/transformers/index]

BinaryCrossEntrophyLoss are used for each class, and two losses are added to obtain final loss
final_loss = MLM_loss + NLI_loss

## 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: 2e-05
- train_batch_size: 128
- eval_batch_size: 128
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- num_epochs: 3

### Training results



### Framework versions

- Transformers 4.26.0
- Pytorch 1.13.1+cu116
- Datasets 2.9.0
- Tokenizers 0.13.2