burakkececi
commited on
Commit
•
4f9057f
1
Parent(s):
293cfb1
Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
license: mit
|
3 |
+
language:
|
4 |
+
- tr
|
5 |
+
library_name: transformers
|
6 |
+
---
|
7 |
+
# 🇹🇷 Turkish BERT Model for Software Engineering
|
8 |
+
|
9 |
+
This repository was created within the scope of computer engineering undergraduate graduation project.
|
10 |
+
|
11 |
+
This research aims to perform an exploratory case study to determine the functional dimensions of user requirements or use cases for software projects.
|
12 |
+
In order to perform this task we created two models, [SE-BERT](https://huggingface.co/burakkececi/bert-software-engineering) and SE-BERTurk.
|
13 |
+
|
14 |
+
You can find a detailed description of the project at the [link](https://github.com/burakkececi/software-size-estimation-nlp).
|
15 |
+
|
16 |
+
# SE-BERTurk
|
17 |
+
|
18 |
+
SE-BERT is a BERT model trained for domain adaptation in a software engineering context.
|
19 |
+
|
20 |
+
We applied Masked Language Modeling (MLM), an unsupervised learning technique, for domain adaptation. MLM enhances the model understanding of domain-specific language by masking portions of the input text and training the model to predict the masked words based on the surrounding context.
|
21 |
+
|
22 |
+
## Stats
|
23 |
+
Created a bilingual [SE corpus](https://drive.google.com/file/d/1IgnJTaR2-pe889TdQZtYF8SKOH92mi1l/view?usp=drive_link) (166Mb) ➡️ [Descriptive stats of the corpus](https://docs.google.com/spreadsheets/d/1Xnn_xfu4tdCtWg-nQ8ce_LHe9F-g0BSmUxzTdi5g1r4/edit?usp=sharing)
|
24 |
+
* 166K entry = 886K sentence = 10M words
|
25 |
+
* 156K training entry + 10K test entry
|
26 |
+
* Each entry has a maximum length of 512 tokens
|
27 |
+
|
28 |
+
The final training corpus has a size of 166MB and 10.554.750 words.
|
29 |
+
|
30 |
+
## MLM Training (Domain Adaptation)
|
31 |
+
Used ``AdamW`` optimizer and set ``num_epochs = 1``, ``lr = 2e-5``, ``eps = 1e-8``
|
32 |
+
* For T4 GPU ➡️ Set ``batch_size = 6`` (13.5Gb memory)
|
33 |
+
* For A100 GPU ➡️ Set ``batch_size = 50`` (37Gb memory) and ``fp16 = True``
|
34 |
+
|
35 |
+
**Perplexity**
|
36 |
+
* ``3,665`` PPL for SE-BERTurk
|
37 |
+
|
38 |
+
### Evaluation Steps:
|
39 |
+
1) Calculate ``PPL`` (perplexity) on the test corpus (10K context with a maximum length of 512 tokens)
|
40 |
+
2) Calculate ``PPL`` (perplexity) on the requirement datasets
|
41 |
+
3) Evaluate performance on downstream tasks:
|
42 |
+
* For size measurement ➡️ ``MAE``, ``MSE``, ``MMRE``, ``PRED(30)``, ``ACC``
|
43 |
+
|
44 |
+
## Usage
|
45 |
+
|
46 |
+
With Transformers >= 2.11 our SE-BERT uncased model can be loaded like:
|
47 |
+
|
48 |
+
```python
|
49 |
+
from transformers import AutoModel, AutoTokenizer
|
50 |
+
tokenizer = AutoTokenizer.from_pretrained("burakkececi/bert-turkish-software-engineering/model")
|
51 |
+
model = AutoModel.from_pretrained("burakkececi/bert-turkish-software-engineering/tokenizer")
|
52 |
+
```
|
53 |
+
|
54 |
+
# Huggingface model hub
|
55 |
+
|
56 |
+
All models are available on the [Huggingface model hub](https://huggingface.co/burakkececi).
|