File size: 3,184 Bytes
39d7aeb 0796289 6e9cc79 0796289 7290dd3 e3136b4 b8d248a bc0ad39 b8d248a f2a5b8c 7f00080 bdf9146 6336d7b bdf9146 673a0b0 c498e18 eb81f8d 673a0b0 c498e18 d8e2057 673a0b0 c249ba4 2525ce5 e94525d 94bc06c 2525ce5 27bd5ab e94525d 163cbdc 27bd5ab 2525ce5 163cbdc e94525d |
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 |
---
widget:
- text: "create array containing the maximum value of respective elements of array `[2, 3, 4]` and array `[1, 5, 2]"
- text: "check if all elements in list `mylist` are identical"
- text: "enable debug mode on flask application `app`"
- text: "getting the length of `my_tuple`"
- text: 'find all files in directory "/mydir" with extension ".txt"'
---
```
```
[](https://paperswithcode.com/sota/code-generation-on-conala?p=mariancg-a-code-generation-transformer-model)
```
```
# MarianCG: a code generation transformer model inspired by machine translation
This model is to improve the solving of the code generation problem and implement a transformer model that can work with high accurate results. We implemented MarianCG transformer model which is a code generation model that can be able to generate code from natural language. This work declares the impact of using Marian machine translation model for solving the problem of code generation. In our implementation, we prove that a machine translation model can be operated and working as a code generation model. Finally, we set the new contributors and state-of-the-art on CoNaLa reaching a BLEU score of 30.92 and Exact Match Accuracy of 6.2 in the code generation problem with CoNaLa dataset.
MarianCG model and its implemetation with the code of training and the generated output is available at this repository:
https://github.com/AhmedSSoliman/MarianCG-NL-to-Code
CoNaLa Dataset for Code Generation is available at
https://huggingface.co/datasets/AhmedSSoliman/CoNaLa
This is the model is avialable on the huggingface hub https://huggingface.co/AhmedSSoliman/MarianCG-CoNaLa
```python
# Model and Tokenizer
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
# model_name = "AhmedSSoliman/MarianCG-NL-to-Code"
model = AutoModelForSeq2SeqLM.from_pretrained("AhmedSSoliman/MarianCG-CoNaLa")
tokenizer = AutoTokenizer.from_pretrained("AhmedSSoliman/MarianCG-CoNaLa")
# Input (Natural Language) and Output (Python Code)
NL_input = "create array containing the maximum value of respective elements of array `[2, 3, 4]` and array `[1, 5, 2]"
output = model.generate(**tokenizer(NL_input, padding="max_length", truncation=True, max_length=512, return_tensors="pt"))
output_code = tokenizer.decode(output[0], skip_special_tokens=True)
```
This model is available in spaces using gradio at: https://huggingface.co/spaces/AhmedSSoliman/MarianCG-CoNaLa
---
Tasks:
- Translation
- Code Generation
- Text2Text Generation
- Text Generation
---
# Citation
We now have a [paper](https://doi.org/10.1186/s44147-022-00159-4) for this work and you can cite:
```
@article{soliman2022mariancg,
title={MarianCG: a code generation transformer model inspired by machine translation},
author={Soliman, Ahmed S and Hadhoud, Mayada M and Shaheen, Samir I},
journal={Journal of Engineering and Applied Science},
volume={69},
number={1},
pages={1--23},
year={2022},
publisher={SpringerOpen}
url={https://doi.org/10.1186/s44147-022-00159-4}
}
```
|