Text2Text Generation
Transformers
PyTorch
t5
codet5
text-generation-inference
File size: 6,249 Bytes
6f0f019
 
 
1a4c722
6f0f019
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
85ae120
 
 
 
 
 
 
 
5fe3699
85ae120
6f0f019
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
---
license: apache-2.0
tags:
- codet5
datasets:
- code_search_net
---

# CodeT5 (base-sized model) 

Pre-trained CodeT5 model. It was introduced in the paper [CodeT5: Identifier-aware Unified Pre-trained Encoder-Decoder Models
for Code Understanding and Generation](https://arxiv.org/abs/2109.00859) by Yue Wang, Weishi Wang, Shafiq Joty, Steven C.H. Hoi and first released in [this repository](https://github.com/salesforce/CodeT5). 

Disclaimer: The team releasing CodeT5 did not write a model card for this model so this model card has been written by the Hugging Face team (more specifically, [nielsr](https://huggingface.co/nielsr)).

## Model description

From the abstract:

"We present CodeT5, a unified pre-trained encoder-decoder Transformer model that better leverages the code semantics conveyed from the developer-assigned identifiers. Our model employs a unified framework to seamlessly support both code understanding and generation tasks and allows for multi-task learning. Besides, we propose a novel identifier-aware pre-training task that enables the model to distinguish which code tokens are identifiers and to recover them when they are masked. Furthermore, we propose to exploit the user-written code comments with a bimodal dual generation task for better NL-PL alignment. Comprehensive experiments show that CodeT5 significantly outperforms prior methods on understanding tasks such as code defect detection and clone detection, and generation tasks across various directions including PL-NL, NL-PL, and PL-PL. Further analysis reveals that our model can better capture semantic information from code."

## Intended uses & limitations

This repository contains the pre-trained model only, so you can use this model for masked span prediction, as shown in the code example below. However, the main use of this model is to fine-tune it for a downstream task of interest, such as:
* code summarization
* code generation
* code translation
* code refinement
* code defect detection
* code clone detection. 

Supervised datasets for code can be found [here](https://huggingface.co/datasets?languages=languages:code).
See the [model hub](https://huggingface.co/models?search=salesforce/codet) to look for fine-tuned versions on a task that interests you.

### How to use

Here is how to use this model:

```python
from transformers import RobertaTokenizer, T5ForConditionalGeneration

tokenizer = RobertaTokenizer.from_pretrained('Salesforce/codet5-base')
model = T5ForConditionalGeneration.from_pretrained('Salesforce/codet5-base')

text = "def greet(user): print(f'hello <extra_id_0>!') </s>"
inputs = tokenizer(text, return_tensors="pt").input_ids

# simply generate a single sequence
generated_ids = model.generate(input_ids, max_length=8)
print(tokenizer.decode(generated_ids[0], skip_special_tokens=True))
# this prints {user.name}

# or, generating 20 sequences with maximum length set to 10
outputs = model.generate(input_ids=input_ids, 
                          num_beams=200, num_return_sequences=20,
                          max_length=10)

_0_index = text.index('<extra_id_0>')
_result_prefix = text[:_0_index]
_result_suffix = text[_0_index+12:]  # 12 is the length of <extra_id_0>

def _filter(output, end_token='<extra_id_1>'):
    # The first token is <pad> (indexed at 0), the second token is <s> (indexed at 1)
    # and the third token is <extra_id_0> (indexed at 32099)
    # So we only decode from the fourth generated id
    _txt = tokenizer.decode(output[3:], skip_special_tokens=False, clean_up_tokenization_spaces=False)
    if end_token in _txt:
        _end_token_index = _txt.index(end_token)
        return _result_prefix + _txt[:_end_token_index] + _result_suffix
    else:
        return _result_prefix + _txt + _result_suffix

results = list(map(_filter, outputs))
print(results)
# this prints:
#["def greet(user): print(f'hello {user.name} {user!') </s>",
# "def greet(user): print(f'hello {user.username} {user!') </s>",
# "def greet(user): print(f'hello {user.name}: {user!') </s>",
# "def greet(user): print(f'hello {user}') print(f!') </s>",
# "def greet(user): print(f'hello {user.name} �!') </s>",
# "def greet(user): print(f'hello {user}') print ( f!') </s>",
# "def greet(user): print(f'hello {user.username}: {user!') </s>",
# "def greet(user): print(f'hello {user}' ) print(f!') </s>",
# "def greet(user): print(f'hello {user.username} �!') </s>",
# "def greet(user): print(f'hello {user.name}, {user!') </s>",
# "def greet(user): print(f'hello {user.login} {user!') </s>",
# "def greet(user): print(f'hello {user} →!') </s>",
# "def greet(user): print(f'hello {user}!') print(!') </s>",
# "def greet(user): print(f'hello {user.name} ({user!') </s>",
# "def greet(user): print(f'hello {user.email} {user!') </s>",
# "def greet(user): print(f'hello {user}!') print (!') </s>",
# "def greet(user): print(f'hello {user.username}, {user!') </s>",
# "def greet(user): print(f'hello {user}' ) print ( f!') </s>",
# "def greet(user): print(f'hello {user.nickname} {!') </s>",
# "def greet(user): print(f'hello {user} {user.name!') </s>"]
```

## Training data

The CodeT5 model was pretrained on CodeSearchNet [Husain et al., 2019](https://arxiv.org/abs/1909.09436). Additionally, the authors collected two datasets of C/CSharp from [BigQuery1](https://console.cloud.google.com/marketplace/details/github/github-repos) to ensure that all downstream tasks have overlapped programming languages with the pre-training data. In total, around 8.35 million instances are used for pretraining. 

## Training procedure

### Preprocessing

This model uses a code-specific BPE (Byte-Pair Encoding) tokenizer. One can prepare text (or code) for the model using RobertaTokenizer, with the files from this repository.

## Evaluation results

For evaluation results on several downstream benchmarks, we refer to the paper.

### BibTeX entry and citation info

```bibtex
@misc{wang2021codet5,
      title={CodeT5: Identifier-aware Unified Pre-trained Encoder-Decoder Models for Code Understanding and Generation}, 
      author={Yue Wang and Weishi Wang and Shafiq Joty and Steven C. H. Hoi},
      year={2021},
      eprint={2109.00859},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
```