File size: 985 Bytes
eb823ba
f066ef4
 
 
eb823ba
f066ef4
 
 
eb823ba
f066ef4
 
 
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
eb823ba
f066ef4
a424f3c
f066ef4
 
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
# Tokenizer

We trained our tokenizer using [sentencepiece](https://github.com/google/sentencepiece)'s unigram tokenizer. Then loaded the tokenizer as MT5TokenizerFast.

## Model

We used [MT5-base](https://huggingface.co/google/mt5-base) model.

## Datasets

We used [Code Search Net](https://huggingface.co/datasets/code_search_net)'s dataset and some scrapped data from internet to train the model. We maintained a list of datasets where each dataset had codes of same language.

## Plots

### Train loss

![train loss](https://i.ibb.co/x53Wm8n/train-loss.png)

### Evaluation loss

![eval loss](https://i.ibb.co/McB2jnf/eval-loss.png)

### Evaluation accuracy

![eval accuracy](https://i.ibb.co/YDGhLdn/eval-accuracy.png)

### Learning rate

![learning rate](https://i.ibb.co/CMStzWv/learning-rate.png)

## Fine tuning (WIP)

We fine tuned the model with [CodeXGLUE code-to-code-trans dataset](https://huggingface.co/datasets/code_x_glue_cc_code_to_code_trans), and scrapper data.