w11wo commited on
Commit
51b3647
1 Parent(s): cef1891

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +79 -0
README.md ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: id
3
+ tags:
4
+ - gpt2-indo-small-kids-stories
5
+ license: mit
6
+ widget:
7
+ - text: "Archie sedang mengendarai roket ke planet Mars."
8
+ ---
9
+
10
+ ## GPT-2 Indonesian Small Kids Stories
11
+
12
+ GPT-2 Indonesian Small Kids Stories is a causal language model based on the [OpenAI GPT-2](https://cdn.openai.com/better-language-models/language_models_are_unsupervised_multitask_learners.pdf) model. The model was originally the pre-trained [GPT2 Small Indonesian](https://huggingface.co/flax-community/gpt2-small-indonesian) model, which was then fine-tuned on Indonesian kids' stories from [Room To Read](https://literacycloud.org/) and [Let's Read](https://reader.letsreadasia.org/).
13
+
14
+ 10% of the dataset was kept for evaluation purposes. The pre-trained model was fine-tuned and achieved an evaluation loss of 3.777 and an evaluation perplexity of 43.68.
15
+
16
+ Hugging Face's `Trainer` class from the [Transformers](https://huggingface.co/transformers) library was used to train the model. PyTorch was used as the backend framework during training, but the model remains compatible with other frameworks nonetheless.
17
+
18
+ ## Model
19
+
20
+ | Model | #params | Arch. | Training/Validation data (text) |
21
+ | ------------------------------ | ------- | ---------- | --------------------------------- |
22
+ | `gpt2-indo-small-kids-stories` | 124M | GPT2 Small | Indonesian Kids' Stories (860 KB) |
23
+
24
+ ## Evaluation Results
25
+
26
+ The model was fine-tuned for 10 epochs.
27
+
28
+ | Epoch | Training Loss | Validation Loss |
29
+ | ----- | ------------- | --------------- |
30
+ | 1 | 4.259600 | 4.020201 |
31
+ | 2 | 3.979100 | 3.911295 |
32
+ | 3 | 3.818300 | 3.849313 |
33
+ | 4 | 3.691600 | 3.809931 |
34
+ | 5 | 3.589300 | 3.789201 |
35
+ | 6 | 3.506200 | 3.778665 |
36
+ | 7 | 3.439200 | 3.774871 |
37
+ | 8 | 3.387600 | 3.774859 |
38
+ | 9 | 3.351300 | 3.776672 |
39
+ | 10 | 3.330100 | 3.776935 |
40
+
41
+ ## How to Use (PyTorch)
42
+
43
+ ### As Causal Language Model
44
+
45
+ ```python
46
+ from transformers import pipeline
47
+
48
+ pretrained_name = "bookbot/gpt2-indo-small-kids-stories"
49
+
50
+ nlp = pipeline(
51
+ "text-generation",
52
+ model=pretrained_name,
53
+ tokenizer=pretrained_name
54
+ )
55
+
56
+ nlp("Archie sedang mengendarai roket ke planet Mars.")
57
+ ```
58
+
59
+ ### Feature Extraction in PyTorch
60
+
61
+ ```python
62
+ from transformers import GPT2LMHeadModel, GPT2TokenizerFast
63
+
64
+ pretrained_name = "bookbot/gpt2-indo-small-kids-stories"
65
+ model = GPT2LMHeadModel.from_pretrained(pretrained_name)
66
+ tokenizer = GPT2TokenizerFast.from_pretrained(pretrained_name)
67
+
68
+ prompt = "Archie sedang mengendarai roket ke planet Mars."
69
+ encoded_input = tokenizer(prompt, return_tensors='pt')
70
+ output = model(**encoded_input)
71
+ ```
72
+
73
+ ## Disclaimer
74
+
75
+ Do consider the biases which come from both the pre-trained GPT-2 model and the Indonesian Kids' Stories dataset that may be carried over into the results of this model.
76
+
77
+ ## Author
78
+
79
+ GPT-2 Indonesian Small Kids Stories was trained and evaluated by [Wilson Wongso](https://w11wo.github.io/). All computation and development are done on Google Colaboratory using their free GPU access.