Commit
•
fa013b0
1
Parent(s):
26a826f
Update README.md
Browse files
README.md
CHANGED
@@ -1,18 +1,19 @@
|
|
1 |
## GPT-Neo Small Portuguese
|
2 |
|
3 |
-
|
4 |
This is a finetuned version from GPT-Neo 125M by EletheurAI to Portuguese language.
|
5 |
|
6 |
-
|
7 |
It was training from 227,382 selected texts from a PTWiki Dump. You can found all the data from here: https://archive.org/details/ptwiki-dump-20210520
|
8 |
|
9 |
-
|
10 |
Every text was passed through a GPT2-Tokenizer with bos and eos tokens to separate it, with max sequence length that the GPT-Neo could support. It was finetuned using the default metrics of the Trainer Class, available on the Hugging Face library.
|
11 |
|
12 |
##### Learning Rate: **2e-4**
|
13 |
##### Epochs: **1**
|
14 |
|
15 |
-
|
|
|
16 |
My true intention was totally educational, thus making available a Portuguese version of this model.
|
17 |
|
18 |
How to use
|
@@ -45,8 +46,8 @@ sample_outputs = model.generate(generated,
|
|
45 |
|
46 |
# Decoding and printing sequences
|
47 |
for i, sample_output in enumerate(sample_outputs):
|
48 |
-
print(">> Generated text {}
|
49 |
-
|
50 |
{}".format(i+1, tokenizer.decode(sample_output.tolist())))
|
51 |
|
52 |
# >> Generated text
|
|
|
1 |
## GPT-Neo Small Portuguese
|
2 |
|
3 |
+
#### Model Description
|
4 |
This is a finetuned version from GPT-Neo 125M by EletheurAI to Portuguese language.
|
5 |
|
6 |
+
#### Training data
|
7 |
It was training from 227,382 selected texts from a PTWiki Dump. You can found all the data from here: https://archive.org/details/ptwiki-dump-20210520
|
8 |
|
9 |
+
#### Training Procedure
|
10 |
Every text was passed through a GPT2-Tokenizer with bos and eos tokens to separate it, with max sequence length that the GPT-Neo could support. It was finetuned using the default metrics of the Trainer Class, available on the Hugging Face library.
|
11 |
|
12 |
##### Learning Rate: **2e-4**
|
13 |
##### Epochs: **1**
|
14 |
|
15 |
+
#### Goals
|
16 |
+
|
17 |
My true intention was totally educational, thus making available a Portuguese version of this model.
|
18 |
|
19 |
How to use
|
|
|
46 |
|
47 |
# Decoding and printing sequences
|
48 |
for i, sample_output in enumerate(sample_outputs):
|
49 |
+
print(">> Generated text {}\\
|
50 |
+
\\
|
51 |
{}".format(i+1, tokenizer.decode(sample_output.tolist())))
|
52 |
|
53 |
# >> Generated text
|