tunji17
commited on
Commit
•
213bf26
1
Parent(s):
a60711d
adds jsonl dataset
Browse files- fakenews_data.jsonl +0 -0
- process_data.py +41 -0
fakenews_data.jsonl
ADDED
The diff for this file is too large to render.
See raw diff
|
|
process_data.py
ADDED
@@ -0,0 +1,41 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datasets import load_dataset
|
2 |
+
|
3 |
+
dataset_name = "2nji/makebelieve-480"
|
4 |
+
|
5 |
+
# Load the dataset
|
6 |
+
dataset = load_dataset(dataset_name)
|
7 |
+
|
8 |
+
# Shuffle the dataset and slice it
|
9 |
+
dataset = dataset['train'].shuffle(seed=42).select(range(480))
|
10 |
+
|
11 |
+
|
12 |
+
# Define a function to transform the data
|
13 |
+
def transform_conversation(example):
|
14 |
+
conversation_text = example['text']
|
15 |
+
segments = conversation_text.split('###')
|
16 |
+
|
17 |
+
reformatted_segments = []
|
18 |
+
|
19 |
+
# Iterate over pairs of segments
|
20 |
+
for i in range(1, len(segments) - 1, 2):
|
21 |
+
human_text = segments[i].strip().replace('Human:', '').strip()
|
22 |
+
|
23 |
+
# Check if there is a corresponding assistant segment before processing
|
24 |
+
if i + 1 < len(segments):
|
25 |
+
assistant_text = segments[i+1].strip().replace('Assistant:', '').strip()
|
26 |
+
|
27 |
+
# Apply the new template
|
28 |
+
reformatted_segments.append(f'<s>[INST] {human_text} [/INST] {assistant_text} </s>')
|
29 |
+
else:
|
30 |
+
# Handle the case where there is no corresponding assistant segment
|
31 |
+
reformatted_segments.append(f'<s>[INST] {human_text} [/INST] </s>')
|
32 |
+
|
33 |
+
return {'text': ''.join(reformatted_segments)}
|
34 |
+
|
35 |
+
|
36 |
+
# Apply the transformation
|
37 |
+
transformed_dataset = dataset.map(transform_conversation)
|
38 |
+
|
39 |
+
# Upload the dataset to the Hub
|
40 |
+
# Don't forget to replace the token with your own
|
41 |
+
transformed_dataset.push_to_hub(dataset_name, token="...")
|