Upload convert_dataset_v2.py with huggingface_hub
Browse files- convert_dataset_v2.py +59 -0
convert_dataset_v2.py
ADDED
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
from tqdm import tqdm
|
3 |
+
import os
|
4 |
+
|
5 |
+
# GAIA
|
6 |
+
data_path = "data/mat_train.json"
|
7 |
+
|
8 |
+
with open(data_path, "r") as f:
|
9 |
+
dataset = json.load(f)
|
10 |
+
|
11 |
+
def _convert(image_path_map, conversations):
|
12 |
+
output = []
|
13 |
+
for turn in conversations:
|
14 |
+
role = turn["role"]
|
15 |
+
content = turn["content"]
|
16 |
+
turn_new = dict()
|
17 |
+
turn_new["from"] = role
|
18 |
+
pid = 1
|
19 |
+
keys = sorted(list(image_path_map.keys()))
|
20 |
+
for k in keys:
|
21 |
+
v = image_path_map[k]
|
22 |
+
if k in content:
|
23 |
+
content = content.replace(k, f"Picture {pid}: <img>{v}</img>\n")
|
24 |
+
content = content.replace(f"</img>\n\n", "</img>\n")
|
25 |
+
pid += 1
|
26 |
+
turn_new["value"] = content
|
27 |
+
output.append(turn_new)
|
28 |
+
return output
|
29 |
+
|
30 |
+
|
31 |
+
for item in tqdm(dataset):
|
32 |
+
#print(item["image"])
|
33 |
+
#print(item.keys())
|
34 |
+
conversations = item["conversations"]
|
35 |
+
#print(len(conversations), conversations[1])
|
36 |
+
image_path_map = dict()
|
37 |
+
if "image" not in item:
|
38 |
+
pass
|
39 |
+
elif type(item["image"]) == str:
|
40 |
+
image_path_map["<image>"] = item["image"]
|
41 |
+
item['image'] = f"{os.getcwd()}/data/{item['image']}"
|
42 |
+
else:
|
43 |
+
for k, v in item["image"].items():
|
44 |
+
image_path_map[k] = v
|
45 |
+
item["image"][k] = f"{os.getcwd()}/data/{v}"
|
46 |
+
item["conversations"] = _convert(image_path_map, conversations)
|
47 |
+
|
48 |
+
from datetime import datetime
|
49 |
+
import json
|
50 |
+
|
51 |
+
now = "20241209_1731"
|
52 |
+
print("write to", f"data/train_{now}.json")
|
53 |
+
with open(f"data/train_{now}.json", "w") as f:
|
54 |
+
json.dump(dataset, f, indent=4, ensure_ascii=False)
|
55 |
+
|
56 |
+
import random
|
57 |
+
with open(f"data/train_{now}_subset.json", "w") as f:
|
58 |
+
random.shuffle(dataset)
|
59 |
+
json.dump(dataset[:1000], f, indent=4, ensure_ascii=False)
|