intheashtray commited on
Commit
6f0bb8a
·
verified ·
1 Parent(s): 79c4323

Rename script/preprocess.py to scripts/preprocess.py

Browse files
Files changed (1) hide show
  1. {script → scripts}/preprocess.py +12 -11
{script → scripts}/preprocess.py RENAMED
@@ -1,20 +1,20 @@
1
  import json
2
  import os
3
  from datasets import Dataset, Image
4
- from transformers import CLIPProcessor, CLIPModel, AutoTokenizer
5
  from PIL import Image as PILImage
6
  import torch
7
  from tqdm import tqdm
8
 
9
- def preprocess_multimodal_data(raw_json_path, image_dir, dataset_name, tokenizer_name, clip_model_name, output_dir):
10
  """
11
- Preprocesses raw JSON with the specified format and image data into a Hugging Face Dataset.
12
 
13
  Args:
14
  raw_json_path (str): Path to the raw JSON file.
15
  image_dir (str): Directory containing the images.
16
  dataset_name (str): Name of the Hugging Face Dataset.
17
- tokenizer_name (str): Name of the tokenizer from Hugging Face Transformers.
18
  clip_model_name (str): Name of the CLIP model from Hugging Face Transformers.
19
  output_dir (str): Directory where the dataset repository will be saved.
20
  """
@@ -43,16 +43,17 @@ def preprocess_multimodal_data(raw_json_path, image_dir, dataset_name, tokenizer
43
  sections = []
44
  languages = []
45
 
46
- # Load CLIP and Tokenizer
47
  clip_processor = CLIPProcessor.from_pretrained(clip_model_name)
48
  clip_model = CLIPModel.from_pretrained(clip_model_name).to("cuda")
49
- tokenizer = AutoTokenizer.from_pretrained(tokenizer_name)
 
50
 
51
  # Process each data point
52
  for i, item in enumerate(tqdm(data, desc="Processing data")):
53
  question = item.get('question', '')
54
  answer = item.get('answer', '')
55
- image = None # Initialize image as None
56
  image_path = item.get('image', None)
57
  id_field = str(i)
58
 
@@ -115,11 +116,11 @@ def preprocess_multimodal_data(raw_json_path, image_dir, dataset_name, tokenizer
115
  print(f"Dataset '{dataset_name}' saved to {os.path.join(output_dir, dataset_name)}")
116
 
117
  # Example Usage:
118
- raw_json_path = 'your_data.json' # replace with your json path.
119
- image_dir = 'your_image_dir' # replace with your image directory.
120
  dataset_name = 'my_multimodal_dataset'
121
- tokenizer_name = 'deepseek-ai/deepseek-coder-1.3b-instruct'
122
  clip_model_name = 'openai/clip-vit-base-patch32'
123
  output_dir = 'my_datasets'
124
 
125
- preprocess_multimodal_data(raw_json_path, image_dir, dataset_name, tokenizer_name, clip_model_name, output_dir)
 
1
  import json
2
  import os
3
  from datasets import Dataset, Image
4
+ from transformers import CLIPProcessor, CLIPModel, AutoTokenizer, AutoModel
5
  from PIL import Image as PILImage
6
  import torch
7
  from tqdm import tqdm
8
 
9
+ def preprocess_multimodal_data(raw_json_path, image_dir, dataset_name, text_model_name, clip_model_name, output_dir):
10
  """
11
+ Preprocesses raw JSON and image data into a Hugging Face Dataset, using DeepSeek-R1 for text embeddings.
12
 
13
  Args:
14
  raw_json_path (str): Path to the raw JSON file.
15
  image_dir (str): Directory containing the images.
16
  dataset_name (str): Name of the Hugging Face Dataset.
17
+ text_model_name (str): Name of the text model (DeepSeek-R1) from Hugging Face Transformers.
18
  clip_model_name (str): Name of the CLIP model from Hugging Face Transformers.
19
  output_dir (str): Directory where the dataset repository will be saved.
20
  """
 
43
  sections = []
44
  languages = []
45
 
46
+ # Load CLIP, DeepSeek-R1, and Tokenizer
47
  clip_processor = CLIPProcessor.from_pretrained(clip_model_name)
48
  clip_model = CLIPModel.from_pretrained(clip_model_name).to("cuda")
49
+ text_model = AutoModel.from_pretrained(text_model_name).to("cuda")
50
+ tokenizer = AutoTokenizer.from_pretrained(text_model_name)
51
 
52
  # Process each data point
53
  for i, item in enumerate(tqdm(data, desc="Processing data")):
54
  question = item.get('question', '')
55
  answer = item.get('answer', '')
56
+ image = None
57
  image_path = item.get('image', None)
58
  id_field = str(i)
59
 
 
116
  print(f"Dataset '{dataset_name}' saved to {os.path.join(output_dir, dataset_name)}")
117
 
118
  # Example Usage:
119
+ raw_json_path = 'your_data.json'
120
+ image_dir = 'your_image_dir'
121
  dataset_name = 'my_multimodal_dataset'
122
+ text_model_name = 'deepseek-ai/deepseek-rl-7b-chat' #Deepseek R1
123
  clip_model_name = 'openai/clip-vit-base-patch32'
124
  output_dir = 'my_datasets'
125
 
126
+ preprocess_multimodal_data(raw_json_path, image_dir, dataset_name, text_model_name, clip_model_name, output_dir)