Update projects/instructblip/run_demo.py, README.md, scienceqa_data_preprocess.py files

README.md

@@ -71,11 +71,22 @@ export HF_ENDPOINT=https://hf-mirror.com
 SCNet快速下载链接：   
 - http://113.200.138.88:18080/aimodels/findsource-dependency/vicuna-13b-v1.1  
 
+数据集结构为：   
+input  
+├── scienceqa  
+│   ├── images  
+│   │   ├── train  
+│   │   ├── test  
+│   │   ├── val  
+│   ├── scienceqa_problems_path.json  
+│   ├── scienceqa_pid_splits.json    
+
 对ScienceQA进行预处理：  
 ```
-# 修改预处理代码中scienceqa数据集解压位置
+# 修改预处理代码中scienceqa数据路径为数据解压位置，上面的数据结构将数据解压至input文件夹中
 python scienceqa_data_preprocess.py
 ```
+运行该命令后会在/input/scienceqa/目录下生成scienceqa_train.json、scienceqa_test.json、scienceqa_val.json文件  
 该命令将scienceQA转为指令微调数据集，指令格式为： 
 ``` 
 <Image> Context: { {hint} {lecture} } Question: { {question} } Options: { {choices} } Answer: (a) { {answer} }

projects/instructblip/run_demo.py

@@ -116,4 +116,4 @@ if __name__ == '__main__':
         inputs=[image_input, prompt_textbox, min_len, max_len, beam_size, len_penalty, repetition_penalty, top_p, sampling],
         outputs="text",
         allow_flagging="never",
-    ).launch()
+    ).launch(share=True, server_name="0.0.0.0")

scienceqa_data_preprocess.py

 import json
 from tqdm import tqdm
     
-with open("scienceqa_problems_path.json", 'r') as file:
+with open("/input/scienceqa/scienceqa_problems_path.json", 'r') as file:
     data = json.load(file)
     
-with open("scienceqa_pid_splits.json") as file:
+with open("/input/scienceqa/scienceqa_pid_splits.json") as file:
     pid_splits = json.load(file)
 
 train_ids = pid_splits['train']
@@ -18,7 +18,7 @@ for id in tqdm(train_ids):
     train_data = data[str(id)]
     if train_data['image'] is None:
         continue
-    image_url = f"scienceqa/images/train/{id}/image.png"
+    image_url = f"/input/scienceqa/images/train/{id}/image.png"
     if train_data['answer'] == 0:
         answer = "(a) " + train_data['choices'][train_data['answer']]
     elif train_data['answer'] == 1:
@@ -46,7 +46,7 @@ for id in tqdm(val_ids):
     val_data = data[str(id)]
     if val_data['image'] is None:
         continue
-    image_url = f"scienceqa/images/val/{id}/image.png"
+    image_url = f"/input/scienceqa/images/val/{id}/image.png"
     if val_data['answer'] == 0:
         answer = "(a) " + val_data['choices'][val_data['answer']]
     elif val_data['answer'] == 1:
@@ -74,7 +74,7 @@ for id in tqdm(test_ids):
     test_data = data[str(id)]
     if test_data['image'] is None:
         continue
-    image_url = f"scienceqa/images/test/{id}/image.png"
+    image_url = f"/input/scienceqa/images/test/{id}/image.png"
     if test_data['answer'] == 0:
         answer = "(a) " + test_data['choices'][test_data['answer']]
     elif test_data['answer'] == 1:
@@ -102,4 +102,4 @@ with open("/input/scienceqa/scienceqa_test.json", 'w') as file:
     json.dump(test_annotation, file)
 
 with open("/input/scienceqa/scienceqa_val.json", 'w') as file:
-    json.dump(val_annotation, file)
+    json.dump(val_annotation, file)
\ No newline at end of file