Merge pull request #3144 from opendatalab/release-2.1.3

Release 2.1.3

README.md

@@ -43,7 +43,7 @@
 </div>
 
 # Changelog
-- 2025/07/22 2.1.2 Released
+- 2025/07/22 2.1.3 Released
   - Bug Fixes
     - Fixed the issue of excessive memory consumption during the `MFR` step in the `pipeline` backend under certain scenarios #2771
     - Fixed the inaccurate matching between `image`/`table` and `caption`/`footnote` under certain conditions #3129

README_zh-CN.md

@@ -43,7 +43,7 @@
 </div>
 
 # 更新记录
-- 2025/07/22 2.1.2发布
+- 2025/07/22 2.1.3发布
   - bug修复
     - 修复`pipeline`后端中`MFR`步骤在某些情况下显存消耗过大的问题 #2771
     - 修复某些情况下`image`/`table`与`caption`/`footnote`匹配不准确的问题 #3129

mineru/model/mfr/unimernet/Unimernet.py

@@ -104,6 +104,10 @@ class UnimernetModel(object):
 
         # Create dataset with sorted images
         dataset = MathDataset(sorted_images, transform=self.model.transform)
+
+        # 如果batch_size> len(sorted_images)，则设置为不超过len(sorted_images)的2的幂
+        batch_size = min(batch_size, 2 ** (len(sorted_images).bit_length() - 1))
+
         dataloader = DataLoader(dataset, batch_size=batch_size, num_workers=0)
 
         # Process batches and store results