diff --git "a/docs/source/LLM/\350\207\252\345\256\232\344\271\211\344\270\216\346\213\223\345\261\225.md" "b/docs/source/LLM/\350\207\252\345\256\232\344\271\211\344\270\216\346\213\223\345\261\225.md"
index 65dee97901..708d70919e 100644
--- "a/docs/source/LLM/\350\207\252\345\256\232\344\271\211\344\270\216\346\213\223\345\261\225.md"
+++ "b/docs/source/LLM/\350\207\252\345\256\232\344\271\211\344\270\216\346\213\223\345\261\225.md"
@@ -107,6 +107,16 @@ AAAAA,BBBBB,CCCCC
 {"query": "AAAAA", "response": "BBBBB", "rejected_response": "CCCCC"}
 ```
 
+**CogAgent模型**
+
+```jsonl
+{"query": "55555", "response": "66666", "image": "some-local-image-path"}
+{"query": "eeeee", "response": "fffff", "history": [], "image": "some-http-image-path"}
+{"query": "EEEEE", "response": "FFFFF", "history": [["AAAAA", "BBBBB"], ["CCCCC", "DDDDD"]], "image": "some-local-image-path"}
+```
+
+image字段支持本地图片文件和http可访问的image url两类。
+
 ### 注册数据集的方式
 
 以下是一个**注册数据集**的案例. 完整的py文件可以查看[custom.py](https://github.com/modelscope/swift/blob/main/examples/pytorch/llm/custom.py), sh脚本可以查看[custom](https://github.com/modelscope/swift/tree/main/examples/pytorch/llm/scripts/custom).
diff --git a/examples/pytorch/llm/scripts/cogagent_chat/lora/sft.sh b/examples/pytorch/llm/scripts/cogagent_chat/lora/sft.sh
index 0b642444db..093d672859 100644
--- a/examples/pytorch/llm/scripts/cogagent_chat/lora/sft.sh
+++ b/examples/pytorch/llm/scripts/cogagent_chat/lora/sft.sh
@@ -26,7 +26,7 @@ python llm_sft.py \
     --eval_steps 100 \
     --save_steps 100 \
     --save_total_limit 2 \
-    --logging_steps 10
+    --logging_steps 10 \
     --push_to_hub false \
     --hub_model_id cogagent-chat-lora \
     --hub_private_repo true \
diff --git a/swift/llm/utils/dataset.py b/swift/llm/utils/dataset.py
index 3c8769f209..8d2dc7c9c6 100644
--- a/swift/llm/utils/dataset.py
+++ b/swift/llm/utils/dataset.py
@@ -22,7 +22,7 @@
                          PreprocessFunc, RenameColumnsPreprocessor,
                          SmartPreprocessor, TextGenerationPreprocessor)
 from .template import History
-from .utils import download_dataset
+from .utils import dataset_map, download_dataset
 
 
 def _remove_useless_columns(dataset: HfDataset) -> HfDataset:
@@ -1120,7 +1120,21 @@ def load_dataset_from_local(
                 'for more information.')
         dataset = HfDataset.from_dict(df.to_dict(orient='list'))
         dataset_list.append(preprocess_func(dataset))
-    return concatenate_datasets(dataset_list)
+
+    dataset = concatenate_datasets(dataset_list)
+
+    def load_image(row):
+        from PIL import Image
+        import requests
+        if not os.path.exists(row['image']):
+            row['image'] = requests.get(row['image'], stream=True).raw
+        row['image'] = Image.open(row['image'])
+        return row
+
+    if 'image' in dataset.features and isinstance(dataset[0]['image'], str):
+        dataset = HfDataset.from_list(
+            dataset_map(dataset, load_image, num_proc=4).data)
+    return dataset
 
 
 def get_custom_dataset(_: str, train_subset_split_list: Union[str, List[str]],