Merge branch 'master' into update-clean-strategy

.gitignore

@@ -162,3 +162,4 @@ Qwen*/
 settings.jsonc
 settings.json
 dataset/blocked_words.json
+dataset/wechat/*

settings.template.jsonc

@@ -1,5 +1,5 @@
 {
-    "version": "0.2.2",
+    "version": "0.2.21",
     "common_args": {
         "model_name_or_path": "./Qwen2.5-7B-Instruct",
         "adapter_name_or_path": "./model_output", //同时做为train_sft_args的output_dir
@@ -14,7 +14,8 @@
     "make_dataset_args": {
         //数据处理配置
         "include_type": [
-            "文本"
+            "text",
+            // "image"
         ],
         "blocked_words": [ // 禁用词
             "例如 姓名",
@@ -34,11 +35,11 @@
                 "accept_score": 2, //可以接受的llm打分阈值,1分最差，5分最好,低于此分数的数据不会用于训练
             }
         },
-        "online_llm_clear":false,
-        "base_url":"https://xxx/v1",
-        "llm_api_key":"xxxxx",
-        "model_name":"xxx",   //建议使用参数较大的模型，例如DeepSeek-V3
-        "clean_batch_size":10
+        "online_llm_clear": false,
+        "base_url": "https://xxx/v1",
+        "llm_api_key": "xxxxx",
+        "model_name": "xxx", //建议使用参数较大的模型，例如DeepSeek-V3
+        "clean_batch_size": 10
     },
     "train_pt_args": {
         //预训练微调配置
@@ -91,4 +92,4 @@
         "max_length": 50,
         "top_p": 0.65
     }
-}
+}

weclone/data/chat_parsers/wechat_parser.py

@@ -0,0 +1,8 @@
+class WeChatParser:
+    def decrypt_wechat_image(self, encrypted_path, output_path):
+        """解密微信加密的图片文件"""
+        pass
+
+    def parse_chat_records(self, db_path):
+        """解析聊天记录数据库"""
+        pass