更新pyproject.toml版本至0.2.2，优化vllm_infer函数以支持前缀缓存，调整qa_generator.py以处理结果并保存为JSON格式。

Author: xming521

pyproject.toml

@@ -20,7 +20,7 @@ dependencies = [
 
 [tool.weclone]
 # 配置文件的版本号，当配置文件结构或重要默认值发生变化时，应增加此版本号
-config_version = "0.2.1"
+config_version = "0.2.2"
 
 # 配置文件更新日志
 config_changelog = """

weclone/core/inference/vllm_infer.py

@@ -139,6 +139,7 @@ def vllm_infer(
         "pipeline_parallel_size": pipeline_parallel_size,
         "disable_log_stats": True,
         "enable_lora": model_args.adapter_name_or_path is not None,
+        "enable_prefix_caching": True,  # 是否启用前缀缓存
     }
     if template_obj.mm_plugin.__class__.__name__ != "BasePlugin":
         engine_args["limit_mm_per_prompt"] = {"image": 4, "video": 2, "audio": 2}
@@ -155,5 +156,3 @@ def vllm_infer(
     print("*" * 70)
     print(f"{len(prompts)} generated results have been saved at {save_name}.")
     print("*" * 70)
-
-

weclone/data/clean/clean_dataset.py

@@ -0,0 +1 @@
+#请检查sorce字段分布是否正常 调整accept_score  调整后按Y保存继续清洗

weclone/data/models.py

@@ -22,6 +22,18 @@ class CutMessage:
     CreateTime: Timestamp
 
 
+# TODO 未使用QaPair
+@dataclass
+class QaPair:
+    id: int
+    system: str
+    instruction: str
+    output: str
+    history: list[ChatMessage]
+    time: Timestamp
+    score: int
+
+
 skip_type_list = [
     "添加好友",
     "推荐公众号",

weclone/data/qa_generator.py

@@ -75,7 +75,7 @@ def main(self):
             template=self.c["template"],
             interval=self.c["cutoff_len"],
         )
-        logger.success(f"聊天记录处理成功，共{len(qa_res)}条，保存到./dataset/res_csv/sft/sft-my.json")
+        logger.success(f"聊天记录处理成功，共{len(qa_res)}条，保存到 ./dataset/res_csv/sft/sft-my.json")
 
     def get_csv_files(self):
         """遍历文件夹获取所有CSV文件路径"""
@@ -348,13 +348,17 @@ def process_text(self, chat_message: ChatMessage):
         pass
 
     def save_result(self, qa_res: List[Dict]):
-        # 保存结果
+        processed_qa_res = []
+        for idx, item in enumerate(qa_res):
+            if isinstance(item, dict):
+                item = {"id": idx, **item}
+                processed_qa_res.append(item)
         with open(
             "./dataset/res_csv/sft/sft-my.json",
             "w",
             encoding="utf-8",
         ) as f:
-            json.dump(qa_res, f, ensure_ascii=False)
+            json.dump(processed_qa_res, f, ensure_ascii=False)
 
 
 if __name__ == "__main__":