fix(graphgen): standardize types for qa_form

Author: ChenZiHong-Gavin

graphgen/configs/graphgen_config.yaml

@@ -3,7 +3,7 @@ input_file: resources/examples/raw_demo.jsonl
 tokenizer: cl100k_base
 quiz_samples: 2
 traverse_strategy:
-  qa_form: open
+  qa_form: aggregated
   bidirectional: true
   edge_sampling: max_loss
   expand_method: max_width

graphgen/graphgen.py

@@ -237,10 +237,12 @@ async def async_traverse(self):
                                                             self.traverse_strategy,
                                                             self.text_chunks_storage,
                                                             self.progress_bar)
-        else:
+        elif self.traverse_strategy.qa_form == "aggregated":
             results = await traverse_graph_by_edge(self.synthesizer_llm_client, self.tokenizer_instance,
                                                    self.graph_storage, self.traverse_strategy, self.text_chunks_storage,
                                                    self.progress_bar)
+        else:
+            raise ValueError(f"Unknown qa_form: {self.traverse_strategy.qa_form}")
         await self.qa_storage.upsert(results)
         await self.qa_storage.index_done_callback()
 

graphgen/models/strategy/travserse_strategy.py

@@ -6,7 +6,7 @@
 @dataclass
 class TraverseStrategy(BaseStrategy):
     # 生成的QA形式：原子、多跳、开放性
-    qa_form: str = "multi_hop" # "atomic" or "multi_hop" or "open"
+    qa_form: str = "atomic" # "atomic" or "multi_hop" or "aggregated"
     # 最大边数和最大token数方法中选择一个生效
     expand_method: str = "max_tokens" # "max_width" or "max_tokens"
     # 单向拓展还是双向拓展